Upload
hoangcong
View
220
Download
5
Embed Size (px)
Citation preview
Prognostic Factors in Aneurysmal Subarachnoid Hemorrhage: Pooled
Analyses of Individual Patient Data and Development of Novel Risk
Scores in Large Cohorts of International Patients
by
Blessing Nathan Romey Jaja
A thesis submitted in conformity with the requirements
for the degree of Doctor of Philosophy,
Graduate Department of the Institute of Medical Science
University of Toronto
© Copyright by Blessing Nathan Romey Jaja 2014
ii
Prognostic Factors in Aneurysmal Subarachnoid Hemorrhage: Pooled Analyses of
Individual Patient Data and Development of Novel Risk Scores in Large Cohorts of
International Patients
Blessing Nathan Romey Jaja
Doctor of Philosophy
Institute of Medical Science
University of Toronto
2014
Abstract
Primary studies reporting prognostic associations in aneurysmal subarachnoid hemorrhage
(SAH) are often insufficiently powered, use data of limited representativeness and scarcely
examine the added value of prognostic factors above those of other known factors. Hence,
considerable knowledge gaps and conflicting results exist in the literature on the nature and
extent of prognostic associations in SAH. Prognostic factors have been combined to develop
prediction models and risk scores for early outcome prediction after SAH. None is routinely
applied in clinical or research settings; some major constraints relate to lack of evidence on the
predictive accuracy, reliability and generalizability of reported risk scores. The global aim of this
research was to address these challenges and contribute to improved understanding of prognostic
associations in SAH by analysing large cohorts of SAH patients reflecting a broad spectrum of
settings. Pooled analyses of patient-level data from multiple randomized clinical trials and
prospective hospital registries involving 10963 patients demonstrated a strong prognostic effect
iii
of admission neurologic status on 3-month outcome according to Glasgow outcome score. Age
had a moderate effect on outcome; premorbid hypertension and subarachnoid clot burden on the
Fisher scale were weak predictors of outcome. Patient’s sex had no independent predictive
value. Prognostic effect of aneurysm size and location depended on treatment modality. Novel
prognostic scores were developed combining these predictors for early prediction of mortality
and unfavorable outcomes at 3 months, and demonstrated adequate performance at bootstrap
(AUC: 0.77 – 0.83) and at cross validation. Using 2 nationally representative administrative
datasets, socioeconomic status and race/ethnicity were explored as latent prognostic factors in
SAH. Multinomial logistic regression analysis demonstrated socioeconomic status, measured as
neighborhood income status, was associated with inpatient mortality risk after admission for
SAH. The extent of the association could be related to health care system under which treatment
was provided. Race/ethnicity was independently associated with inpatient mortality. Patients of
Hispanic ethnicity had the best outcomes and Asia/Pacific Islanders experienced the worst
outcomes during the inpatient course. This research has provided higher level evidence than prior
studies on studied prognostic factors and reliable tools for early prediction of outcome after
hospitalization for SAH.
iv
Acknowledgements
Certain persons and institutions were instrumental to the successful completion of this
project. They deserve mention and my vote of thanks, for without them this project would have
been only a pipe dream. Of special mention is my supervisor, Dr. R. Loch Macdonald, who I
would like to thank immensely for giving me the opportunity to undertake this research, and for
his patience, understanding, continued guidance and support through the course of the research
program. I would like to express gratitude to my Program Advisory Committee members, Drs.
Mohammed Mamdani, Gustavo Saposnik, Kevin Thorpe and Peter Austin for providing
expertise and insightful comments which gave direction to and helped shape the final outcome of
the research. May I also acknowledge the consultations and invaluable statistical guidance that I
received from Prof. Ewout W. Steyerberg and Dr. Hester Lingsma, Department of Public Health,
Erasmus MC - University Medical Center Rotterdam, Netherland.
I am grateful to members of the Subarachnoid Hemorrhage International Trialists
(SAHIT) collaboration for contributing institutional clinical data on subarachnoid hemorrhage
patients which served as the primary source of patient data for the research. Also deserving of
mention are the Canadian Institute for Health Information for providing free access to their
Discharge Abstract Database, and the Agency for Healthcare Quality and Research (Rockville,
MD), USA, for access to the Healthcare Cost and Utilization Project Nationwide Inpatient
sample.
May I also acknowledge funding support from the Canadian Institute for Health
Research, University of Toronto Open fellowship and conference grant, Division of
Neurosurgery, St. Michael’s Hospital, and the Rivers State Government of Nigeria.
v
To my friends, Dr. Ken Simiyu and Khwaka Kukubo, thank you for your encouragement,
support, and the many opportunities for relaxation and comic relief from the long hours of data
management and computation. To my parents, thank you for your persistent prayers,
unconditional love and for inculcating in me the values of academics. To my family, particularly
Benita, thank you for your personal sacrifices, and for enduring the long periods of my absence.
Finally, I thank the good Lord for the blessing of life, health and faith to start and
complete the onerous task of doctoral-level training.
vi
Table of Contents
Abstract ............................................................................................................................................ii
Acknowledgements ......................................................................................................................... iv
List of Tables ................................................................................................................................... x
List of Figures ................................................................................................................................. xi
List of appendices .......................................................................................................................... xii
Symbols and Abbreviations .......................................................................................................... xiii
Chapter 1 ......................................................................................................................................... 1
Introduction and literature review ................................................................................................... 1
1.1 Introduction .......................................................................................................................... 2
1.2 Pathobiology of aneurysmal subarachnoid hemorrhage ...................................................... 2
1.2.1 Risk factors for aneurysmal subarachnoid hemorrhage ................................................... 3
1.2.1.1 Modifiable risk factors ..................................................................................................... 4
1.2.1.2 Non modifiable risk factors ...................................................................................... 4
1.2.2 Pathophysiology of aneurysmal subarachnoid hemorrhage ............................................. 7
1.3 Diagnosis and clinical presentation ..................................................................................... 9
1.4 Treatment of aneurysmal subarachnoid hemorrhage ......................................................... 12
1.5 Prognosis of aneurysmal subarachnoid hemorrhage .......................................................... 13
1.6 Prognostic Factors in aneurysmal subarachnoid hemorrhage ............................................ 15
1.6.1 Demographic factors ...................................................................................................... 17
1.6.2 Clinical factors ............................................................................................................... 22
1.6.3 Neuroimaging factors ..................................................................................................... 28
1.6.4 Physiologic factors and biomarkers ............................................................................... 29
1.7 Limitations of prognostic studies ....................................................................................... 30
1.8 Prognostic models in SAH: How reliable are available models to predict outcome of
patients with SAH? ....................................................................................................................... 32
1.8.1 Results of systematic review of prognostic models in SAH .......................................... 34
1.8.2 Implications of findings of the systematic review ......................................................... 38
1.9 Conclusion ......................................................................................................................... 41
Chapter 2 ....................................................................................................................................... 50
vii
Study Methodology ....................................................................................................................... 50
2.1 Introduction ........................................................................................................................ 51
2.2 Study rationale: Towards providing higher level evidence of prognostic associations and
better prognostic models in SAH .................................................................................................. 51
2.3 Hypothesis and research contributions .............................................................................. 54
2.4 Research Ethics Board Approvals ..................................................................................... 57
2.5 Study population ................................................................................................................ 57
2.6 Independent (Predictors) variables .................................................................................... 60
2.7 Dependent (Outcome) Variables........................................................................................ 60
2.8 Approach to statistical analyses ......................................................................................... 61
2.8.1 Descriptive analysis........................................................................................................ 61
2.8.2 Univariable prognostic association ................................................................................ 61
2.8.3 Multivariable prognostic association ............................................................................. 62
2.8.4 Quantification of the magnitude of prognostic associations .......................................... 62
2.8.5 Secondary analyses: interaction effects and subgroup analysis ..................................... 63
2.9 Handling of missing data ................................................................................................... 64
2.10 Statistical software ............................................................................................................. 64
Chapter 3 ....................................................................................................................................... 67
Prognostic value of hospital admission characteristics in aneurysmal subarachnoid hemorrhage:
Meta-analyses of individual participant data in the subarachnoid hemorrhage international
trialists (SAHIT) repository .......................................................................................................... 67
3.1 Introduction ........................................................................................................................ 68
3.2 PART A: Demographic factors - age and sex .................................................................... 68
3.2.1 Methods .......................................................................................................................... 69
3.2.2 Results ............................................................................................................................ 70
3.2.3 Discussion ...................................................................................................................... 71
3.3 PART B: Clinical factors - premorbid hypertension and admission neurologic status ..... 77
3.3.1 Methods .......................................................................................................................... 77
3.3.2 Results ............................................................................................................................ 78
3.3.3 Discussion ...................................................................................................................... 78
viii
3.4 PART C: Neuroimaging factors: Fisher CT clot burden, aneurysm location and size ...... 89
3.4.1 Methods .......................................................................................................................... 90
3.4.2 Results ............................................................................................................................ 91
3.4.3 Discussion ...................................................................................................................... 93
Chapter 4 ....................................................................................................................................... 99
SAHIT score: Novel prognostic scores for early prediction of outcome after aneurysmal
subarachnoid hemorrhage ............................................................................................................. 99
4.1 Introduction ...................................................................................................................... 100
4.2 Methods............................................................................................................................ 101
4.2.1 Study population .......................................................................................................... 101
4.2.2 Variable selection ......................................................................................................... 102
4.2.3 Outcome measure ........................................................................................................ 102
4.2.4 Model development ...................................................................................................... 102
4.2.5 Model performance ...................................................................................................... 103
4.2.6 Model Validation.......................................................................................................... 104
4.3 Results .............................................................................................................................. 105
4.3.1 Model performance ...................................................................................................... 105
4.3.2 Model presentation ....................................................................................................... 106
4.4 Discussion ........................................................................................................................ 107
Chapter 5 ..................................................................................................................................... 123
Investigating socioeconomic status and race/ethnicity as latent prognostic factors in aneurysmal
subarachnoid hemorrhage ........................................................................................................... 123
5.1 Introduction ...................................................................................................................... 124
5.2 Part A: Socioeconomic status and inpatient mortality risk after SAH ............................. 124
5.2.1 Methods ........................................................................................................................ 125
5.2.2 Results .......................................................................................................................... 128
5.2.3 Discussion .................................................................................................................... 129
5.3 Part B: Race/ethnicity and inpatient mortality risk after SAH ........................................ 136
5.3.1 Methods ........................................................................................................................ 136
5.3.2 Results .......................................................................................................................... 138
ix
5.3.3 Discussion .................................................................................................................... 139
Chapter 6 ..................................................................................................................................... 146
General discussion and conclusion ............................................................................................. 146
6.1 Introduction ...................................................................................................................... 147
6.2 Strength of the research ................................................................................................... 147
6.3 Summary of research findings and contributions to knowledge ...................................... 149
6.4 Limitations of the research ............................................................................................... 154
6.5 Directions for future research .......................................................................................... 157
6.5.1 Confirmatory study of the prognostic value of other factors ....................................... 157
6.5.2 Validation of Risk Scores in Subarachnoid Aneurysm (VISA) ................................... 158
6.5.3 Center variability in outcomes of SAH ........................................................................ 159
6.5.4 Type and timing of outcome assessment ..................................................................... 161
6.6 Conclusion ....................................................................................................................... 163
Appendices .................................................................................................................................. 189
x
List of Tables
Table 1.1 – Approaches adopted by previous studies to analyze age effect on outcome ............. 43
Table 1.2 – Studies identifying independent predictors of poor outcome in multivariable analysis
....................................................................................................................................................... 45
Table 1.3 – Characteristics of studies reporting prognostic models in SAH ................................ 48
Table 1.4 – Approach to model development in previous studies reporting prognostic models in
SAH............................................................................................................................................... 49
Table 1.5 – Approach to model validation in previous studies..................................................... 49
Table 2.1 – Characteristics of studies in the SAHIT repository ................................................... 65
Table 3.1 – Results of adjusted analysis of the prognostic effect of age and sex ......................... 76
Table 3.2 – Distribution of premorbid hypertension and neurologic status by 3-month GOS ..... 84
Table 3.3 – Adjusted effects of premorbid hypertension and neurologic status ........................... 84
Table 3.4 – Relation of premorbid hypertension to comorbid conditions and complications ...... 85
Table 3.5 – Distribution of ruptured aneurysm location, diameter and Fisher clot burden by study
....................................................................................................................................................... 95
Table 3.6 – Distribution of aneurysm location by treatment modality ......................................... 95
Table 3.7 – Distribution of aneurysm location, diameter and Fisher clot burden by GOS .......... 95
Table 3.8 – Adjusted effects of studied neuroimaging factors ..................................................... 96
Table 4.1 – Baseline distribution of variables by study cohort................................................... 112
Table 4.2 – Association between predictors and 3-month outcome ........................................... 113
Table 4.3 – Performance indices of the six models at bootstrap validation ............................... 114
Table 4.4 – Performance indices at leave-one-study-out cross validation.................................. 115
Table 4.5 – Model coefficients to obtain linear predictors for computing outcome probabilities
..................................................................................................................................................... 122
Table 5.1 – Baseline characteristics of US patients by quartile of median household income .. 133
Table 5.2 – Baseline characteristics of Canadian patients by quintile of median household
income ......................................................................................................................................... 134
Table 5.3 – Relation of neighborhood income to in-hospital mortality and discharge to iPAC for
US patients .................................................................................................................................. 134
Table 5.4 – Relation of neighborhood income status to in-hospital mortality and discharge to
iPAC for Canadian patients ........................................................................................................ 135
Table 5.5 – Baseline distribution of variables according to race/ethnicity status ....................... 143
Table 5.6 – Results of multivariable analysis of the relation of race/ethnicity to in-hospital
mortality ...................................................................................................................................... 144
xi
List of Figures
Figure 2.1 – Framework of multivariable analysis in SAHIT repository ..................................... 66
Figure 3.1 – Boxplot of age by study cohort ................................................................................ 73
Figure 3.2 – Spline plot of the relation of age to outcome at different dichotomization split points
of the GOS .................................................................................................................................... 74
Figure 3.3 – Forest plot demonstrating consistency in the effects of age and sex ........................ 75
Figure 3.4 – Relative prognostic value of studied prognostic factors expressed as Nagelkerke’s
partial R2 ....................................................................................................................................... 76
Figure 3.5 – Percentage distribution of neurologic status in included studies.............................. 86
Figure 3.6 – Forest plot of the effect of premorbid hypertension across studies .......................... 87
Figure 3.7 – Forest plot demonstrating consistency in the effect of neurologic status across
studies ........................................................................................................................................... 88
Figure 3.8 – U-shaped relation of aneurysm size to GOS outcome with change point at 5.5mm 96
Figure 3.9 – Forest plot to examine consistency in the relation of aneurysm location and diameter
to outcome across studies.............................................................................................................. 97
Figure 3.10 – Forest plot illustrating prognostic strength of SAH clot burden across studies ..... 98
Figure 4.1 – Spline plot of the relationship between age and 3-month GOS outcome ............... 114
Figure 4.2 – Cross validation plots in CONSCIOUS I cohorts .................................................. 116
Figure 4.3 – Cross validation plots in Tirilazad cohorts ............................................................. 117
Figure 4.4 – Cross validation plots in IHAST cohorts................................................................ 118
Figure 4.5 – Cross validation plots in D-SAT cohorts................................................................ 119
Figure 4.6 – Cross validation plots in SHOP cohorts ................................................................. 120
Figure 4.7 – Plots of predicted probabilities of mortality by sum score with 95% confidence
intervals based on the core model ............................................................................................... 121
Figure 5.1 – Plots of predicted probabilities of mortality showing the effect of neighborhood
income status in the US increases with advancing age ............................................................... 135
Figure 5.2 – Plots of race/ethnicity differences in risk of mortality by age, expressed as predicted
probabilities (y axis) ................................................................................................................... 145
xii
List of appendices
Appendix A: PUBMED search history for review of prognostic models in SAH……….….....189
Appendix B: Copyright licenses…………………………………………………...…………...192
xiii
Symbols and Abbreviations
ACA Anterior cerebral artery
API Asian/ Pacific Islander
ASA aspirin;
AUC Area under the receiver operation curve
BASIC Brain Attack Surveillance in Corpus Christi
C-1 CONSCIOUS-1 Trial
CCI Charlson-deyo comorbidity index score
CIL Calibration in the large
CNS Central nervous system
CSF Cerebrospinal fluid
CT Computed tomography
DAD Discharge abstract database
DCI Delayed cerebral ischemia
DIND Delayed Ischemic Neurologic deficits
D-SAT Database of Subarachnoid Treatment of the University of Washington
EBI Early brain injury
FINMONICA Finnish contribution to the world health organization multinational monitoring of
trends and determinants of cardiovascular disease
GCS Glasgow Coma Score
GOF Le Cessie - van Houwelingen - Copas - Hosmer test of goodness of fit
GOS Glasgow outcome score
HHU Heinrich-Heine University
ICA Internal cerebral artery
ICD International classification of diseases
ICH Intracerebral hemorrhage
ICU Intensive care unit
IHAST Intraoperative hypothermia for aneurysm surgery trial
IMASH Intravenous magnesium sulphate for aneurysmal subarachnoid hemorrhage
IMPACT International mission for prognosis and clinical trial
iPAC Institutional post-acute care
ISAT International subarachnoid aneurysm trial
MAPS Matrix and platinum science trial
MAR Missing at random
MASH Magnesium sulfate in aneurysmal subarachnoid hemorrhage
MCA Middle cerebral artery
Mg Magnesium
MICE Multiple imputations by chained equations
MRI Magnetic resonance imaging
NA North America
NICSAH Intravenous Nicardipine after aneurysmal subarachnoid hemorrhage study
NIHSS National institute of health stroke scale
NIS Nationwide Inpatient sample
xiv
NOMAS Northern Manhattan study
PCQ Posterior circulation
PRISMA Preferred reporting items for systematic reviews and meta-analyses
PTSD Post traumatic stress disorder
QoL Quality of life
RCTs Randomized clinical trials
RS Rankin score /scale
SAH Subarachnoid hemorrhage (aneurysmal)
SAHIT Subarachnoid hemorrhage international trialists
SES Socioeconomic status
SHOP Subarachnoid hemorrhage outcomes project
SPARC Statewide planning and research cooperative system
TAR Target aneurysm recurrence
US United States
WFNS World Federation of Neurological Surgeons
1
Chapter 1
Introduction and literature review
A part of this chapter is adapted with permission from the following publication:
Jaja BN, Cusimano MD, Etminan N, Hanggi D, Hasan D, Ilodigwe D, Lantigua H, Le Roux P,
Lo B, Louffat-Olivares A, Mayer S, Molyneux A, Quinn A, Schweizer TA, Schenk T, Spears J,
Todd M, Torner J, Vergouwen MD, Wong GK, Singh J, Macdonald RL; for the SAHIT
Collaboration. Clinical prediction models for aneurysmal subarachnoid hemorrhage: a systematic
review. Neurocritical Care 2013; 18(1):143-53.
2
1.1 Introduction
Subarachnoid hemorrhage (SAH) from ruptured intracranial aneurysms is responsible for
only 5% of strokes but produces a disproportionate 25% mortality rate compared with other
stroke types. Though the annual incidence is estimated at 9 per 100,000 persons, considerable
regional variations are present in annual incidence with the highest incidence rates reported in
Japan and Finland.1 Unlike other stroke types that are relatively more common in the elderly,
SAH occurs more commonly in younger adults with an average age between 40 and 60 years and
has an estimated mortality rate of 40% to 60%.1-3
About one-third of untreated patients die
within 6 months of the condition4, 5
and survivors have varying degrees of neurocognitive and
functional deficits which limit their capacity for independent living.6, 7
Less than one-third of
patients are able to return to their previous occupation and lifestyle.8 SAH is therefore an
important cause of premature death and disability.
1.2 Pathobiology of aneurysmal subarachnoid hemorrhage
Trauma may be the most common cause of SAH although spontaneous SAH is due to
rupture of intracranial aneurysms in 85% of cases.9 Non aneurysmal causes are relatively rare
and include a non-exhaustive list comprising idiopathic perimesencephalic SAH; congenital and
acquired coagulation defects; non inflammatory lesions and vasculopathies such as arteriovenous
malformations (AVM), fusiform aneurysms, cavernous malformation, moya moya; pituitary
apoplexy, cerebral venous thrombosis, reversible cerebral vasoconstriction syndrome,
inflammatory lesions, for example mycotic aneurysm, polyarteritis nodosa, Churg–Strauss
syndrome and Wegener granulomatosis; other rare causes are sickle cell related coagulopathies,
CNS tumors, and drugs commonly cocaine.9, 10
3
Intracranial aneurysms are localized dilatations of cerebral blood vessels that are due to
weakness in the vessel wall. They are commonly saccular in shape but may be occasionally
fusiform in shape. Saccular aneurysms have an out pouched fundus that tapers into a proximal
neck attaching the aneurysm to the parent vessel. The fundus may be lobulated in which case it
may involve neighbouring branching vessels. Though of obscure and complex etiology,
intracranial aneurysms are generally believed to be a chronic, acquired, degenerative disorder of
cerebral arteries. They could be found at any age, but they are rare in children and are present in
2-5% of the general population; this proportion is double in individuals with a positive family
history of intracranial aneurysms. Intracranial aneurysms are typically found at the bifurcations
of the major vessels of the circle of Willis. The anatomical distribution is variable depending on
the population studied and the design of the study.11-13
Approximately 20% of intracranial
aneurysms are multiple in nature. Approximately 85% are found in the carotid or “anterior”
circulation, and about 15% occur in the vertebrobasilar or “posterior” circulation. Approximately
30% arise in the intracranial portion of the internal carotid artery, usually at or near the origin of
the posterior communicating artery. Another 30% occur at the region of the anterior
communicating artery. Approximately 25% arise from the middle cerebral artery, usually at its
major branch point.11-13
Posterior circulation aneurysms are more likely to be found at the tip of
the basilar artery, but may occur more proximally along its trunk.11-13
Ninety percent of
aneurysms measure less than 10mm and remain asymptomatic throughout life.10
1.2.1 Risk factors for aneurysmal subarachnoid hemorrhage
Multiple factors contribute to the formation, growth and eventual rupture of intracranial
aneurysms. Though some risk factors are inherently constitutive, hence non-modifiable; other
risk factors are modifiable, presenting opportunities to prevent or delay the formation, growth
4
and rupture of aneurysms. Modifiable risk factors are commonly classic vascular risk factors and
have been implicated in about 2/3 of cases of SAH.14
1.2.1.1 Modifiable risk factors
Cigarette smoking appears to be the single most important factor for aneurysm formation,
growth and rupture. The prevalence of cigarette smoking among SAH patients is twofold that of
the general population. According to one review, about 40% of SAH cases could be attributed to
cigarette smoking.4 Unfortunately, more than 1/3 of cigarette smokers continue to smoke after
surviving SAH; these are usually those patients who started smoking at a younger age.15
The role
of hypertension as a risk factor for the occurrence of SAH is less well established compared with
its role in other stroke types. The prevalence of hypertension among SAH patients is slightly
higher than that in the general population. A systematic review of longitudinal and case control
studies found that hypertension increased the risk of SAH by 2.5 times, and it appeared relatively
more hazardous in women.14, 16
Each 10 mmHg increase in systolic blood pressure was
associated with a 31% increase in the risk of SAH.14
Alcohol abuse increases the risk of SAH
independently of other vascular risk factors.17
Longitudinal and case control studies suggest a
10% to 50% increase in risk of SAH with consumption of ≥ 150 grams alcohol per week,
equivalent to 12 grams of alcohol. In one study, drinking 300 g/week of alcohol accounted for
20% of SAH in the population.18
1.2.1.2 Non modifiable risk factors
Non-modifiable risk factors for aneurysmal SAH include age, sex, aneurysm morphology
and positive family history. Aneurysms may form early and continue to grow throughout life,
with estimated mean time to rupture of 14 years.1 The risk of rupture increases with advancing
age, stabilizing after the age of 40 years.1, 4
Though the incidence of SAH has been estimated to
5
be 24% higher in women than men, this sex difference becomes apparent only after the age of 55
years.1 Aneurysm size, location, aspect ratio (ratio of aneurysm height to neck width), and
geometry are morphologic characteristics that have been associated with the risk of aneurysm
rupture.19
Natural history studies show increased aneurysm size, posterior circulation/posterior
communicating artery location, daughter sac and documented growth over time are associated
with rupture whereas other factors like aspect ratio and size ratio differ in observational studies
comparing ruptured and unruptured aneurysms.19, 20
The risk of rupture increases with aneurysm
size,10, 17
however the optimal cut point for rupture is uncertain. Though ruptured aneurysms are
commonly less than 10 mm, aneurysms larger than 10 mm are five times more likely to rupture
than those that are 10 mm or less.3, 17
Aneurysms that grow on serial imaging or that have
daughter sacs are at higher risk of rupture.20
A review of Japanese studies found the annual risk
of rupture was 1.5% for patients with aneurysms smaller than 10 mm and 9.3% for those with
aneurysms larger than 10 mm.21
Increasing size is a risk factor for rupture in studies of the
natural history of unruptured aneurysms.20, 22
Greater burden of vascular risk factors may lead to
rupture at smaller aneurysm size.23
These studies also show posterior aneurysms are prone to
rupture relative to anterior aneurysms, particularly in men.24
In contrast to the above factors (size, location, daughter sac, growth), a number of studies
compared the characteristics of ruptured and unruptured aneurysms. In these studies, a higher
aspect ratio has been associated with ruptured aneurysms. Different studies have used different
aspect ratio values to describe the threshold point for ruptured aneurysm, including threshold
points of 1.56, 1.6 or 1.8.19
According to Lall et al.,19
it is likely aneurysms with aspect ratios
greater than 3 are highly likely to be ruptured aneurysms, whereas aneurysms with aspect ratio
less than 1.4 are much less likely to be ruptured. As noted, these data are based on comparison of
6
ruptured and unruptured aneurysms rather than prospective follow up of unruptured aneurysms
that go on to rupture.
There is consistent evidence for inherited susceptibility to aneurysm formation, which has
been found in 10% of patients with SAH.4, 25
Consanguineous first-degree relatives of patients
with SAH have a fourfold increase in risk of SAH.26
A positive family history of aneurysm is
associated with rupture at a younger age, with larger and multiple aneurysms than in patients
with sporadic aneurysm.10, 27
A recent study developed absolute risk prediction models to
estimate incidence and life time risk of SAH based on risk factors profiles of current cigarette
smoking, positive family history, hypertension and hypercholesterolemia.28
Whether individuals
at statistically higher risk of rupture would benefit from screening is debatable and has been
suggested for those with 2 or more first degree relatives with aneurysms, identical twins, where
one twin has an aneurysm, or autosomal dominant polycystic kidney disease, though the
possibility requires assessment in cost-effectiveness studies as well as consideration of the
quality of life. Some genetic conditions increase the risk of intracranial aneurysm formation,
notably the autosomal dominant conditions polycystic kidney disease, and Ehlers-Danlos
syndrome type IV. 4, 8, 17, 21, 25
Extensive genetic studies have yet to find a common molecular
basis. Nonetheless, candidate genes implicated in genome wide linkage studies to identify
genetic loci for intracranial aneurysms include: chromosome 7q11, 14q22 and 5q22–31 in
Japanese families, chromosome 19q13.3 in Finnish families, chromosome 2p13 in Dutch
families, chromosome 1p34.3–36.13 in United States (US) families.8 Potential candidate genes
code for structural proteins of the extracellular matrix such as collagen and matrix
metalloproteinases.27
7
There is less robust evidence in support of other etiologic risk factors. One review found
inconclusive evidence in support of protective effects of white ethnicity compared with non-
white ethnicity, oral contraceptive use, hormone replacement therapy, hypercholesterolemia,
high body mass index, and diabetes in the etiology of SAH.14
Evidence further indicates that no
increased risk of aneurysm rupture is present during pregnancy, delivery and pueperium.29
One
study found a protective role of increase vegetable consumption for aneurysm rupture suggesting
that dietary factors may play a role in aneurysm pathogenesis.30
Though some studies suggest
SAH is commoner in the later hours of the morning between 6am and 12pm, on weekends and
during colder weather, 17, 31
other studies and reviews suggest the evidence for this
meteorological pattern is tenuous.32-34
1.2.2 Pathophysiology of aneurysmal subarachnoid hemorrhage
The immediate factors precipitating aneurysm rupture with bleeding into the
subarachnoid space of the brain have not been fully defined. In about 20% of patients rupture is
precipitated by a sudden increase in intramural pressure related to physical exercise and straining
or other Valsalva-like maneuvers.10
Following aneurysm rupture, blood extravasates into the
subarachnoid space triggering a sequence of pathophysiologic events that are mediated by a
multiplicity of complex, interacting, often concurrent cellular processes and molecular
mechanisms with overlapping outcomes. These events usually have been studied in the context
of angiographic arterial vasospasm and the sequelae of cerebral ischemia.35
However, the failure
of anti-vasospastic agents to prevent cerebral ischemia and improve clinical outcomes despite
markedly attenuating the appearance of angiographic vasospasm has led to a recent
reconceptualization of the pathophysiology of SAH with gradual transition from arterial spasm
and narrowing as the principal mechanisms underlying delayed ischemic injuries and poor
8
outcome to the concept that delayed ischemia is multifactorial and the pathophysiology may
include cortical spreading ischemia and microthrombosis in addition to angiographic vasospasm.
There is also the need for a better understanding of the mechanisms mediating early brain injury
(EBI), a term that has been coined to describe the series of events preceding the development of
arterial vasospasm, events which occur within the first 24-72 hours after the initial bleeding.35-39
Angiographic vasospasm, on the contrary, occurs 3 to 14 days after a single SAH, with a peak
incidence at 7-8 days. Research suggests EBI results from rapid alterations in intracranial
pressure, cerebral perfusion pressure and dysautoregulation of cerebral blood flow as the
immediate consequences of the increase in CSF volume, drainage obstruction, vasoparesis and
brain tissue hypoxia and ischemia following SAH; with the severity of physiologic derangement
reflected in the initial bleeding.40, 41
Progressive breakdown in brain tissue ion homeostasis
occurs with aneurysm rupture, affecting predominantly sodium, potassium, calcium, and
magnesium exchange; and evoking vasoconstriction of cerebral arteries and electrical changes in
brain activity such as the cortical spreading depolarization.41
These physiologic and ionic
derangements are believed to initiate blood brain barrier dysfunction, stimulate the inflammatory
and oxidative cascades that lead to global cerebral edema and ischemic injuries, and consequent
vascular and neuronal cell death.35, 41
Cell death after SAH may occur as a result of necrosis,
apoptosis and autophagy.37
Although a plethora of molecular pathways can be conceptualized as
probable mediators of EBI and angiographic vasospasm, and present potential targets for
therapeutic interventions, recent research attention has increasingly focused on the role of nitric
oxide/nitric oxide synthase pathways, endothelin-1, the antioxidant systems, platelet and
coagulation systems and the inflammatory cytokine cascades.38
9
1.3 Diagnosis and clinical presentation
Intracranial aneurysms are usually asymptomatic until rupture. SAH patients who can provide a
history complain of the classic thunderclap headache characterised by sudden onset of diffuse
headache of immediate high intensity, nausea and vomiting, and frequently loss of
consciousness. These clinical features have been reported in at least 70% of patients.42
On
admission, 50% of patients are in deep coma.10, 43
Less commonly, patients have neck stiffness;
seizures, which may be helpful to differentiate a perimesencephalic SAH from an aneurysmal
SAH;44
intraocular hemorrhage with bleeding into the vitreous humor (Terson’s sign); and focal
neurologic deficits involving the third and sixth cranial nerves.9, 10
About 40% of patients have a
warning leak or sentinel hemorrhage which usually is a preceding minor undiagnosed SAH and
increases the odds of subsequent major bleeding 10-fold.9, 25, 45
Because it typically occurs 24
hours to 2 weeks prior to bleeding, warning leak or sentinel headache provides a window for
early intervention. In the absence of classic clinical features, misdiagnoses are common
occurring in 20% of cases, and increases the likelihood of one-year mortality and morbidity in
good grade patients 4-fold.25, 46
Non-contrast cranial computed tomographic (CT) scan would be done for patients with
suspected SAH to confirm the presence of subarachnoid hematinic densities, the presence of
other concurrent intracranial hemorrhage, the likelihood of developing hydrocephalus, and to
estimate the blood burden for risk of angiographic vasospasm.4, 25
CT sensitivity is, however,
time dependent; extravasated blood being visible on CT scan images in more than 95% of
patients within 24 hours of bleeding, this proportion progressively falling to 50% after one
week.47
In suspected patients with CT negative results, xanthochromia on lumbar puncture done
preferably 12 hours after onset of headache is helpful to confirm SAH.4, 10
Despite advances in
neuroimaging technology, catheter angiography remains the diagnostic imaging procedure of
10
choice for demonstrating and localizing ruptured aneurysm; however, its invasive nature implies
that this neuroimaging modality is not without complications; specifically ischemic neurologic
injury and rebleeding may occur in about 1% of patients.4, 25
CT angiography has been reported
to have a comparable sensitivity to catheter angiography and may be sufficient alone to guide
surgical repair but arguably not endovascular repair.25
Significant advances have taken place in
magnetic resonance imaging technology but its place in the diagnostic work up is yet to be
clearly established, because of reasons that could be attributed to limited accessibility, high cost,
longer study time, the likelihood of motion artifacts, and difficulty with use in acutely ill
patients, among other reasons.25
Almost all patients with SAH who survive to hospital admission develop one or more
complications during the inpatient course; the major complications include angiographic
vasospasm, hydrocephalus, seizures, medical complications, and rebleeding. Vasospasm is the
most potentially devastating treatable complication of SAH.48
Characterised by narrowing of
cerebral arteries, commonly at multiple sites, which may or may not be associated with ischemic
symptoms, the condition occurs most frequently on the second week after rupture and resolves
spontaneous after the third week in many patients.49
The reported incidence varies widely in the
literature between 16% and 71%, most probably because authors have defined this condition
differently including use of terminologies such as symptomatic vasospasm, angiographic
vasospasm, and delayed ischemic neurologic deficits (DIND), among others.9 Mean blood flow
velocity ≥ 200cm/s in the middle cerebral artery on transcranial Doppler ultrasonography is
considered diagnostic of vasospasm, however the sensitivity is only moderate.8, 9
The
recommended term for the arterial narrowing is angiographic vasospasm. Delayed neurological
deterioration that cannot be attributed to identifiable causes is called delayed cerebral ischemia
11
(DCI or, in the past, symptomatic vasospasm). Delayed cerebral ischemia is the most important
predictor of cerebral infarction, which increases the risk of poor outcome by fivefold.50
The
recognition that CT clot burden correlates with the risk of angiographic vasospasm has resulted
in the widespread use of the Fisher grade of CT clot burden or some modification of the scale to
quantify the risk of developing vasospasm.51
Therapeutic interventions to improve cerebral blow
flow following angiographic vasospasm include the use of induced hypertensive and several
endovascular interventions, and less recommended at present, hypertensive hypervolemic
hemodilution (so called triple H therapy)25
Clinical improvements with these regimen and
interventions have been demonstrated in observational series, but their efficacy has not been
studied in randomized clinical trials. Acute hydrocephalus (occurring within 3 days of bleeding)
develops in at least 20% of SAH patients and chronic hydrocephalus (occurring after day 3) may
be seen in as many as 50% of patients.4, 9, 52
Hydrocephalus may result from CSF obstruction or
bleeding into the intraventricular space (intraventricular hemorrhage), the latter more likely
associated with neurologic deterioration. The majority of patients with acute hydrocephalus
experience clinical improvement with external ventricular drainage.53
Ventriculitis is a
complication though especially with prolonged drainage.10
About 5% of patients develop
seizures usually on the first day;54
the proportion of seizures that is truly epileptic is uncertain
and more patients develop abnormal movements likely nonconvulsive in nature. Predictors of
seizure risk include middle cerebral artery aneurysm, intracerebral hematoma, rebleeding, and
infarction, among others.25
Whether seizure worsens outcome has not been proven. Prophylactic
anticonvulsant therapy when indicated has been recommended with caution about potential
adverse effects.4
12
Medical complications are common after SAH and have been reported in about 50% of
admitted patients and are associated with poor outcome.55, 56
The commonest is fever, a marker
of systemic inflammatory state.25
A non-exhaustive list of other medical complications includes
anaemia, glucose imbalance, hypertension and hypotension, hypernatremia, hyponatremia,
cardiac failure, pulmonary edema, pneumonia and deep venous thrombosis. Management rests
on active monitoring and maintenance of a high index of clinical suspicion to identify underlying
causes for prompt treatment in the setting of multidisciplinary team anchored by intensive care
physicians. Rebleeding is the most worrisome complication of SAH and is most likely to occur
in the first few hours after SAH. It is estimated that half of rebleeds occur within 6 hours of
symptom onset, however in patients with unsecured aneurysms the incidences reduces to about
1-2% per day for the next 4 weeks.9, 25
Prognosis in patients who rebleed is poor; about 80% of
these patients die or remain disabled.10
Patients at highest risk of rebleeding include those
presenting in poor neurologic states, those receiving delayed treatment, or those harboring larger
aneurysms. Adoption of early aggressive exclusion of aneurysm has reduced significantly the
incidence of rebleeding.10
1.4 Treatment of aneurysmal subarachnoid hemorrhage
The best treatment outcomes of aneurysmal SAH are achieved at high volume centers
providing multidisciplinary management involving experienced cerebrovascular surgeons,
neurointensivists and interventional neuroradiologists, among other specialists with interest in
aneurysm treatment. Unfortunately, over 60% of hospitals treating SAH fall into the lowest case-
volume quartile.57
The primary goal of treatment is prompt exclusion of the ruptured aneurysm
from circulation once physiologic stabilization is achieved, the aneurysm causing intracranial
bleeding is identified and its anatomy and that of contiguous structures is characterised.
13
Definitive repair is provided with either surgical clip ligation or endovascular coil embolization,
preferably within 72 hours of the initial rupture to minimize risk of rebleeding. The best modality
for aneurysm repair depends on available expertise and ‘extent of interdisciplinarity’, and other
patient and aneurysm specific factors.4, 25
For patients at equipoise for clipping or coiling, the
international subarachnoid aneurysm trial (ISAT) provided evidence to support coiling as the
preferred treatment.58
Otherwise, clip ligation would be preferred for younger patients, in the
presence of space occupying lesions for example intracerebral hemorrhage (ICH), aneurysms of
the middle cerebral artery and pericallosal aneurysms and for aneurysms with wide neck and
multiple associated vessels.4 In contrast, coil embolization would be preferred for the elderly, in
the absence of a space occupying ICH, aneurysms in the posterior circulation, aneurysms with
small neck and unilobar shapes.4 There is no consensus on the optimal approach to treatment in
the poorest grade patients. A growing number of studies have been published to defend and
support early aggressive treatment with endovascular coiling or surgical clipping in all World
Federation of Neurological Surgeons (WFNS) grade V patients.59-63
Some authorities, however,
still prefer a modified conservative approach, particularly in the elderly, anchored on evidence of
some neurologic improvement during the very early phase of stabilization to justify definitive
procedures for aneurysm exclusion;63-65
an option that bears the possibility of a self-fulfilling
prophesy in the outcomes of some patients.
1.5 Prognosis of aneurysmal subarachnoid hemorrhage
Subarachnoid hemorrhage is an acute cerebrovascular event which can have devastating
effects in affected persons, their family members and society in general. Patients with SAH are at
high risk of death or residual brain injury; patients who are fortunate to survive the condition
have a lower life expectancy and quality of life than the general population.10
The mortality rate
14
of SAH varies widely across studies and region, ranging from 8% to 67%.1-7
Among
industrialized nations, median mortality rate is highest in Europe (44%) compared with North
America (32%), and lowest in Japan (27%).1-3
Relatively fewer studies have addressed the
natural history of SAH. One frequently cited study which was reported as far back as 1967
indicated a cumulative case fatality rate of 25-30 on day 1; 40-45% at the end of week 1; 50-60%
by the first month; 65% by the first year 65-70% by the end of the 5th
year.5 An estimated 12% of
patients do not survive to hospital admission.66
In the study by Pakarinen et al,5 43% of patients
did not survive the initial bleeding, most (74%) died on the first day post-SAH. However, over
the past 25 years mortality and case fatality rates of SAH have declined significantly, with the
decline attributed to earlier diagnosis of SAH, prompt aneurysm repair, improved medical
management, and use of nimodipine.67
In the US, SAH mortality rates declined an average of 1%
per year between 1979 and 1994.68
One meta-analysis of 33 prospective, population based
studies published between 1995 and 2007 reported a 17% decline in case fatality rate between
1973 and 2002.69
In contrast, the average age of affected persons has increased from 52 years to
60 years.68, 69
Declining case fatality rates and increasing incidence in the elderly has raised
concerns about the functional status of SAH survivors. In a meta-analysis by Nieuwkamp et al,69
data on functional outcome were found in only 6 of the 33 studies included in the analysis, and
these data showed that on average 55% of patients were independent for activities of daily living
within one year of SAH. Patients who achieved functional independence may still experience
considerable neurocognitive impairment. In a review of 61 studies reporting cognitive and
functional outcomes of patients with SAH, Al Khindi and colleagues7 noted large variations in
the prevalence of anxiety, depression, impaired quality of life, memory, executive functioning
and other neurocognitive deficits. Their review found that patients with SAH showed frequent
15
impairment in the cognitive domains of memory, executive function and language. Residual
cognitive deficits may persist in the long term contributing to impairments in social roles and
lack of satisfaction with quality of life long after the acute event. The proportion of years of
potential life lost from SAH is reported to be similar to that of ischemic stroke and intracerebral
hemorrhage.6, 68
Attention has scarcely focused on the experience of significant others of patients who
suffered an SAH, and on the societal burden of the condition. Many family members of patients
with SAH have been reported to experience symptoms of psychosocial distress and sleep
dysfunction; in one study caregiving partners reported reduced quality of life 18 months after
their partner’s SAH, especially in the domains of emotional behaviour, social interactions, work;
with half of the caregiving partners who had a job working shorter hours or in positions of less
responsibility or losing their job after their partners had SAH.70
Many significant others express
fears about a possible recurrence of SAH in their partners; an emotion that could hamper the
ability to care for the affected loved one, potentially limiting recovery of patients from the
sequelae of SAH.71
The economic burden of SAH is enormous and underappreciated. A study in
the United Kingdom reported that in 2005, SAH resulted to 80,356 life years and 74,807 quality
adjusted life years lost. The economic burden associated with SAH in the United Kingdom was
estimated as 510 million pounds annually.72
In the US, the mean hospital charge for SAH
corrected for inflation has been estimated at $65,000 per patient.73
1.6 Prognostic Factors in aneurysmal subarachnoid hemorrhage
Prognostic factors are measureable patient, disease and treatment characteristics that are
associated with subsequent clinical outcome in people with a given health condition.74
For any
given disease condition, these factors could be multiple extending over a broad spectrum of
16
characteristics including demographic factors such as age, sex, race/ethnicity, socioeconomic
status; markers of structural damage and disease progression for example measures of clinical
severity, and biologic markers identified from metabolomics, proteomic, genetic, genomic and
epigenomic research, among other factors. Prognostic factors could play important roles at
different positions along the continuum of the translational pathways to improved outcomes.74-77
Understanding which factors are prognostic in a given condition could be helpful to distinguish a
group of people with a different average prognosis in order to inform and redefine the disease
diagnosis, to monitor changes in disease status and treatment response over time so as to better
inform treatment recommendations and individualize patient management, and potentially to
lead to better understanding of pathophysiology and treatment by identifying outliers and such.
Prognostic factors could be potential modifiable targets for therapeutic interventions and serve as
building blocks for prognostic models, risk scores or prognostic scores which serve a variety of
roles in clinical practice and research. Because prognostic factors are potential confounders that
may mask treatment effect, accurate estimation of the magnitude of their associations with
outcome is fundamental to aiding the design and analysis of interventional studies, for example
for optimal use of stratification or minimization randomization in the design or use of statistical
adjustment in the analysis of randomized clinical trials.74-77
Researchers have investigated
prognostic associations in SAH to identify potential prognostic factors, quantify the relative
magnitude of their associations with outcome, and combined the information to develop
prognostic models, risk scores or prognostic scores in SAH. Though a plethora of factors have
been identified to be prognostic in SAH, but upon review, the majority have major and multiple
limitations including small sample sizes (often a few hundred or less patients), single center
retrospective reviews, no multivariable analysis or if done, no inclusion of key suspected
17
prognostic factors, and many others. Their prognostic value has yet to be adequately estimated.
Even for widely studied conventional prognostic factors, considerable knowledge gap and
disagreements exist in the literature. Nevertheless, a body of evidence supports a relation
between some prognostic factors and outcomes of patients with SAH, but for other prognostic
factors the relation is less well established.
1.6.1 Demographic factors
Increasing age has been identified to be associated with poorer outcomes.78-83
This
finding most likely reflect the prevalence of extracerebral comorbidities among the elderly and
reduced plasticity of the aging brain for optimal response to the primary brain injury.84, 85
Many
studies analysed the effect of age using different approaches, particularly the application of
threshold values to describe age effect on outcome; which suggested the possibility of change
points in the prognostic effect of age around which the outcomes of patients may worsen with
increasing age.86-92
The different approaches adopted in the literature to summarize the effect of
age is shown in Table 1.1. Unfortunately the use of multiple threshold values by most previous
studies has made the comparison of prognostic associations across studies very challenging.
Hence, the precise value of age for prognostication in SAH cannot be gleaned from the literature.
In an old case controlled study investigating the prognostic effect of hypertension and age in 319
cases of SAH among veterans who were discharged from a single hospital in the US in the
period 1958 through 1962, age was categorized into 5 categories, including under 36 years, 36-
44 years, 45-54 years, 55-64 years, and 65 years and older.93
Studies reported thereafter have
used different categorizations including use of decades, or quartiles of age, or dichotomization of
age into 4, 3 or 2 categories. No reasons were given for the categorization of age as was done.
18
Deruty et al.79
investigated the role of level of consciousness and age as prognostic factors in
SAH, and applied a cut-off point value of 50 years in describing the prognostic effect of age. The
authors reported better prognosis in patients who were younger than 50 years compared with
those who were 50 years or older. A single center study evaluating predictors of 12 month
outcome on the modified Rankin score in poor grade SAH patients dichotomized age using the
threshold value of 65 years and reported a 6-fold increase in the hazard ratio of poor outcome in
poor grade patients who were ≥ 65 years compared with those who were younger than 65 years.
The small number of patients analysed in this study (98 patients) may have informed the decision
to collapse age as was done.94
Studies focusing on elderly subgroup of patients have also applied
different cut-off point values to define the elderly in whom prognosis is expected to markedly
worsen. This population has been defined using upper limit age values of 50 years in older
studies of over 2 decades ago86
, to 60 years65, 78, 87
, 65 years88, 95
, 70 years85, 90, 96, 97
or even 75
years.91, 92
The focus of the studies were usually to ascertain optimal treatment approaches in this
population. There is some evidence to indicate that the prognostic effect of age may differ with
levels of neurologic status. Some studies have noted interaction effect between age and
neurologic status assessed on the WFNS scale82, 98, 99
which was not the finding of other
studies.78
There is also evidence to suggest that the prognostic effect of age may differ with
duration of time to surgery.82
Though SAH is commoner in women than men, whether sex
differences exist in prognosis is doubtful.17, 100
Recent review indicating improved survival after
SAH over the last two decades also reported that the improvement is unevenly distributed over
sex, drawing attention to the need for a better understanding of the role of sex differences to
SAH outcomes.101
19
Studies evaluating the relationship of race/ethnicity to SAH outcomes have documented
higher mortality rates in “non-white” populations compared with “white” populations,100, 102-109
however whether this finding is due to higher incidence of SAH or greater case fatality in non-
white populations compared with white populations is uncertain. Most studies reported crude
incidence and case fatality rates. Apparent from the literature is inconsistency in the
categorization of race/ethnicity across studies. Epidemiologic studies comparing black
populations with white populations have found double incidence rates and mortality rates from
SAH in blacks compared with non-Hispanic whites. Available data were predominantly those
from population based studies in the Cincinnati and New York metropolitan areas of the United
States. Researchers reviewed medical and autopsy records, and CT scan of all patients with a
likely diagnosis of intracerebral or SAH in the greater Cincinnati area between 1988 and 1989.107
They reported that African Americans were 2.1 times at higher risk of stroke than their white
counterparts. The incidence rate in African Americans who were 75 years or older was 25%
higher than that of whites in same age bracket. A different study which reviewed hospital records
in the greater Cincinnati metropolitan area in 1994 found the incidence of SAH to be 80% higher
in African Americans than in whites.110
Case fatality rates were statistically similar between the
two populations. Eden et al. investigated ethnicity differences in the processes of care between
African Americans and whites in the greater Cincinnati area from 1997 to 2005 and reported
there were no differences between the two groups in time to treatment of ruptured aneurysm.111
Comparative studies including data on Hispanic populations have reported higher incidence risk
ratios in Hispanic groups compared with non-Hispanic whites, which were slightly lower than
the incidence risk ratios seen in African Americans. The prospective, population based Northern
Manhattan Study (NOMAS)103
compared SAH incidence rate and case fatality rate between
20
whites, African Americans and Hispanic groups and found African Americans and Hispanics had
higher incidence risk ratios than whites but similar case fatality risk to whites. Compared with
whites, the incidence risk ratio in African Americans was 1.6 (95% confidence intervals; 0.8-
2.8); in Hispanics, it was 1.3 (95% confidence intervals; 0.7-2.3). African Americans were found
to have 3.5 times higher odds of death within 30 days than whites (95% confidence intervals,
0.4-28.4). The odds ratio for Hispanics was 0.9 with 95% confidence intervals of 0.2-5.4. The
NOMAS study identified only 52 hospitalized cases of SAH in the Manhattan area from July 1,
1993, to June 30, 1997; comprising 9 cases in whites, 9 cases in African Americans and 34 cases
in Hispanic populations; a sample size that is too small to reliably estimate the risk differences.
In a relatively higher powered study, researchers reviewed data on 107 SAH patients from the
population based Brain Attack Surveillance in Corpus Christi (BASIC) project and compared
Mexican Americans with non-Hispanic whites in Southeast Texas.100
The authors found
significantly higher incidence risk ratio in Mexican Americans (1.68; 95% confidence intervals,
1.10-2.38) compared with non-Hispanic whites. Ethnicity differences were not seen in in-
hospital mortality risk and outcome at discharge measured with the modified Rankin scale. The
researchers suggested a possible role for genetic heterogeneity as a likely cause of the risk
difference. However, the study did not provide comparative data on other race/ethnic groups, and
neither did the study provide data comparing risk factors profile of Mexican Americans to non-
Hispanic whites.
The literature suggests a different epidemiologic pattern of SAH in populations of Asian
nativity but the prognostic implications are uncertain. Asians have been reported to have smaller
aneurysms; higher incidence of SAH, suggesting aneurysm rupture at smaller size; and some of
the lowest case fatality rates reported.112, 113
Studies comparing incidence and outcomes of
21
populations of Asian nativity to other race/ethnic groups were mostly conducted in the United
States. There is the possibility of confounding due to the effect of acculturation. Researchers
analysed the United States all-payor Nationwide Inpatient Sample administrative dataset for the
year 1997 to investigate stroke incidence in patients of Asian/ Pacific Islander (API) nativity
compared with non-Hispanic whites.109
APIs were found to be younger, had higher
socioeconomic status, and statistically higher risk of SAH than non-Hispanic whites. The two
populations had statistically similar case fatality rates. Ayala et al.105
investigated stroke
mortality rates among Hispanics, African Americans, American Indians/Alaska Natives, APIs
and whites by age and sex from 1995 to 1998 using the US National Vital Statistics’ death
certificates data. They found higher age-standardized death rates (per 100,000 persons) in
African Americans, American Indians/Alaska Natives, and APIs compared with non-Hispanic
whites for all stroke subtypes. Only a few studies have examined differences in outcomes among
racial/ethnic groups while simultaneously accounting for factors that could potentially confound
the race/ethnicity association with mortality.114
The studies used data on patients who were
enrolled in trials114
or patients in a small geographic cohort.115, 116
The study populations were
more commonly recruited prior to substantial changes in treatment practice, such as early
aneurysm repair and endovascular coiling, which have significantly impacted case fatality. The
conclusions of the studies were inconsistent. For instance, a post hoc analysis predominantly
comparing African-Americans to whites who were recruited into a large clinical trial between
1991 and 1997 found no significant race/ethnic differences in 3-month functional outcome.114
A
different study based on the New York database of hospital discharges in 2003 reported that
white patients with SAH in New York City had better functional outcomes than non-white
patients.115
22
Though socioeconomic factors are frequently implicated as contributors to overall
population health117, 118
and disease-specific outcomes,119
insight is limited as to the role of
socioeconomic factors in SAH. At least two reviews have synthesized evidence pertaining to the
nature of the association between socioeconomic status (SES) and outcomes of patients with
stroke with no data provided on SAH.120, 121
The possibility of socioeconomic differences in
outcomes of SAH was suggested by a study examining the burden of stroke in African
Americans in a US population,106
and a prospective population-based study which examined
socioeconomic patterns in the incidence, mortality and prognosis of SAH in Finland.122
The
former study reported that mortality outcome after SAH is related to household income among
minority populations in the Los Angeles county area but not in white populations.106
The latter
study demonstrated that young adult men with lowest personal income in Finland had higher
case fatality rates than their counterparts with higher personal income.122
One implication of the
findings of both studies may be that the relation of SES to SAH, if any, transcends national
health care systems. Furthermore, insight into the nature of the association between
socioeconomic status and SAH outcomes may be important given that many risk factors for SAH
occur in the context of socioeconomic differences that are present over the life course in many
populations; the nature of the association may further elucidate what role environmental factors,
and differential access to and receipt of the processes of care play in the outcomes of patients
with SAH.
1.6.2 Clinical factors
Patient neurologic status is the single most important indicator of the severity of brain
injury soon after SAH and is critical to treatment decisions and prediction of outcome.60, 123, 124
Neurologic status is generally considered the strongest indicator of prognosis after SAH.
23
However, when to assess, and what measures to use to assess neurologic status for purposes of
prognostication is debatable.125, 126
Available grading scales were commonly developed on the
basis of expert opinion and their measurement properties have not been sufficiently studied.
According to Deruty and coworkers,79
the Botterell grading system which is a 5 categorical scale
is probably the oldest grading scale in SAH. A novel 5 categorical scale was proposed by
Nishioka et al. in the first co-operative study of intracranial aneurysms.127
In 1968, a
modification of the Botterell scale was proposed by Hunt and Hess which over the years has
become widely accepted and adopted for assessing prognosis though the scale was originally
aimed at predicting appropriate timing for surgery after SAH.128
Despite its popularity, the Hunt
and Hess scale has been shown to have moderate inter rater reliability, most probably because of
the vague definitions of the different terms used to define the grades. Moreover, research
evidence has been provided which question the ability of the scale to differentiate outcomes of
patients with SAH. Modifications of the Hunt and Hess scale are available, including the Hunt
and Kosnik scale, among others; however their uptake into clinical practice lags behind the
traditional Hunt and Hess scale.129
The World Federation of Neurological Surgeons (WFNS)
scale is commonly used to assess neurologic status at hospital admission and is now increasingly
used in interventional studies evaluating the efficacy of new therapies.60, 123, 124
The WFNS scale
was developed by a panel of experts as a response to the limitations of prior scales.130
The scale
essentially compressed the Glasgow coma scale (GCS) into 5 categories and adds to this the
presence or absence of focal motor deficits. Some studies have reported poor intergrade
correlation of WFNS scale with outcome, which was not the finding of other studies.123
Many
risk scores have been developed for SAH patients that essentially extend the WFNS scale by
incorporating other clinical as well as neuroimaging characteristics. According to some studies, a
24
simple re-categorization of the GCS using different threshold values may be as useful as the
WFNS scale or the Hunt and Hess scale for measuring clinical severity for prognostication.131-134
Takagi and coworkers135
proposed a new scale called the Prognosis on Admission of
Aneurysmal Subarachnoid Hemorrhage (PAASH) scale which compressed the 13 categories of
the GCS into 5 categories based on statistical parameters, including PAASH grade I (GCS Score
15); grade II (GCS Scores 11–14); grade III (GCS Scores 8–10); grade IV (GCS Scores 4–7);
and grade V (GCS Score 3). Subsequent validation studies comparing the PAASH scale to the
WFNS and Hunt and Hess scales indicated the PAASH scale had better inter observer
agreement with weighted kappa of 0.65 compared with WFNS (Kappa 0.60) and Hunt and Hess
scale (Kappa 0.48).133
A different validation study found similar discriminative ability between
the PAASH scale and WFNS scale though the former had a more graduated relationship to
outcome than the WFNS scale.131
The European Stroke Organization guideline for management
of SAH has recommended the PAASH scale for assessment of neurologic status after SAH.4
Researchers have also revised the GCS into a new scale called the Poor Grade GCS (PGS)
system for predicting long term outcome in SAH patients with poor grade status.134
The
breakpoints were: PGS-A (GCS 10–12); PGS-B (GCS 8–9); PGS-C (GCS 5–7); PGS-D (GCS
3–4). The PGS was reported to predict one year outcome, unlike the Hunt and Hess scale, WFNS
scale and the GCS which did not; though no indication was given as to the accuracy of the
prediction.134
Julien et al. also revised the GCS to develop a new scale called the GCS grading
system with categories that are slightly different than the PAASH scale, and include grade I
(GCS 15); II (GCS 12-14); III (9-11); IV (GCS 6-8) and grade V (GCS 3-5).132
In the study, the
GCS grading system strongly correlated with outcome than the WFNS and Hunt and Hess scales.
Whether the newer scales will replace the WFNS and Hunt and Hess scales as gold standards in
25
the foreseeable future given their simplicity and better inter rater reliability will depend on the
results of further validation studies and the willingness of practitioners to adopt them.
The commonest comorbid condition seen in patients with SAH is hypertension,93, 136-138
but conflicting results are present in the literature as to the nature of its association with
outcome.82, 137-143
Premorbid hypertension has been defined differently in different prognostic
studies, in some inclusive of admission blood pressure values138
which are often elevated soon
after SAH and hence may not necessarily be reflective of premorbid values. Among the earliest
sufficiently powered studies to investigate the prognostic role of premorbid hypertension was the
study by Keller who examined 319 matched patients with aneurysmal SAH receiving treatment
between 1958 and1962. The study found no correlation between the disproportionately higher
incidence of hypertension in black patients and higher mortality rate.93
A prospective Danish
aneurysm study of 1076 patients who were managed between 1978 and 1983 found that patients
with a premorbid history of hypertension had significantly higher mortality rate than those
without hypertension (59% versus 49%) two years after SAH, but not within 30 days of the
insult.144
Contrary to that study, a different Danish study of 118 consecutive patients admitted
into a single tertiary hospital over a three year period found SAH patients with a history of
hypertension had significantly higher mortality rates within 14 days of the insult compared with
those without hypertension (61% vs. 32%).137
In a population based study of 824 patients
admitted to hospitals in the United Kingdom between 1992 and 1996, no association was found
between history of hypertension and mortality at 1, 7 and 30 days.145
However, in the
international cooperative aneurysm study admission blood pressure and premorbid medical
comorbidity were independently associated with 6-month mortality in multivariable analysis but
the magnitude of the risk was not quantified.139
Furthermore, premorbid history of hypertension
26
was one of twenty one comorbid conditions that were investigated and it was unclear whether the
reported effect of admission blood pressure was independent of that of premorbid hypertension.
A twenty year follow up study of 142 Finnish patients diagnosed with unruptured intracranial
aneurysm between 1956 and 1978 reported significantly higher risk of death at rupture in SAH
patients with “definitive hypertension”, which was defined as systolic blood pressure greater
than 160 mmHg or diastolic blood pressure greater than 95 mmHg or use of antihypertensive
drugs.138
The estimated odds ratio associated with the risk of fatal SAH was 12.67 with very
wide 95% confidence intervals between 1.53 and 104.70. Retrospective analysis of the tirilazad
trials dataset by Rosengart and coworkers82
demonstrated elevated admission systolic blood
pressure ≥ 190 mmHg and history of hypertension were both independently predictive of risk of
unfavorable outcome on the GOS at 3 month. Rosen and Macdonald141
reanalysed same dataset
and develop a new risk score for predicting 3-month unfavorable outcomes incorporating
admission systolic blood pressure as well as premorbid history of hypertension as predictor
variables. In contrast, a recursive partitioning tree model developed from 885 patients enrolled
into the cooperative study of intravenous nicardipine failed to incorporate premorbid history of
hypertension as one of the factors predictive of 3-month unfavorable outcome on the GOS.143
Similarly, relatively recent smaller observational studies from single tertiary hospital settings did
not find a history of hypertension independently predictive of unfavorable outcomes142
or long
term survival.94
The prognostic relevance of other comorbid conditions in SAH remains largely
understudied. In the study by Rosengart et al.,82
a history of myocardial infarction and liver
disease were independently associated with poor outcome at 3 months.
Following ruptured aneurysm exclusion by surgical or endovascular procedures, efforts at
improving prognosis are geared towards preventing or managing neurologic and non-neurologic
27
complications which may significantly impact outcome. These complications could be indicative
of or could be the consequence of progressive brain injury or reflect the burden of extracerebral
organ dysfunction. Though complications occurring during the inpatient course may be expected
to provide considerable prognostic information, active therapeutic interventions to mitigate their
adverse effects significantly confound their prognostic value. This scenario plays out very well in
the case of rebleeding which though being the worse complication of SAH has been
inconsistently identified as a prognostic factor.139, 146-148
It is likely that the low prevalence of
rebleeding in most study cohorts, and the time dependent nature of the complication which
makes it less easily handled in multivariable logistic analysis may have been contributory as
well. The occurrence of new infarcts during hospitalization has been consistently identified as a
marker of poor outcome.82, 149, 150
Similarly, studies have shown that the presence of delayed
cerebral ischemia and angiographic vasospasm independently predicts poor outcome.82, 141, 151
The roles of other neurologic complications, such as seizures and hydrocephalus, as prognostic
factors are less established. Studies examining the prognostic role of extracerebral organ
dysfunction have commonly focused on cardiopulmonary events, which have been shown to
correlate with the risk of poor outcome;152, 153
though data is available also showing that other
extracerebral complications do affect outcome and could have added prognostic relevance.149, 150
A single center study including 787 consecutive patients with SAH found patients who were at
risk for renal failure were twice as likely to have a poor outcome at 3 months than those who
were not at risk for renal failure.150
A prospective series investigating determinants of poor
outcome 12 months after SAH in 534 Japanese cohorts showed that while complications
occurring after hospitalization did not predict mortality at 12 months, they significantly predicted
poor outcome on the modified Rankin scale at 12 months.149
28
1.6.3 Neuroimaging factors
Neuroimaging characteristics also have been suggested to be predictors of complications
after SAH and aneurysm repair as well as predictors of clinical outcome. 83, 94, 149, 154-159
Subarachnoid clot volume, density and distribution seen on computer tomographic (CT) scan
images have been established as prognostic factor for angiographic vasospasm.51, 160
The
relationship between the location and amount of subarachnoid blood and the development of
angiographic vasospasm was codified by Fisher et al.161
into a scale which has been applied in
many studies to evaluate the association between CT clot burden and clinical outcome in SAH
with variable results reported. The Fisher scale has been criticised for its subjectivity and poor
reproducibility in an era of advanced high resolution CT imaging.51
Researchers have developed
modifications of the scale and alternative methods of quantifying CT clot burden to improve
predictive performance; but the newer scales are not without their own limitations as well.51, 160,
162 Publications by Japanese investigators show they were aware of the association of
subarachnoid blood clots (so-called high density areas) and subsequent development of
angiographic vasospasm, and indeed considered risk classification systems for its prediction
before Fisher.163
The Fisher scale continues to be routinely used. Aneurysm morphology may be
a predictor of peri-procedural complications, and potentially a predictor of outcome.164
Aneurysms larger than 10 mm and those with wide neck have been associated with higher rates
of thromboembolic events; whereas middle cerebral aneurysms have been associated with
elevated risk of intraoperative rupture during endovascular repair.164
Larger diameter aneurysms
have been correlated with elevated risk of rebleeding.155
Despite consistent evidence of a
relationship between aneurysm morphology and risk of peri-procedural complications, evidence
of a direct relationship of aneurysm morphology and clinical outcome appears conflicting; some
studies suggesting poorer outcomes with larger diameter aneurysm and posterior circulation
29
aneurysms relative to anterior circulation aneurysms,141, 149, 165
other studies not finding such a
relationship. 82, 83, 94, 157, 159, 166, 167
The use of different threshold values to describe the effect of
ruptured aneurysm size further confound any attempt at synthesising evidence from the
literature. 82, 83, 94, 157, 159, 166, 167
1.6.4 Physiologic factors and biomarkers
Interest in the use of biomarkers, including laboratory variables, to assess prognosis has
been growing in recent years with investigators in SAH increasingly focusing on the prognostic
role of markers of cardiac dysfunction, brain injury and oxidative stress, several of which have
been shown to be univariably associated with outcome; though the independent effect is
uncertain and their added incremental predictive value is unknown. A meta-analysis of studies
investigating the prognostic role of electrocardiographic abnormalities and biochemical markers
of cardiac dysfunction168
found evidence in support of a significant association between elevated
risk of death and echocardiographic wall motion abnormality, elevated troponin levels, brain
natriuretic peptide levels, tachycardia, presence of Q waves, ST segment depression, T-waves
abnormalities and bradycardia. In the same meta-analysis, poor outcome was associated with
elevated troponin levels, creatine kinase MB levels and ST-segment depression. The study
concluded that unlike in ischemic heart disease, QT prolongations and ST segment elevations
were not significantly associated with death or poor outcome in SAH. Another meta-analysis
concluded that the expression of the apolipoprotein APOE4 allele is associated with poorer
outcome in patients with SAH.169
Indeed, several other biomarkers with prognostic potential
have been identified including cytokines such as tumor necrosis factor-α, and interleukin 6; the
calcium binding protein S100β; endothelial markers such as endothelin-1; among many others.170
30
The systematic reviews also found possible bias in reported prognostic associations of
biomarkers in SAH. First, the primary studies were found to have been published over a wide
time period during which the diagnosis and treatment of SAH have improved with consequent
decline in poor outcome. Second, the numbers of subjects studied were almost always small,
only in few studies exceeding 50 patients.168, 169
Third, the reference values of studied
biomarkers were scarcely provided. Fourth, the systematic reviews found that inconsistencies are
prevalent in the definition of threshold values across studies. Fifth, the relative contribution and
incremental prognostic value of biomarkers associated with SAH outcome was rarely
investigated. Sixth, identified biomarkers have not been validated. It is noteworthy that
biomarkers are often costly, requiring invasive procedures to obtain; their pathophysiologic
mechanisms or role in SAH have not been fully elucidated.
1.7 Limitations of prognostic studies
Research in other clinical conditions clearly indicates that studies reporting prognostic
associations are usually inadequately powered and the findings less generalizable because of the
limited representativeness of the individual study populations.171-174
Issues exist about data
quality and appropriateness of certain prevalent statistical approaches adopted to investigate
prognostic associations.171-174
A review of studies reporting prognostic factors in complex
regional pain syndrome found few studies to be prospective in design.171
The reviewed studies
were found to include very few patients, and researchers in this condition scarcely applied
reliable and validated measures to assess prognostic factors. Based on the reviewers’ quality
assessment criteria, the review found 75% of included studies to be of poor quality. The
reviewers advocated for large prospective cohort studies to more precisely determine the value of
prognostic factors in complex regional pain syndrome. A different review aimed at assessing the
31
methodological quality of primary studies identifying factors that were predictive of the outcome
of activities of daily living after stroke found 88% of studies were of low quality.172
Included
studies were assessed on a 27-item quality assessment checklist evaluating 6 domains of bias
including study participation, study attrition, predictor measurement, outcome measurement,
statistical analysis, and clinical performance and validity. Only 6 of the 48 studies that were
reviewed had a score of 20 out of a maximum of 27 points. Study attrition rate was reported in
77% of studies, the reason for which was stated in 67% of the studies. The review observed that
missing data were inadequately handled in most studies (67%). Other limitations that the
reviewers identified included the application of cut-off point values to predictor variables and
outcome measures without a clear rationale for why this was done; non reporting of the crude
estimates of effects of prognostic factors in univariable analysis. Approximately 65% of studies
dichotomized predictors, such as age, which are continuous in nature. A systematic review of
prognostic models in patients with acute stroke showed that none of the available models as at
the time of the review had been sufficiently rigorously developed and validated for possible
application to clinical practice or research purposes.175
The similarity of the findings to earlier
reviews in 1986, 1989, and 1996, in the opinion of the authors, indicated a lack of improvement
over time in the quality of research on the development of prognostic models. One major
limitation was the small number of patients included in most studies. A review of liver
transplantation literature noted methodological flaws in the building of prognostic models and
poor quality of reporting.176
All reviewed models did not discriminate well between patients who
died and those who survived after liver transplantation. Similar review of cancer prognostic
studies 174
made observations along same lines; including use of inappropriate methods for
development of multivariable models and poor reporting. Although several reviews have
32
identified poor study quality as the bane of prognostic research, there is some evidence that with
increasing awareness study quality is improving, particularly for those conditions where
prognostic research has been long established. For instance, whereas a relatively older review
indicated the use of small patient samples, inefficient analysis approaches for model
development in traumatic brain injury,177
a more recent review found methodological
improvements in more recently published models in traumatic brain injury, including the use of
large patient cohorts in the development of more recent models and improvements in other
aspects of study design. 178
1.8 Prognostic models in SAH: How reliable are available models to predict
outcome of patients with SAH?
Prognostic factors are building blocks for prognostic models which mathematically
combine a set of prognostic factors to predict the risk of future clinical outcomes in patients.
Prognostic models have multiple applications in clinical practice and research. In clinical
practice, prognostic models could be useful as decision support tools providing empirical
evidence complementing clinicians expert opinion about prognosis, related decisions about
treatment goals; and facilitating evidence based communication with patients and significant
others on outcomes expectations and goals for rehabilitation.75, 77
In research, prognostic models
may be useful to examine the prognostic distributions of patient populations to gain insight into
differences in case-mix of studied populations and compare outcomes among practice centers,
which has potential for clinical audit and benchmarking; enrol patients into randomized clinical
trials (RCTs) based on their baseline prognostic risk, so called prognostic targeting.179
Prognostic
models may be useful in the analysis phase of RCTs to adjust for baseline differences in
prognostic factors, which has been shown, along with prognostic targeting, to result in increased
33
statistical power, potentially enabling reduction in sample size.75, 77, 179
Finally, identification of
patients or subgroups with outcomes that differ from those predicted in outcome models could
identify genetic or other factors that are important in the pathophysiology of SAH, thus
potentially leading to research on new treatments.
Given the potential applications of prognostic models, expectedly several studies have
been published reporting prognostic models in SAH. To our awareness, no studies have been
published synthesizing evidence on the average effects of prognostic factors in SAH, and
providing evidence on the relative or added incremental value of prognostic factors in SAH. No
prognostic model or risk scores in SAH is universally adopted into clinical practice. The question
as to why this is so remains unanswered. It was earlier noted that research in other diseases has
identified substantial methodological flaws in the development and reporting of prognostic
models, which may negatively influence model translation into clinical practice or for research
purposes. Insight into the likely reasons for the limited use of prognostic models in SAH may be
helpful to guide further prognostic research in SAH and guide the development and facilitate the
translation of prognostic models in SAH into clinical and research application.
Therefore, as part of this research project, a systematic review was conducted of such
studies with the objectives to identify prognostic factors that are commonly included in
prognostic models in SAH, and to evaluate the methodological approaches adopted to develop
the models to ascertain the reliability and generalizability of available models. The review
conformed to recommendations contained in the preferred reporting items for systematic reviews
and meta-analyses (PRISMA) publication and utilized validated methods to minimize bias.124
The review was limited to studies reporting prognostic models for mortality or functional
outcomes, studies including at least 2 prognostic factors in the models, studies presenting new
34
models or validating previously reported models. The review excluded studies that aimed to
explore the predictive value of a single prognostic factor, studies that aimed to rank prognostic
factors, without presenting a model to predict mortality or functional outcome. Also excluded
were studies that developed prognostic models for secondary outcome events such as
angiographic or DCI, cerebral infarction, seizures, or hydrocephalus; studies that reported
models which were derived only from expert opinion; and lastly studies that presented models
for traumatic SAH and perimensencephalic SAH.
Relevant studies were identified by a search of MEDLINE through PUBMED, EMBASE
through OVID, and Web of science, without language restrictions. The search strategy
(Appendix A) was designed for high sensitivity rather than high specificity using previously
validated strategies.180
Since there have been substantial changes in the management of patients
with SAH in the last 2 decades; literature search was restricted to studies published between
January 1, 1995 and June 30, 2012; and updated to include studies published up to December
2013 (Table 1.2). For additional studies, further reviewed was performed of the reference lists of
the studies that met the inclusion criteria.
1.8.1 Results of systematic review of prognostic models in SAH
The initial search yielded 2,745 records that were screened by title and abstract to
identify 69 articles as potentially relevant, which were then read in full. Eleven studies satisfied
the inclusion criteria. The 58 excluded studies were for reasons related to: risk factors identified
and ranked for the outcome of interest, but not proceeding further to present prognostic model (n
= 30); general models not specific to SAH (n = 5); presented or validated or compared single
predictor scales or their modifications (n = 14); prognostic model for unruptured aneurysms (n =
35
1); complications and secondary events (n = 6); and an artificial neural network or graphical
chain model presented (n = 2).
In terms of target populations, 2 studies proposed models to predict unfavorable outcome
in patients with poor-grade SAH94, 142
and 2 developed models based on physiologic parameters
or a combination of clinical predictors and biomarkers of brain injury.146, 157
The remaining 7
models are applicable to SAH patients of any clinical severity.83, 141, 143, 158, 159, 167, 181
The
objectives of the research were specifically stated to predict outcome in all grades of patients in 4
studies,141, 158, 159, 181
whereas the others studies did not specify the target population. Some
prognostic models were developed on predominately good-grade patients.83
In 8 studies, models
were developed from retrospective analysis of single center datasets from hospitals in the US,
Spain, and France. In 3 studies, the models were developed from datasets of phase 3 randomized
clinical trials: the International Subarachnoid Aneurysm Trial (ISAT), the tirilazad trials, and the
cooperative aneurysm study of intravenous nicardipine after aneurysmal SAH (NICSAH). Data
for the 11 studies were collected between 1983 and 2006, and average duration of data collection
was 6 years (range 2–10 years, Table 1.3).
The multivariable nature of prognostic research makes estimation of required sample size
challenging. Simulation studies suggest the number of events observed per candidate predictors
considered in the analysis provides a good approximation of study power.182
A commonly used
rule of thumb is for each candidate predictor studied at least 10 events are required to produce
reasonably stable estimates from logistic regression models. Studies reporting prognostic models
in SAH included a median of 441 patients (interquartile range: 149–733). However, only the
studies developing models from RCT datasets were sufficiently powered, having included at
least 10 events for each candidate predictor considered to develop the prognostic models.
36
A wide range of predictors were considered for model development (range: 5–89 candidate
predictors), the median number was 12. All studies considered baseline demographic, and
clinical factors, and usually basic admission radiology findings. Three studies included pre-
existing medical conditions.94, 141, 142
The final models, on average, contained 5 predictors
(range: 3–8); the most frequently retained predictors in the prognostic models were age (n = 8),
Fisher or modified Fisher grade (n = 6), World Federation of Neurological Surgeons (WFNS)
grade (n = 5), aneurysm size (n = 5), or Hunt and Hess grade (n = 3). When age was included, it
was dichotomized in 6 of the 8 studies. In the other 2 studies, it was unclear how age was coded
in the final model. Physiologic and biochemical measurements were considered for model
development in 4 studies.94, 142, 146, 157
The additional biochemical variables were retained in the
final models in 3 studies, including heart-fatty acid binding protein, S100B, nucleoside
diphosphate kinase A, ubiquitin fusion degradation protein-1, and troponin 1 in one study,157
admission glucose in 2 studies,94, 146
and arterio–alveolar gradient and bicarbonate in another.146
Although most studies reported using baseline hospital data, the time at which predictor variables
in the final model were obtained was described in 8 studies. The primary outcome measure was
the Glasgow outcome scale (8 of 11 studies) or the modified Rankin scale (3 of 11 studies). The
outcome was assessed 3–12 months after SAH. Models were used to predict mortality (2 of 11
studies) or unfavorable outcome (severe disability, vegetative, or death on Glasgow outcome
scale or modified Rankin score of 4–6) in 9 of the 11 studies (Table 1.1).
Though no consensus exists on the optimal statistical strategy for developing prognostic
models, growing consensus are developing as to what is good practice at key steps of the
modelling process. Because missing data is a common problem in medical research, how missing
data were handled should be explicit.183
In all reviewed studies, investigators discarded patients
37
who did not have data on all candidate predictors and the outcome measure (complete case
analysis), an approach that could potentially lead to biased estimates where the proportion of
patients excluded is substantial or the development sample is small. The approach adopted to
select predictors into the model is important. Two approaches are prevalent: one approach is
based on the use of automated statistical selection algorithms; the other approach is based on a
priori subject knowledge of the prognostic value of the candidate predictors. A third approach
which combines the 2 methods is possible.182
How predictors were selected into prognostic
models was uncertain in 5 studies; in the remaining 6 studies predictor selection was performed
by use of automatic selection algorithms using stepwise logistic regression with either backward
elimination or forward selection based on some nominal significance level of probability (Table
1.4). Where study power is low, as was the case in most reviewed studies, this approach may
result to exclusion of low prevalence but clinically important predictors such as rebleeding,
overestimation of regression coefficients (overfitting), and likelihood of worse predictions in
new patients, also called optimism.182, 184
The predictive accuracy of new prognostic models require validation in new cohorts
(external validation) using adequate performance measures to assess how well the model
differentiates between patients who develop the outcome and those who do not (discrimination),
and how close the model`s predicted risks are to the actual observed risks (calibration).182
The
model may be tested in the development cohort to assess the predictive accuracy in the
underlying population from which the development cohort was derived (internal validation).
Discrimination is commonly evaluated with the area under the receiver operating curve and its
equivalent, the c-statistics; and calibration is commonly assessed using a calibration plot and a
goodness of fit test, frequently the Hosmer Lemeshow test. The systematic review found no
38
studies providing external validation of available prognostic models. For internal validation, the
adopted methods were bootstrapping (n = 1), cross-validation (n = 1), split sampling (n = 2), or
by apparent validation (n = 5, Table 1.5). Model performance was reported by calibration in 2
studies, with goodness of fit test in the range of 0.65–0.86, or by discrimination in 7 studies with
area under the receiver operating characteristics curve or c-statistics values in the range 0.70–
0.86. In 4 studies, model performance was reported in terms of the proportion of patients
correctly classified by the model. The prognostic models were presented as score charts (in 5
studies), normogram (in 1 study) and as regression formula in 2 studies (Table 1.4).
1.8.2 Implications of findings of the systematic review
Despite the narrow scope of the review and its other potential limitations, it provided
useful insight to guide further research on prognostic associations in SAH, though it also
illustrated several important limitations that potentially constrain the reliability and
generalizability of available prognostic models in SAH. The review showed that the most
common factors associated with outcome are admission neurologic status (10 of 11 studies), age
(8 of 11 studies), and amount of blood seen on admission CT scan (6 of 11 studies). Other factors
were less consistently identified as such; pre-existing hypertension, vasospasm on admission
angiography, intraventricular hemorrhage, aneurysm location and size, method of aneurysm
repair, and some laboratory measurements. The prognostic factors identified in the review cover
the relevant domains for prediction of outcome in SAH, and could be useful to inform a priori
selection of predictors in further research to develop reliable prediction models. Some studies
provided insight into the relative strength of prognostic factors. In the study by Rosengart et al.,82
the factors contributing most to variation in outcome, in descending order of importance, were
cerebral infarction, neurological grade as measured on the WFNS scale, age, temperature on day
39
8, intraventricular hemorrhage, vasospasm, SAH, intracerebral hematoma, and history of
hypertension. The retention of on average 5 predictors in the prognostic models enhance the face
validity of the models, as probability estimates are easier to compute with a smaller number of
predictors.
However, the reliability of available models, that is their ability to produce unbiased
estimates of the probability of outcomes in new cohorts, is constrained by many limiting factors.
One major limiting factor is the lack of study power with the related problem of overoptimistic
estimates of model predictive ability, as seen in the rather high discrimination performance of
many models in the development cohort despite the small sample size. Models with high
optimism perform poorly when applied in new cohorts.182
Most studies were retrospective raising issues about data quality. Prospective cohort
studies are ideal as they allow for pre-specification of patient selection and variable definition.182
Datasets from large clinical trials have the advantage of study power, with the limitation that the
cohorts may not fully represent the variety of patients and management practice seen in the
condition. Suboptimal statistical approaches were often adopted that could negatively impact on
model reliability. Two major challenges were identified. One was the reliance on statistical
algorithms for predictors screening to the exclusion of a priori subject knowledge of the
prognostic effect of candidate predictors. The other was the use of inefficient approaches to
assess model internal validity, use of inadequate performance indicators, and insufficient insight
into the calibration behaviour of reported models. The lack of explicit consideration of a priori
subject knowledge for predictor selection into prognostic models is a major limitation because
subject knowledge is crucial to creating a credible prediction model in SAH since complications
such as DCI, cerebral infarction, and rebleeding that are strong predictors of poor outcome can
40
be eliminated in the final model when automated procedures are used, particularly if these
complications are of low frequency in the datasets or if they are modeled using logistic
regression which less effectively treats these time-dependent covariates.182
Though a few studies
examined the internal validity of the reported models, inefficient approaches were commonly
adopted such as split sampling which reduces study power and produces biased estimates when
applied in small development samples.174
The bootstrap technique is increasingly recommended
for internal validation since it reflects the process of sampling from an underlying population and
could be used to reliably assess optimism in model predictive performance.182
Incidentally, this
technique was used in just one study.83
The performance of available models were generally poorly assessed, for example
calibration, an index relevant to application of models in clinical setting was infrequently used.
Calibration plots were scarcely provided, which would have allowed for visual examination of
the full range of predicted versus observed outcomes. The 2 studies providing data on model
calibration relied on the Hosmer-Lemeshow goodness of fit test, as do most studies reporting
prognostic models in the literature. The Hosmer –Lemeshow goodness of fit test may not be the
best measure of model calibration: the test has limited power to detect model misspecification or
overfitting of the effects of predictors.182
When applied to the development sample, a non-
significant test has no real implication for model application as it (non-significant finding)
reflects only a lack of fit in the development sample. Experts now recommend assessing two
measures of calibration that address key issues in validation. These are calibration-in-the-large
and the recalibration slope.182
Calibration-in-the-large reflects the difference between the average
of observed outcomes and the average of predicted outcomes, and corresponds to the intercept of
41
the regression model refitted with the linear predictors only. Recalibration slope is the slope of
the refitted model using the linear predictors only.
There are also issues about generalizability of the models to new patient populations.
External validation of prognostic models has seldom been performed. The development data
were to varying extent less representative of the broad spectrum of SAH patients, as they were
derived from single hospital settings, or clinical trial populations, or represent patients with good
neurologic grade.
In summary, the systematic review identified prognostic factors that are commonly
included in prognostic models in SAH. It showed that though a number of prognostic models
have been published to predict outcomes of patients with SAH, the use of suboptimal
methodologies are prevalent in model development, which could limit the reliability and
generalizability of available models in SAH.
1.9 Conclusion
Subarachnoid hemorrhage from ruptured intracranial aneurysms is a life threatening acute
cerebrovascular event requiring emergent management. Though generally uncommon, the
condition has considerable impact and burden on young adults who are at their prime with
respect to family commitments and career. To a large extent, the pathophysiology has not been
fully defined; hence efforts in search of effective therapies and interventions to prevent SAH or
considerably improve the outcomes of patients are yet to yield commensurate value. Prognosis
therefore remains poor, particularly with respect to cognitive outcomes and quality of life among
patients who survive SAH. Early recognition of the likely trajectory of outcome could be helpful
to inform treatment decisions. This has been demonstrated by the recognition that reduction in
time to aneurysm exclusion and an aggressive approach to patient management minimizes the
42
risk of life threatening complications and improves outcome. Research has identified a plethora
of potential outcome predictors in SAH that could help advance our understanding of the
condition and further redefine our approach to patient care. The prognostic information has been
harnessed by previous studies to develop tools for early prediction of outcome. However, quality
of studies reporting prognostic factors and proposing prediction schemes has been largely
suboptimal, raising uncertainties about study conclusions and requiring replication and
confirmatory re-examination of the prognostic value of several factors, and further attempts at
developing reliable outcome prediction tools using more optimal approaches. The body of
research presented in this thesis was aimed at achieving this objective.
43
Table 1.1 – Approaches adopted by previous studies to analyze age effect on outcome
Author &Year N Study design Outcome
measure
Follow up Threshold
for age
Stepwise
categories
Continuous
Keller 1970 319 Single center
(USA)
mortality 5 years 5 categories
Fortuny et
al.,1980
256 Single center mortality discharge 2 categories
Rosenorn et al.,
1987
1076 Multicenter
(Denmark)
Survival/ mental
status
2 years decade
Kassell et al.,
1990
3521 multicenter GOS 6months decades
Inagawa 1993 503 Single center
(Japan)
GOS 6 months 4 categories
O’Sullivan et
al., 1994
186 Single center
(Scotland)
mortality 40 months ≥60 years
Deruty et al.,
1995
74 Single center
(France)
mortality discharge 2 categories
Lanzino et al.,
1996
906 multicenter GOS 3 months quartile
Chung et al.,
2000
129 Single center
(USA)
GOS 1 months ≥70 years
Johansson et
al., 2001
281 Single center
(Sweden)
GOS 3-6 years ≥65 years decade
Lagares et al.
2001
294 Single center
(Spain)
GOS 1 mo. Post-
discharge
quartiles
Horiuchi et al.,
2005
509 Single center
(Japan)
GOS discharge ≥70 years yes
Rabinstein et
al., 2004
81 Single center
(USA)
Rankin score Last follow
up
decades
Kazumata et al.,
2006
168 Single center
(Japan)
GOS 1, 3 months yes
Mocco et al.,
2006
98 Single center
(USA)
Rankin score 12 months 2 categories
Nieuwkamp et
al., 2006
170 Two centers
(Netherland)
GOS Discharge,
2-4months
≥75 years 2 categories
Pereira et al.
2007
386 Multicenter
(RCT)
GOS 6, 12 months
Rosengart et al.,
2007
3567 Multicenter
(RCT)
GOS 3 months yes
Salary et al.,
2007
133 Single center
(USA)
GOS 6 months yes
Ryttlefor et al.
2008
278 Multicenter
(RCT)
Rankin score 2 months ≥65 years 4 categories
Coghlan et al.,
2009
588 Multicenter
(RCT)
GOS 3 months yes
Langham et al.,
2009
2397 Multicenter
(UK, Ireland)
GOS 6 months yes
Sacco et al.,
2009
118 Multicenter
(Italy)
Mortality 7days, 1, 12
months
3 categories
Awe et al.,
2011
150 Single center
(USA)
GOS discharge ≥70 years
Risselada et al.,
2010
2128 Multicenter
(RCT)
Rankin score 2 months decades
Brinjikji et al., 88930 Multicenter mortality discharge 4 categories
44
2012 (NIS)
Degos et al.,
2012
933 Single center
(France)
Rankin score 1 year 3 categories
Karamanakos et
al., 2012
1657 Single center
(France)
Mortality 1 year 3 categories
Chotai et al.,
2013
108 Single center
(Korea)
GOS 6, 12 months 20-39
years
yes
Scholler et al.,
2013
256 Single center
(Germany)
GOS 1-5 years ≥60 years 4 categories
45
Table 1.2 – Studies identifying independent predictors of poor outcome in multivariable analysis
Author &
Year
N Study design Outcome
Measure
Candidate predictors Independent Predictors in
multivariable model
Model for
prediction?
Model
Validation
Model
performance
Model
Presentation
Naval et
al.,185
2013
1134
Single center,
Surgery &
endovascular
cohorts, (USA)
In hospital
mortality
Age, sex, race,
Hunt & Hess grade, WFNS grade,
admission Glasgow coma score,
major medical comorbidities,
Intraventricular hemorrhage, Intracerebral
hemorrhage, global cerebral edema, and
hydrocephalus
Age, GCS, one or more
major comorbidities
Yes Apparent AUC=0.82 Score chart
Degos et
al.,,186
2012
526
Single center,
endovascular
series only,
(USA)
1-year
mortality on
Rankin score
Age, sex, Glasgow coma score,
presence of motor deficit, presence of
clinical seizure, and WFNS
grade, S100β, Troponin-I, hydrocephalus
and intraventricular
hemorrhage, aneurysm location and size
Glasgow Coma Score,
S100β, Troponin-I
Yes Split
sampling
AUC=0.76 Score chart
Karamana
kos et
al.,187
2012
1657
Single center,
Surgery &
endovascular
cohorts,
(Finland)
Mortality at
12 months
Age, sex, saccular IA family aneurysm,
time period of aneurysmal SAH, Hunt &
Hess grade, Intracerebral hemorrhage,
intraventricular hemorrhage, subdural
hematoma, hydrocephalus, Site of
ruptured saccular IA, size of ruptured
saccular IA, 2 Saccular IAs
Age, Hunt & Hess grade,
intraventricular
Hemorrhage, saccular IA,
ruptured saccular IA on
the internal carotid artery
or the basilar artery
bifurcation,
hydrocephalus
No
Taki et
al.,149
2011
534
Multicenter,
Surgery &
endovascular
cohorts,
(Japan)
Modified
Rankin Score
(0-2 versus 3-
6) at 3 & 12
months
Age, sex, WFNS grade, Fisher grade,
rerupture, date of aneurysm obliteration,
interval from admission to aneurysm
obliteration, symptomatic vasospasm,
vasospasm-induced cerebral infarction,
cardiopulmonary dysfunction, infection,
shunt-dependent hydrocephalus, seizure,
ileus, femur fracture, acute renal failure,
dome size, neck size, size classification,
aneurysm location, repair modality,
clipping-related complication, coiling-
related complication, antiplatelet therapy,
CSF drainage, meningitis
Age, admission WFNS
grade, Preadmission
aneurysm rerupture,
Vasospasm-induced
cerebral infarct, Infection
(pneumonia, sepsis),
Shunt-dependent
hydrocephalus, Seizure,
Post-clipping
hemorrhagic
complication, Post-coiling
ischemic complication
No
O’kelly et
al.,188
2010
3120
Multicenter,
Surgery &
endovascular
cohorts,
(Canada)
Time to death
or
readmission
for SAH
Age, sex, Charlson index, aneurysm size,
ventilated, hydrocephalus, aneurysm
location, associated Intracerebral
hemorrhage, repair modality
Age, sex, coiling,
Charlson index,
Intracerebral hemorrhage,
aneurysm size, ventilated,
hydrocephalus
No
Risselada 2143 Multicenter, Mortality Age, sex, prior SAH, Fisher grade of CT Age, WFNS grade, Fisher Yes Bootstrap AUC= 0.7 Regression
46
et al.,83
2010
Surgery &
endovascular
cohorts, (ISAT
trial data)
(Modified
Rankin score)
at
2 months
blood, lumbar puncture findings, WFNS
grade, number of aneurysms, Aneurysm
location, and size, randomization group,
vasospasm on admission angiography
grade, aneurysm size resamplin
g
formula
Zacharia
et al.,150
2009
787
Single center,
Surgery &
endovascular
cohorts, (USA)
Modified
Rankin score
(1-3 versus 4-
6) at 3 and 12
months
Age, sex, premorbid Rankin score, Hunt
& Hess grade, pre-existing renal disease,
pre-existing diabetes, Admission serum
Creatinine level, Admission creatinine
clearance, Peak creatinine, creatinine
clearance decrease ≥25%, clinical
vasospasm, new infarct secondary to
vasospasm, total contrast-enhanced
imaging studies
Age, admission Hunt &
Hess grade, Total scan,
premorbid Modified
Rankin score ≤1, risk of
renal failure, new infarct
secondary to vasospasm,
diabetes
No
Coghlan
et al.,189
2009
588
Multicenter,
surgical series
only, (IHAST
trials data)
GOS (1-4
versus 5; and
ordinal) at 3
months
Age, sex, WFNS grade, Fisher grade of
CT blood, NIHSS score, aneurysm
location, and size, history of
hypertension, history of coronary artery
disease, time from SAH to surgery
Age, WFNS grade,
aneurysm location, and
size
No
Langham
et al.,190
2009
2397
Multicenter,
Surgery &
endovascular
cohorts,
(UK & Ireland)
GOS-
extended (1-4
versus 5-8) at
6 months
Age, sex, WFNS grade, amount of blood
on CT scan, Aneurysm location, and size,
concurrent medical condition, pre-repair
deterioration
Age, WFNS grade,
amount of blood on CT
scan, aneurysm size,
concurrent medical
condition, pre-repair
deterioration
No
De
Toledo et
al.,158
2009
441
Single center,
(Spain)
GOS (1-3
versus 4-5) at
6 months
40 Variables, not specifically stated Age, WFNS grade, Fisher
grade
Yes Apparent AUC=0.82 Nomogram
Guresir et
al.,191
2008
585
Single center,
Surgery &
endovascular
cohorts,
(Germany)
modified
Rankin Score
(0-2 versus 3-
5) at 6 months
Age, Sex, Hunt & Hess grade, Glasgow
Coma score, aneurysm location, size, and
configuration, Intracerebral hemorrhage,
early hydrocephalus, Time to treatment,
rebleeding
Age, time to treatment,
early hydrocephalus,
Intracerebral hemorrhage
No
Rosengart
et al.,82
2007
2695
Multicenter,
surgical series
only, (Tirilazad
trials data)
GOS (1-3
versus 4-5) at
3 months
Age, sex, race,
admission Glasgow coma score,
WFNS grade, admission systolic and
diastolic blood pressures, time from ictus
to admission, admission to surgery,
weight, and body temperature, History of
hypertension, myocardial
infarction, diabetes, liver disease, thyroid
disease, migraine headaches, previous
SAH,
symptomatic vasospasm, cerebral
infarction, hydrocephalus, pulmonary
Age, WFNS grade, SAH
thickness on
initial CT, aneurysm
location, and size,
IVH, ICH,
systolic blood pressure,
history of hypertension,
prior SAH, myocardial
infarction, liver
disease, temperature on
day 8 ≥38°C,
anticonvulsant use, rescue
No
47
edema, repair modality, anticonvulsant
use, cerebral angioplasty, use of rescue
therapy, aneurysm location, and size,
therapy, symptomatic
vasospasm, cerebral
infarction
Claassen
et al.,146
2004
586 Single center,
(USA)
Modified
Rankin
scale (1-3
versus 4-6) at
3 months
Age, sex, ethnicity, tobacco use,
alcohol use, loss of consciousness,
seizure at ictus, Glasgow
coma score, Hunt & Hess grade, Hijdra
grade of CT blood, hydrocephalus,
cerebral infarction, global cerebral
edema, rebleeding, delayed cerebral
ischemia, aneurysm location, and
size, admission angiographic vasospasm,
repair modality, use of rescue therapy,
systemic inflammatory response
syndrome,, arterio–alveolar gradient,
blood urea nitrogen, serum glucose,
albumin, creatine kinase,
troponin-I, physiologic subscore of
APACHE-2 score
Mean arterial
blood pressure, admission
glucose, arterio–alveolar
gradient, bicarbonate
Yes Split
sampling
AUC=0.79 Score chart
Rosen et
al.,141
2004
3567
Multicenter,
surgical series
only, (Tirilizad
trials data)
GOS (1-3
versus 4-5) at
3 months
Age, sex, race, history of hypertension,
liver disease, diabetes
mellitus, admission systolic & diastolic
blood pressures,
WFNS grade, time from SAH to
admission, admission
temperature, whether the patient was alert
at admission, clot
thickness on CT scans, intracerebral
hemorrhage, intraventricular hemorrhage,
aneurysm location, and size, vasospasm
on initial angiograms, and hydrocephalus
Age, WFNS grade,
systolic blood
pressure,
history of
hypertension, Fisher
grade,
aneurysm size,
aneurysm
location,
vasospasm on
admission
Yes Split
sampling
AUC=0.78 Score chart
Lagares
et al.,80
2001
442
Single center,
Surgery &
endovascular
cohorts, (Spain)
GOS 1-2
versus 3-5 at 3
months
Age, Hunt & Hess grade, WFNS grade,
Glasgow Coma Score,
aneurysm size, Fisher grade of CT blood,
aneurysm location
Age, WFNS, Glasgow
Coma Score, Fisher
grade, aneurysm size
Yes Apparent AUC=0.81 Score chart
Ogilvy &
Carter,181
1998
409
Single center,
surgical series
only, (USA)
GOS (ordinal)
at mean time
3.5 years
Age, Hunt & Hess grade at the time of
surgery, aneurysm location and size,
density of haemorrhage assessed by the
Fisher Scale
Age, Hunt and Hess
grade, Fisher grade, and
aneurysm size
Yes Assessed
in new
cohort
unclear Score chart
Germanso
n et al.,143
1998
885
Multicenter,
surgical series
only,
(Cooperative
study of
Nicadipine data)
GOS (1-3
versus 4-5) at
3 months
Age, sex, pre-existing hypertension,
aneurysm location, and size,
CT clot thickness, serum glucose,
Glasgow coma score, level of
consciousness
Age, Glasgow coma
score, Basilar aneurysm
location
Yes Not
reported
Not reported Regression
formula
48
Niskanen
et al.,151*
1993
929
Single center,
surgical series
only,
(Finland)
GOS 1-2
versus 3-5 at 1
year
Age, systolic blood pressure, co-existing
disease, Hunt & Hess grade, Aneurysm
location, and size, vasospasm on
angiography, multiple aneurysms, amount
of blood on CT, intracerebral hematoma,
intraventricular bleeding, hydrocephalus
Age, Hunt & Hess grade,
amount of blood on CT,
intraventricular bleeding,
vasospasm on
angiography
Yes Split
sampling
AUC plot
presented
Regression
formula
Jagger et
al.,192
1989
3521
Multicenter,
surgical series
only, (cooperative
study data)
GOS (1-2
versus 3-5),
and mortality
at 6 months
Eye opening, speech, orientation,
response to commands, motor response,
meningeal signs, cranial nerve
involvement
Eye opening, motor
response
Yes Split
sampling
Error rate
≤ 2%
Score chart
Niskanen: study population consisted of patients with ruptured aneurysm
Table 1.3 – Characteristics of studies reporting prognostic models in SAH
Data
collection
Patients
(N)
Development
sample (N)
Study design Outcome measure endpoint Outcome
events
Candidate
predictors
Events per
variable
Risselada et al.83
1994-2002 2143 2128 Multicenter (Europe, North
America), prospective clinical trial
Rankin scale , death
at 2 months
Death vs.
alive
153 11 14
Turck et al.157
2002-2006 199 28 Single center, prospective (France) GOS at 6 months 1-3 vs. 4-5 50 9 6
De Toledo et al.158
1990-2001 441 441 Single center, retrospective (Spain) GOS at 6 months 1-3 vs. 4-5 38 40 1
Salary et al.167
2003-2004 133 133 Single center, retrospective (US) GOS at 6 months 1-3 vs. 4-5 43 12 4
Mocco et al.94
1996-2002 148 98 Single center, retrospective (US) Rankin scale at 12
months
0-3 vs. 4-6 42 26 2
Lagares et al.159
1990-2001 442 Unclear Single center, retrospective (Spain) GOS at 3 months 1-3 vs. 4-5 unclear 5 unclear
Claassen et al.146
1996-2002 586 413 Single center, prospective (US) Rankin scale at 3
months
1-3 vs. 4-6 167 38 5
Rosen et al.141
1991-1997 3567 1794 Multicenter (Europe, North
America, Australia, Zealand, South
Africa), prospective clinical trial
GOS at 3 months 1-3 vs. 4-5 539 20 27
Ogilvy et al.181
1990-1994 409 Unclear Single center, prospective (US) GOS at 12 months 1-3 vs. 4-5 11 5 2
Germanson et
al.143
1987-1989 885 733 Multicenter (North America),
prospective clinical trial
GOS at 3 months 1-3 vs. 4-5 256 9 28
Le roux et al.142
1983-1993 159 158 Single center, prospective (US) GOS at 6 months 1-3 vs. 4-5 61 89 1
49
Table 1.4 – Approach to model development in previous studies reporting prognostic models in
SAH
Handling of
missing data
Coding of
variables
stated
Analysis
model
Selection of
predictors in
final model
Final Model presentation Model
presentation
Risselada et al.83
Complete
case
Yes Logistic
regression
Backward with
AIC
Coefficients, intercept and
confidence intervals
Regression
formula
Turck et al.157
Complete
case
Yes Logistic
regression
Backward with
uniform p value
Coefficients and confidence
intervals
None
De Toledo et al.158
Not reported Yes Logistic
regression
Backward with
uniform p value
Coefficients and confidence
intervals
normogram
Salary et al.167
Not reported Yes Logistic
regression
Unclear Coefficients and confidence
intervals
Prognostic score
Mocco et al.94
Not reported Unclear Logistic
regression
Backward with
uniform p value
Coefficients and confidence
intervals
Prognostic score
Lagares et al.159
Complete
case
Yes Logistic
regression
Unclear Not reported Prognostic score
Claassen et al.146
Complete
case
Yes Logistic
regression
Forward
selection
Coefficients and confidence
intervals
Prognostic score
Rosen et al.141
Compete
case
Yes Logistic
regression
Backward with
uniform p value
Coefficients only Prognostic score
Ogilvy et al.181
Not reported Yes Logistic
regression
Unclear Coefficients only Prognostic score
Germanson et
al.143
Not reported Yes Logistic
regression
Unclear Coefficients and intercept Regression
formula
Le roux et al.142
Complete
case
Yes Logistic
regression
Unclear Coefficients and confidence
intervals
None
Table 1.5 – Approach to model validation in previous studies
Approach to
validation
Calibration Performance Discrimination Performance
Risselada et al.83
Bootstrap H-L GOF p=0.86 C-statistics 0.70
Turck et al.157
Cross validation Not reported Not reported Sensitivity/
specificity
68% and 100%
De Toledo et al.158
Apparent H-L GOF p=0.65 AUC 0.86
Salary et al.167
Apparent Not reported Not reported AUC 0.82
Mocco et al.94
Apparent Percent
predicted
Not reported Not reported Not reported
Lagares et al.159
Apparent Not reported Not reported AUC 0.81
Claassen et al.146
Split sample Not reported Not reported AUC 0.79
Rosen et al.141
Split sample Not reported Not reported AUC 0.78
Ogilvy et al.181
Temporal Percent
predicted
Multiple p
values
Not reported Not reported
Germanson et al.143
Not reported Not reported Not reported Not reported Not reported
Le roux et al.142
Apparent Percent
predicted
71% Not reported Not reported
50
Chapter 2
Study Methodology
A part of this chapter is adapted from a manuscript that has been accepted for publication as
follows:
Jaja BN, Macdonald RL, Cusimano MD, Etminan N, Hanggi D, Hasan D, Ilodigwe D, Lantigua
H, Le Roux P, Lo B, Louffat-Olivares A, Mayer S, Molyneux A, Quinn A, Schweizer TA,
Schenk T, Spears J, Todd M, Torner J, Vergouwen MD, Wong GK, Singh J; for the SAHIT
Collaboration. Subarachnoid Hemorrhage International Trialists Repository: Advancing clinical
research in aneurysmal subarachnoid hemorrhage. Neurocritical care
51
2.1 Introduction
This thesis consists of analyses performed on large cohorts of international patients with
SAH using approaches that are relatively novel in the context of prognostic studies in SAH. This
chapter sets out the study rationale, the different hypotheses that were tested and the framework
of the study design and statistical methods adopted. It provides a description of the study
populations and methods that are common to all chapters. Any deviations and more specific
details are noted in the individual chapters.
2.2 Study rationale: Towards providing higher level evidence of prognostic
associations and better prognostic models in SAH
There is scope for improvement of primary studies reporting prognostic associations in
SAH, fundamentally the need for greater study power to more precisely estimate the magnitude
of prognostic associations, and the need for more representative samples to examine the
generalizability of findings across different settings, so as to provide higher level evidence on the
nature and extent of prognostic associations and the value of prognostic factors for predicting
outcome. One approach to address the issue of study power and lack of representative data is the
establishment of collaborations which aim to pool individual patient data from multiple
prospective, well conducted, ethically approved studies from different settings, and to use the
data to address questions relating to, and test hypothesis pertaining to the predictive value of
potential prognostic factors. This methodology affords greater power beyond that of individual
primary studies and provides opportunity to examine consistency of prognostic associations
across a broad spectrum of settings; patient case mix, practice pattern, geographic and temporal
settings.193-195
The large size and heterogeneity of pooled data from multiple sources could be
harnessed to develop prediction models which are more likely to be reliable and applicable to
different clinical settings, hence more useful for clinical and research purposes.196
52
It is possible that the less than optimal analysis approaches adopted in previous
prognostic studies in SAH reflect prevalent practice at the time the studies were conducted.
Indeed, suboptimal study design and methodologies are prevalent in medical literature; and the
need for improvement in the quality of the design of and the analysis adopted in prognostic
research is increasingly being advocated. Advances in statistical methods and software
development now affords the opportunity to deal with some common issues that may lead to
confounding and bias, issues such as missing data, nonlinearity of predictor effect, among others.
Researchers now recognize the merit of pooled analyses of individual patient data from
multiple sources. Investigators in traumatic brain injury have accrued datasets from many
clinical trials and organized epidemiologic studies in traumatic brain injury conducted over a
period of 20 years, and conducted pooled analyses of individual patient data in the resource to
more precisely estimate the magnitude of prognostic associations, ascertain the added
incremental value for several conventional prognostic factors in traumatic brain injury, and
explored the relation of some latent factors to outcome, as well as developed better prognostic
models for predicting outcomes of patients with traumatic brain injury.197, 198
Similar approaches
have been adopted in ischemic stroke, and in other conditions.199
The successful adoption of
pooled analysis to address questions relating to the value of prognostic factors in similar
conditions to SAH demonstrate the feasibility of such approach (pooled analyses in large
heterogeneous samples from multiple sources) in SAH to confirm previously reported prognostic
associations, more precisely estimate the magnitude of prognostic effect, addressing knowledge
gaps and disagreements that are present in the literature, and explore hitherto latent prognostic
factors in SAH.
53
In a systematic review, we identified factors that have been most consistently shown in
previous studies to predict the outcome of patients with SAH.124
Most are conventional
demographic, clinical, and neuroimaging parameters information on which are usually readily
obtained at hospital admission. Despite the multiple replication studies in the literature
examining the relation of these factors to outcome in SAH, their value for prognostication has
not been sufficiently determined. Considerable gaps still exist in our knowledge of the nature and
the extent of their effect, which requires further research to fill in. It is not unlikely that greater
insight into the value of prognostic risk factors could help inform and refine diagnosis and our
understanding of the pathophysiology of SAH, individualize patient care, and help us develop
tools to aid research and advance patient care. The current absence of a reliable tool to inform
clinical decision about outcomes of patients with SAH represents a significant management
lacuna; more so as the variable course of the condition potentially confound clinicians’ ability to
estimate outcome based on their experience only. The availability of an objective tool which
reliably predicts patient outcome could therefore have profound effect on current patient care
standards in SAH. Moreover, the prevalence of negative trials in SAH may require a rethinking
of the approaches to conducting and analysing randomized clinical trials in SAH, to which end
objective prediction tools may facilitate. As currently known prognostic risk factors do not fully
account for the variability in patient outcome, the search for latent prognostic factors continues.
A part of this thesis will therefore ask and address questions relevant to understanding the role of
socioeconomic status (SES) and race/ethnicity as latent prognostic factors in SAH. The rationale
being that most preventable etiologic risk factors for the occurrence of SAH occur in the context
of socioeconomic differences that are present through the life course; socioeconomic differences
have been implicated as the single most important determinant of health outcomes.
54
Race/ethnicity differences in outcome may also shed some light on the pathobiology of SAH.
Better understanding of the role of SES and race/ethnicity differences in outcome could be
helpful to design interventions relevant to improving outcomes for all populations.
The overall goal of this research project is to provide a better understanding of prognostic
associations in SAH by investigating the relationship between some known and latent prognostic
factors and outcomes of SAH using large cohorts of patients who are representative of diverse
case mix, practice, geographic and temporal settings in SAH. Outcome prediction tools shall be
developed for translation of the knowledge into clinical and research applications. The overall
goal shall be achieved by pooled analyses of individual patient data of SAH patients whose
information are contained in large clinical and administrative repositories.
2.3 Hypothesis and research contributions
Chapter 3: Primary studies reporting prognostic associations in SAH are often
insufficiently powered, use data of limited representativeness and scarcely examine the added
value of prognostic factors above those of other known factors. Hence, considerable knowledge
gaps and conflicting results exist in the literature as to the nature and extent of prognostic
associations in SAH even for those risk factors that have been widely studied. This chapter
investigated the relation between 7 conventional factors and clinical outcomes in SAH using
individual patient data in the Subarachnoid Hemorrhage International Trialists (SAHIT)
repository. The objective of Chapter 3 is to determine the role of patient’s age, sex,
premorbid history of hypertension, admission neurologic status, admission Fisher CT clot
burden, aneurysm location and size as independent risk factors for clinical outcome
following hospital admission for SAH.
To achieve the above objective, the following specific hypotheses were tested.
55
1. That a change point is present in the prognostic effect of age and aneurysm size
2. That the effect of prognostic factors differ between patients who are enrolled into
randomized clinical trials and unselected hospital cohorts
3. That each conventional factor has added incremental predictive value above those of
other factors
Individual participant data (patient-level) meta-analyses were conducted to investigate
univariable association between each predictor and 3-month Glasgow outcome score in studies in
the SAHIT repository. Multivariable analyses were performed to determine the adjusted effect of
each predictor on sequentially accounting for the effects of other predictors as adjustment
factors. Prognostic associations were estimated over the ordered categories of Glasgow outcome
score using proportional odds models. Nagelkerke’s R2 was used to measure the added
prognostic value of the predictor above those of the adjustment factors. The analyses were
conducted separately for each of the 7 prognostic factors. The results most probably provide a
higher level of evidence than any previous studies of the prognostic value of the 7 prognostic
factors studied in this chapter.
Chapter 4: Many prognostic models or risk scores have been reported to predict the
outcome of patients with SAH. However, none is routinely used in clinical practice or for
research purposes. Major constraints are lack of evidence as to the reliability or generalizability
of available models in SAH. The use of suboptimal methods to develop prognostic models in
SAH leaves much room for improvement. The objective of this chapter is to develop better
clinical prediction models in SAH. Therefore, chapter 4 tests the hypothesis that prognostic
models based on the readily obtained parameters at hospital admission of patients age,
premorbid hypertension, neurologic status on the WFNS scale, Fisher CT clot burden,
56
aneurysm location and size, and envisioned treatment modality will have adequate
discrimination and calibration in a large, heterogeneous development sample, and potential
to perform satisfactorily in new cohorts. A novel set of prognostic scores were developed from
patient information in the SAHIT repository to predict 3 months mortality or unfavorable
outcome in patients with SAH. Given the large size of the development cohort and the broad
spectrum of patients and practice settings represented, the scores could have wide application for
clinical practice and research purposes.
Chapter 5: Known prognostic factors in SAH explain only a small proportion of the
variance in outcome indicating a need to further explore other possible determinants of and
associative factors for poor outcome in patients with SAH. Whether SES is a predictor of patient
outcome after SAH is uncertain. Studies have documented race/ethnic differences in SAH
mortality rates, but whether the mortality differences are due to only differential prevalence of
SAH or are also related to differences in case fatality among race/ethnic groups is uncertain. The
objective of Chapter 5 is to investigate whether socioeconomic status, measured as
neighbourhood income status, race/ethnicity are independently associated with patient
outcome after hospitalization for SAH.
The hypotheses tested in this chapter include:
1. That SES measured as neighbourhood income status is associated with the risk of
inpatient mortality after admission for SAH.
2. That SES measured as neighbourhood income status is associated with the secondary
outcome of discharge to institutional post-acute care after admission to acute care.
57
3. That the association between SES and the outcomes studied, if any, is not related to the
type of health care system under which care is provided (universal insurance coverage
versus private insurance coverage).
4. That race/ethnicity differences are present in the risk of inhospital mortality following
hospitalization for SAH.
5. That race/ethnicity status is associated with the pattern of discharge to institutional post-
acute care.
Chapter 6: This final chapter provides a general discussion of the research work detailed in
the thesis. The contributions of the thesis to a better understanding of prognostic associations in
SAH is discussed, as well as the translational implications of the body of work for clinical and
research applications. Chapter 6 further examined the limitations of the research and directions
for moving forwards.
2.4 Research Ethics Board Approvals
The research reported in this thesis was conducted within the framework of 2 research
ethics board approvals obtained from the St. Michael’s Hospital, Toronto, ON, Canada:
REB#12-021, REB#10-357.
2.5 Study population
Patient population for the research was derived from 3 sources: the Subarachnoid
Hemorrhage International Trialists (SAHIT) repository, the Discharge Abstract Database (DAD)
of the Canadian Institute for Health Information, and the Nationwide Inpatient sample (NIS) of
the Healthcare Cost and Utilization Project by the Agency for Healthcare Quality and Research
(Rockville, MD), USA. The SAHIT repository is a collaborative effort of investigators who are
58
interested in SAH research to pool and centrally warehouse anonymized individual patient data
from rigorously conducted, ethically approved randomized clinical trials (RCT)58, 200-205
and
well-designed prospective observational studies and hospital registries in SAH206-209
for purposes
of exploring novel hypothesis with the potential to advance the design, conduct and analysis of
RCTs in SAH. The repository presently contain data on 11 443 SAH patients derived from 14
datasets, of which 9 are datasets of clinical trials and 5 are datasets of prospective hospital
registries or observational studies. Some characteristics of the original studies and registries are
shown in Table 2.1. The RCT studies were multicenter phase II, III and IV trials. Patient
enrolment took place in 4 continents over a 14 year period (1997-2011). All trials did not find
statistically significant differences between patients with or without the trial target intervention,
except for the ISAT trial; enabling the pooling of individual patient data of both trial arms with
an indicator variable to identify patient trial arm. The prospective hospital registries are those of
the Columbia University, USA (Subarachnoid Hemorrhage Outcomes Project, SHOP);
University of Washington (Database of Subarachnoid Treatment, D-SAT), USA, and University
of Chicago. The primary outcome measure was commonly the GOS at 2 months in the ISAT, 3
months in 6 studies, and 6 months in the University of Washington dataset. Data on Rankin score
are available in 7 studies. Key variables were standardized across datasets by recoding, and then
extracted along with the identity number provided in the source dataset, and pooled into a single
dataset for further quality checks. Data validation was performed by comparing results of
frequency distributions and cross tabulations between source data, published results on source
data and the pooled data. All data manipulations were done independently by two trained persons
and any inconsistencies resolved by discussions and further consultations with data providers and
experienced clinicians. Data validation and quality checks were deemed satisfactory prior to all
59
analysis. All datasets are archived in a secured St. Michael’s hospital network and backups
performed on a regular basis. The SAHIT repository was the source of patient data for
exploratory analysis of prognostic associations presented in chapter 3, and the development of
novel prognostic scores in SAH presented in chapter 4.
The Discharge Abstract Database (DAD), managed by the Canadian Institute for Health
Information, and the Nationwide Inpatient sample (NIS), managed as part of the Healthcare Cost
and Utilization Project by the Agency for Healthcare Quality and Research (Rockville, MD) are
large administrative databases which contain patient-level sociodemographic, diagnostic,
therapeutic, and administrative information on hospital discharges in Canada and the United
States respectively. Although the DAD mandatorily captures information from all hospital
discharges in Canada, the NIS is designed as a representative 20% subsample of discharges from
acute care hospitals in the United States. DAD data for the fiscal years 2004 to 2010 and NIS
data for the fiscal years 2005 to 2010 were used for this project in order to analyze a cohort most
reflective of current treatment practices in SAH. In both databases, diagnoses were coded
according to the International Classification of Diseases (ICD). Data were abstracted for all
patients with a principal diagnosis of SAH using the diagnostic code appropriate to each database
(ICD, Ninth Revision, Clinical Modification [ICD-9-CM] code 430 for NIS and ICD-10-CM
code I60 for DAD). The accuracy of SAH ICD coding in administrative database has been
validated.210, 211
To minimize the chances of including patients with traumatic SAH, data were
excluded on patients with primary diagnosis of SAH but with a secondary diagnosis related to
head trauma (NIS: ICD-9-CM codes 800.0– 801.9, 803.0–804.9, 850.0–854.1, or 873.0–873.9
and for the DAD: ICD-10 S00–S09). The DAD and NIS data were the sources of patient
60
information for the analyses of the relations between socioeconomic status and race/ethnicity and
SAH outcomes presented in chapter 5.
2.6 Independent (Predictors) variables
Prognostic associations were investigated for 7 risk factors including the demographic
factors age and sex; the clinical factors premorbid history of hypertension and neurologic status
measured on the WFNS scale; and the neuroimaging factors CT clot burden measured on the
Fisher scale, ruptured aneurysm location and diameter. Variables explored as latent prognostic
factors were patient socioeconomic status measured as neighbourhood income status, and
race/ethnic groups were defined as white, black, Hispanic, Asian/ Pacific Islander (API), and
Native Americans/others.
2.7 Dependent (Outcome) Variables
The endpoints for prognostic analyses were the Glasgow outcome score (GOS) at 3
months to explore the prognostic effect of conventional factors and develop novel risk scores
(Chapters 3 and 4); in-hospital mortality and discharge to institutional post-acute care (iPAC) to
explore the effect of socioeconomic status and race/ethnicity (chapter 5). The GOS is an ordered
5 categorical level outcome measure categorized as GOS 1: dead; GOS 2: persistent vegetative
state; GOS 3: severe disability; GOS 4: moderate disability; GOS 5: Good recovery. Rather than
dichotomize the GOS into unfavorable outcome (GOS 1, 2, 3 versus 4, 5) for prognostic analysis
as is traditionally done, it was considered more appropriate to estimate prognostic associations
across the ordinal categories of the GOS to take advantage of the full range of the GOS.197
Three
month GOS was estimated from 3-month Rankin score in the Magnesium for Aneurysmal SAH
(MASH) trials and imputed from 2-month GOS in the International Subarachnoid Aneurysm
61
Trial (ISAT) data and 6 months GOS in the University of Washington D-SAT data. These
approaches have been validated previously.212, 213
2.8 Approach to statistical analyses
2.8.1 Descriptive analysis
Prior to formal analysis, data were examined graphically using bar charts and boxplots
and in frequency tables. The shape of the relationship between continuous variables (age and
aneurysm diameter) and outcome was explored using restricted cubic splines. The splines are
cubic transformations that are constrained to be linear at the tails. Because they are smooth and
flexible functions, restricted cubic splines allow for adequate modeling of nonlinear
relationships.214
Spline function requires the specification of knots around which the curves can
bend. The recommended maximum number of knots is 5, which was applied in the analysis of
continuous variables; though, fitted spline functions with 3 and 4 knots were also investigated in
sensitivity analysis to examine whether the shape of associations is robust to the confounding
effect of changes in number of knots.214
Whether allowing for nonlinearity in the effect of
continuous predictors significantly improved model fit was tested with likelihood ratio tests
comparing nested models with main effects of the continuous predictor fitted with as a linear
term to model extending the predictor effect with a nonlinear term (square or cubic spline).
2.8.2 Univariable prognostic association
Meta-analyses of individual patient data were performed to investigate univariable
prognostic associations for each conventional factor. This was done by fitting proportional odds
logistic models to obtain estimates of prognostic association over the ordered categories of the
GOS for each study and then pooling the summary estimates across studies using a random
62
effects model. The random effects model was considered appropriate as it assumes that studies
included in the analyses were drawn from patient populations that differ from each other in ways
that could impact on prognostic associations; the assumptions of the random effects model is also
consistent with the goal of the analyses which is to derive average estimates of prognostic
associations that are generalizable to a wide range of patient and practice settings. Forest plots
were used to display the common odds ratios from proportional odds models, enabling the
illustration of consistency in prognostic associations from study to study. Between-study
heterogeneity was tested using the I2 statistic and associated probability value. The likelihood of
publication bias was examined with Egger’s test of small study effect.
2.8.3 Multivariable prognostic association
The framework for multivariable analysis is shown in figure 2.1. The adjusted effect of
each predictor was obtained by fitting proportional odds models over the ordered categories of
the GOS sequentially accounting for a consistent set of adjustment factors. Nested sets of
adjustment factors were sequentially included in the analysis in the order in which they are
encountered in the clinical course. Model A accounted for the fixed effect of study. Model B
adjusted for age and neurologic status. Model C adjusted for the additional effect of
neuroimaging factors of Fisher CT clot burden, ruptured aneurysm location and size. Model D
provided further adjustment for the confounding effect of treatment modality (whether aneurysm
was repaired by clipping or coiling, or treated conservatively).
2.8.4 Quantification of the magnitude of prognostic associations
The strength of prognostic associations was quantified as odds ratios and Nagelkerke’s R2
statistic. For categorical variables, odds ratio greater than 1 indicates an increased risk of a poor
63
outcome whereas odd ratios lesser than 1 indicates a decreased risk of a poor outcome relative to
the reference category. The effect size of continuous predictors was scaled to correspond to the
odds ratio associated with a change in the interquartile range of the continuous predictor (the
odds ratios corresponding to changing from the 25th
percentile of the prognostic factor to the 75th
percentile). This standardization allows direct comparison to be made of the prognostic strength
of different prognostic factors irrespective of the unit or scale they were recorded.214, 215
Hence,
an odds ratio greater than one for a continuous predictor variable indicates an increase in the risk
of poor outcome on comparing a patient with the 75th
percentile value of that continuous
predictor variable to a patient with the 25th
percentile value of same predictor variable.
The predictive strength of individual prognostic factors was further assessed as the
difference in Nagelkerke’s R2 of models with and models without the prognostic factor.
Nagelkerke’s R2 is the proportion of the variance of the outcome measure that is explained by
predictor variables included in the model. This so called partial R2
values are considered to
reflect the added incremental effect of each prognostic factor when the prognostic factor is added
to models already containing other adjustment factors.215
They have been used in a similar study
in traumatic brain injury. Because partial R2 is robust to changes in sample size, it provides a
measure for direct comparisons to be made between prognostic factors.215
Significance level was
set at p<0.05, except as specified otherwise in individual chapters or studies.
2.8.5 Secondary analyses: interaction effects and subgroup analysis
Secondary analyses were performed to examine for clinically relevant, pre-specified
interactions and to ascertain how prognostic associations compare at the different
dichotomizations of the GOS. The latter was done by fitting binary logistic regression models to
examine prognostic associations at each dichotomization split of the GOS (GOS1 vs. GOS2-5,
64
GOS 1-2 vs. GOS3-5, GOS 1-3 vs. GOS 4-5, and GOS 1-4 vs. GOS 5). Depending on the
research question, analyses were performed to assess the relation between predictor variables and
a number of secondary outcomes including pre-existing comorbid events, medical and
neurological complications after SAH, and discharge to institutional post-acute care (iPAC).
2.9 Handling of missing data
Given the scope of the research, challenges of missing data were anticipated. This was
handled by the technique of multiple imputations which uses the distributions of observed data to
estimate a set of plausible values for missing data. The pattern of missing data was assumed to be
Missing At Random (MAR), which implies that the reason for the missing values is not
attributable to the variable on which data are missing but is related to other variables available in
the dataset.216
Multiple imputations were performed for missing data using the MICE (multiple
imputations by chained equations) algorithm to sample imputed values from the posterior
predictive distributions of missing data.216
The imputation models were specified on both
independent and dependent variables as well as any other variables that could help explain the
reason for missing values. For each study, 20 imputed datasets were generated for analysis with
the datasets being identical with respect to non-missing data but could vary on imputed values.
2.10 Statistical software
Analyses were performed in Stata version 12.1 (Stata corporation Texas, USA) and R (R
Foundation for Statistical computation, Vienna, Austria). The R packages used were the rms,
MICE, Foreign and gplot packages.
65
Table 2.1 – Characteristics of studies in the SAHIT repository
MASH-I
& II
IMASH TIRILAZ
AD
C- 1 ISAT IHAST MAPS HHU SHOP D-SAT Chicago Leeds Durham
Type P-II & P-
III
P-III P-III P-II P-III P-III P-IV P-II Hospital
registry
Hospital
registry
Hospital
registry
Observat
ional
Observation
al study
Purpose Mg. to
reduce
DCI &
ASA to
reduce
DIND
Mg. to
improve
clinical
outcome
Tirilazad
to reduce
poor
outcome
Safety
of
clazose
ntan
Coiling vs.
clipping
Intraop.
cooling to
improve
outcome
Establish
TAR rates
for Matrix
2 & GDC
coils
effect of
head-
motion &
fibrinolysis
Prospectiv
e hospital
dataset
Prospectiv
e hospital
dataset
Prospectiv
e hospital
dataset
Association
between
PTSD &
QoL
Centers Multiple:
Netherland
, Scotland
& Chile
Multiple:
Mostly
Asia
Multiple:
NA, Asia,
Europe
Multipl
e: NA,
Europe
Multiple:
mostly
Europe
Multiple:
NA, Europe
& Asia
Muliple:
NA,
Europe &
Asia
Single
center:
Germany
Columbia
U, USA
U
Washingto
n, USA
U Chicago,
USA
U Leeds,
UK
New castle
& Durham,
UK
Enrolment 2000 -
2011
2002 -
2008
1991 -
1997
2005 -
2006
1997 -
2002
2000 - 2003 2007- 2011 2008-2011 1996- 2012 1983-1993 1995-2002 2005-2006
Time to
treatment
≤ 4 days ≤48 hrs ≤ 48 hrs ≤ 56
hrs
≤ 28 days < 14 days <24 hrs ≤72 hrs ≤48 hrs ≤48 hrs ≤48 hrs
Outcomes DCI & RS
GOSE GOS GOSE RS GOSE TAR & RS GOS, clot
clearance
rate, DCI
GOS, RS,
Barthel,
NIHSS
GOS, RS,
NIHSS,
Barthel
RS RS,
Neurcog
nitve
battery
Neurocogniti
ve battery
Follow up 3 mos 3 mos 3 mos 3 mos 2 & 12mos 3 mos 12mos 3 mos. 3 mos 6 mos Variably 3 & 13mos
Size 1484 327 3552 413 2143 1001 228 60 1500 439 75 117 105
Result Nil effect Nil effect Nil effect Nil
effect
Coiling is
better
Nil effect Yet to be
published
Nil effect - - - - Positive
SAHIT
data
elements
25 34 110 >1000 39 400 14 30 34 182 28 24 322
Notes: Size; sample size; MASH: Magnesium Sulfate in Aneurysmal Subarachnoid Hemorrhage : A Randomized controlled trial; IMASH: Intravenous Magnesium Sulphate for
Aneurysmal Subarachnoid Hemorrhage; C-1: CONSCIOUS-1 Trial; ISAT: International Subarachnoid Aneurysm Trial; IHAST: Intraoperative Hypothermia for Aneurysm
Surgery Trial; MAPS: Matrix And Platinum Science Trial; HHU: Heinrich-Heine University Concomittant Intraventricular Fibrinolysis and Low-Frequency Rotation After Severe
Subarachnoid Hemorrhage Trial; SHOP: Subarachnoid Hemorrhage Outcomes project; D-SAT : Database of Subarachnoid Treatment, University of Washington; Chicago:
University of Chicago SAH registry; Leeds: University of Leeds Neurocognitve observation study; Durham: Observational neurocognitive study from University of Durham; Mg:
Magnesium; ASA: Aspirin; PTSD: Post traumatic stress disorder; QoL: Quality of life; NA: North America; DCI: Delayed cerebral ischemia; RS: Rankin score; GOS/GOSE:
Glasgow outcome score and its extended variant; TAR: Target aneurysm recurrence; NIHSS: National Institute of Health Stroke scale. P: Phase
66
Figure 2.1 – Framework of multivariable analysis in SAHIT repository
Model A
•Predictor & Study fixed effect
Model B
•Model A + Age & WFNS grade
Model C
•Model B + Fisher, Location & Size
Model D (full model)
• Model C + Repair modality
67
Chapter 3
Prognostic value of hospital admission characteristics in aneurysmal
subarachnoid hemorrhage: Meta-analyses of individual participant data in
the subarachnoid hemorrhage international trialists (SAHIT) repository
This chapter is adapted from 3 manuscripts presently under consideration for publication as
follows:
Jaja BN, Lingsma H, Thorpe KE, Schweizer TA, Steyerberg EW, Macdonald RL, for the SAHIT
Collaboration. Prognostic effect of age and sex in aneurysmal subarachnoid hemorrhage: meta-
analyses in 10 951 patients in an international database. Journal of Neurosurgery
Jaja BN, Lingsma H, Thorpe KE, Schweizer TA, Steyerberg EW, Macdonald RL, for the SAHIT
Collaboration. Prognostic value of premorbid hypertension and baseline neurologic status in
aneurysmal subarachnoid hemorrhage: pooled analyses of individual patient data in the SAHIT
repository. Journal of Neurosurgery. Provisionally accepted
Jaja BN, Lingsma H, Thorpe KE, Schweizer TA, Steyerberg EW, Macdonald RL, for the SAHIT
Investigators. Admission neuroimaging characteristics as predictors of outcome after aneurysmal
subarachnoid hemorrhage: pooled analyses of individual patient data in the SAHIT repository.
Journal of Neurosurgery
68
3.1 Introduction
This chapter of the thesis reports results of detailed analyses of prognostic associations in
the SAHIT repository with regard to 7 risk factors obtained at hospital admission. The aim of the
investigation is to provide more definitive and accurate estimation of their value as prognostic
factors in aneurysmal SAH; given the size and heterogeneity of the sample population. The study
confirmed some variables and refuted others with regard to some conventional factors, but
provided more precise estimation of their prognostic strength and the shape of the relationship of
the variable to outcome. The study also contributed to settling unresolved issues in the literature
with regard to some prognostic factors.
3.2 PART A: Demographic factors - age and sex
Patient’s age and sex are important demographic factors influencing the etiopathogenesis
and outcomes of aneurysmal subarachnoid hemorrhage (SAH). Advancing age has been shown
to correlate with poorer outcomes after SAH, 78-83
but the prognostic effect of age has yet to be
adequately quantified. Issues remain as to the upper age limit beyond which a considerable
increase in prognosis may be expected with advancing age; an uncertainty reflected in the
literature as the use of different cut points to describe the optimal change point in the effect of
age,86-92
making comparison across studies challenging. The sex distribution of SAH is skewed
towards a higher incidence and rupture rates of intracranial aneurysms in women.17, 100
However,
whether sex differences are present in SAH outcomes could be debated. While some studies have
reported sex related differences in mortality rates and greater risk factors burden in women, 16, 217
other studies suggest the overall outcome of SAH is unrelated to patient’s sex.82, 83, 218, 219
While
69
survival after SAH has improved recently, recent evidence indicates this may not have been
evenly distributed over age and sex.101
Part A of this chapter aimed to (1) investigate the change point in the prognostic effect of
age, if any; (2) more accurately estimate the prognostic strength of age and sex for 3-month
outcomes on the Glasgow outcome scale (GOS); and (3) examine whether the effects of age and
sex differ between trial and non-trial patients.
3.2.1 Methods
Patient-level meta-analyses were conducted on 8 randomized clinical trials and 2
observational studies involving 10951 patients to investigate univariable association between
age, sex and 3-month GOS. The adjusted effects of age and sex were estimated by fitting
proportional odds models to estimate prognostic associations after accounting for the effect of
other prognostic factors. Nested sets of adjustment factors were sequentially included in the
analyses in the order in which they are encountered in the clinical course. Model A accounted for
the fixed effect of study. Model B adjusted for age (in the analysis of sex effect) and neurologic
status. Model C adjusted for the additional effect of neuroimaging factors of Fisher CT clot
burden, ruptured aneurysm location and size. Model D (full model) provided further adjustment
for the confounding effect of treatment modality (whether aneurysm repair was performed by
clipping, or coiling, or treated conservatively).
Restricted cubic spline function was used to study non-linearity in the effect of age; the
effect of age was scaled as the odds ratio over the difference between the 75th and 25th
percentiles. The percentage of total data points imputed was 11.8%. To test whether the
prognostic effect of age or sex differed between trial and non-trial patients, an interaction term
70
was included in the fully adjusted models between age or sex and studies stratified by trial status
with significance level set at a probability value of 0.01 to account for multiple testing and the
effect of large sample size.
3.2.2 Results
The distribution of age across studies is shown in Figure 3.1. Median age was 53 years
with IQR of 44-62 years. Proportion of patients who were women was 71%. Spline plots of the
shape of the relationship between age and outcome demonstrated the probability of poor
outcome increases with advancing age, with a steep increase in poor outcome around the age of
60-65 years; suggesting a change point around this age group though the actual age value
differed somewhat with dichotomization split point of GOS (Figure 3.2). There was only a
marginal difference in model χ2
on comparing the full model (model D) fitted with nonlinear
transformation of age, using restricted cubic splines, with the full model with age fitted as a
linear term (χ2: 2888 versus 2833 respectively); suggesting the effect of age was adequately
modelled with a linear term. The negative effect of increasing age on outcome was consistently
demonstrated across studies (Figure 3.3). The unadjusted effect of age was estimated in the meta-
analysis as Odds ratio (OR) = 1.78 (95% CI, 1.61 – 1.98) for 62 versus 44 years old (75th
vs. 25th
percentile). The effect of age was only slightly reduced on sequentially accounting for the effect
of other prognostic factors (Table 3.1). The fully adjusted model (Model D) estimated a 69%
increase in the odds of poor outcome when patients of 62 years of age were compared with
patients of 44 years of age (OR, 1.69; 95% CI: 1.59-1.81).
Meta-analyses (Figure 3.3) demonstrated that, overall, prognosis was relatively poorer in
women compared with men (OR 1.20; 95% CI: 1.04 – 1.39). This association was not significant
on full adjustment for other prognostic factors (Model D: OR, 1.04 95% CI: 0.84-1.29). No
71
evidence was found in support of a small study effect (publication bias) for age (p = 0.552) nor
for sex (p= 0.373) in the meta-analyses.
Figure 3.4 comparing the relative strength of studied prognostic factors quantified as
partial R2 values shows that age had added incremental predictive value over and above the
combined prognostic effect of other prognostic factors adjusted for in the analysis. Age
independently explained 3% of the total variability in GOS outcome. However, patient’s sex had
negligible incremental predictive value above that due to other prognostic factors.
Examination of interaction effects after full adjustment showed that the effect of age did
not differ between patients who are enrolled into clinical trials and hospital cohorts (p=0.70); did
not vary between men and women (p=0.79); but differed with patient’s neurologic status
(p=0.006). Secondary analyses showed that the magnitude of prognostic associations were
similar at each dichotomization point of the GOS and across the full range of the GOS.
3.2.3 Discussion
Whether an age limit exists beyond which a considerable increase in risk in poor outcome
may be expected is uncertain. Though some studies have suggested a linear association between
age and outcomes of SAH78, 91
, it is not uncommon to find age dichotomized at different cut
point values to investigate prognostic associations, including the use of upper limit values of 50
years in older studies of over 2 decades ago86
, to 60 years65, 78, 87
, 65 years88, 95
, 70 years85, 90, 96, 97
or even 75 years91, 92
to define the elderly in whom prognosis is expected to worsen considerably.
The only prior study78
, to our awareness, that systematically investigated a change point in the
prognostic effect of age analysed 906 US and Canadian patients who were enrolled into the
international cooperative study of SAH between 1989 and 1991. The study reported that a break
72
point value of 60 years best discriminated between patients at low and high risk of 3-month
favorable outcome on the dichotomized GOS, suggesting this age value be regarded as the cut
point beyond which outcomes of SAH worsen markedly with increasing age. The present study
demonstrated that prognosis worsens around the age of 60-65 years after SAH, though no clear
cut-off value could be recommended given our finding that such a threshold value would differ
with outcome (GOS split point). Moreover, we found the relation could be adequately modeled
as a linear relationship.
This study confirmed established knowledge that age is an independent prognostic factor
in SAH78, 80-83, 143
, but the study further indicated that the overall effect of age is rather moderate.
SAH occurs more commonly in women than men and patient’s sex has been shown to interact
with many risk factors for aneurysm formation, growth and rupture16, 101
. In consonance with
most prior studies82, 83, 218, 219
the present study found a negligible effect of patient’s sex on
clinical outcome. Because patients enrolled into clinical trials are selected with respect to the
overall SAH cohort, prognostic associations may be expected to differ in these patients compared
with non-selected hospital cohorts. No evidence was found in support of this assumption with
respect to the effect of age and sex.
In summary, part A of this chapter demonstrated that prognosis worsens around age 60-
65 years, though presuming the effect of age to be continuous and linear was adequate. Age had
an overall independent but moderate effect on outcome. The prognostic effect of sex was
negligible.
73
Figure 3.1 – Boxplot of age by study cohort
C-1: CONSCIOUS 1 trial (N=433); IHAST: Intraoperative Hypothermia for Aneurysm Surgery Trial (N=998);
IMASH: Intravenous Magnesium Sulphate for Aneurysmal Subarachnoid Hemorrhage (N=327); ISAT:
International Subarachnoid Aneurysm Trial (N=2143); MASH: Magnesium Sulfate in Aneurysmal Subarachnoid
Hemorrhage (N=315 for MASH-1 and 1204 for MASH-2); HHU: Heinrich-Heine University Concomitant
Intraventricular Fibrinolysis and Low-Frequency Rotation After Severe Subarachnoid Hemorrhage Trial (N=60);
D-SAT: Database of Subarachnoid Treatment of the University of Washington (N=439); SHOP: Subarachnoid
Hemorrhage Outcomes project of Columbia University(N=1500).
0 20 40 60 80 100
Age in years
TIRILAZAD
SHOP
MASH-2
MASH-1
ISAT
IMASH
IHAST
HHU
D-SAT
C-1
74
Figure 3.2 – Spline plot of the relation of age to outcome at different dichotomization split points
of the GOS
1= Probability of less than good outcome (GOS 1 versus GOS2-5)
2=Probability of unfavorable outcome (GOS 1-2 versus GOS 3-5)
3= Probability of death/vegetative outcome (GOS1-3 versus GOS4-5)
4= Probability of death (GOS1-4 versus GOS5)
75
Figure 3.3 – Forest plot demonstrating consistency in the effects of age and sex
NOTE: Weights are from random effects analysis
Overall (I-squared = 63.9%, p = 0.003)
HHU
MASH-2
MASH-1
ISAT
TIRILAZAD
IMASH
Study
IHAST
D-SAT
SHOP
C-1
1.78 (1.61, 1.98)
1.84 (0.78, 4.36)
2.12 (1.81, 2.48)
1.60 (1.19, 2.15)
1.43 (1.25, 1.64)
1.93 (1.75, 2.13)
1.99 (1.49, 2.68)
ratio (95% CI)
Odds
1.75 (1.44, 2.13)
1.99 (1.58, 2.52)
1.93 (1.69, 2.22)
1.25 (0.91, 1.71)
1.78 (1.61, 1.98)
1.84 (0.78, 4.36)
2.12 (1.81, 2.48)
1.60 (1.19, 2.15)
1.43 (1.25, 1.64)
1.93 (1.75, 2.13)
1.99 (1.49, 2.68)
ratio (95% CI)
Odds
1.75 (1.44, 2.13)
1.99 (1.58, 2.52)
1.93 (1.69, 2.22)
1.25 (0.91, 1.71)
Better Poorer 1.229 1 4.36
Age
NOTE: Weights are from random effects analysis
Overall (I-squared = 60.8%, p = 0.006)
MASH-2
Study
ISAT
SHOP
IMASH
HHU
MASH-1
D-SAT
TIRILAZAD
IHAST
C-1
1.20 (1.04, 1.39)
1.64 (1.30, 2.07)
ratio (95% CI)
1.45 (1.22, 1.73)
1.23 (0.98, 1.55)
1.16 (0.77, 1.74)
0.52 (0.19, 1.44)
1.03 (0.67, 1.59)
1.01 (0.71, 1.44)
1.06 (0.90, 1.26)
0.93 (0.72, 1.21)
1.58 (1.06, 2.36)
Odds
1.20 (1.04, 1.39)
1.64 (1.30, 2.07)
ratio (95% CI)
1.45 (1.22, 1.73)
1.23 (0.98, 1.55)
1.16 (0.77, 1.74)
0.52 (0.19, 1.44)
1.03 (0.67, 1.59)
1.01 (0.71, 1.44)
1.06 (0.90, 1.26)
0.93 (0.72, 1.21)
1.58 (1.06, 2.36)
Odds
Better Poorer 1.188 1 5.31
Sex
76
Table 3.1 – Results of adjusted analysis of the prognostic effect of age and sex
Model A: Predictor + Study
Model B: Model A + WFNS+Age
Model C: Model B + Neuroimaging data (Fisher grade+ Aneurysm location + Ruptured aneurysm size)
Model D: Model C + Repair (clipping vs. coiling vs. conservative)
Figure 3.4 – Relative prognostic value of studied prognostic factors expressed as Nagelkerke’s
partial R2
Bars represent differences in R2 values of adjustment models with and without predictor of interest.
0
5
10
15
20
25
% o
f e
xpla
ine
d v
aria
nce
in G
OS
Prognostic factors
Model A
Model B
Model C
Model D
Model A Model B Model C Model D
Age 1.86 (1.75-1.94) 1.80 (1.70-1.90) 1.71 (1.60-1.82) 1.69 (1.59-1.81)
Female 1.21 (1.05-1.40) 1.10 (0.91-1.33) 1.03 (0.84-1.28) 1.04 (0.84-1.29)
77
3.3 PART B: Clinical factors - premorbid hypertension and admission
neurologic status
It is now recognised that aneurysmal SAH is the end result of a chronic disorder of
cerebral arteries220, 221
and that the formation, growth and eventual rupture of aneurysms is
facilitated by pre-existing hypertension.17, 222
Hypertension has been identified as the commonest
comorbid disease seen in patients with SAH with reported prevalence estimates exceeding 40%
in some studies.93, 136-138
Despite a growing body of literature examining the relevance of
hypertension to clinical outcomes of SAH, no clear evidence exists as to whether a premorbid
history of hypertension is independently associated with outcomes of SAH.82, 137-143
Patient neurologic status is the single most important indicator of the severity of brain injury
soon after SAH and is critical to treatment decisions and prediction of outcome.60, 123, 124
However, accurate estimation of the prognostic strength of neurologic status has been difficult as
investigators have usually investigated prognostic associations using only cohorts of clinical
trials82, 143
or cohorts representing experience from a single hospital, 94, 142
with risk adjustment
for a variable set of confounders.
The primary objective of Part B of this chapter was to investigate the role of premorbid
history of hypertension and patient neurologic status as prognostic factors for 3-month outcome
on the Glasgow outcome scale.
3.3.1 Methods
Patient-level meta-analyses were conducted to investigate univariable association
between premorbid hypertension (in 6 studies; N=7,249), admission neurologic status measured
on the WFNS scale (in 10 studies; N=10,869) and the primary endpoint of 3- month GOS.
Multivariable analyses were performed to sequentially adjust the effect of premorbid
78
hypertension and neurologic status for age, CT clot burden, aneurysm location, size, and
treatment modality. Prognostic associations were estimated over the ordered categories of GOS
using proportional odds models. Secondary analyses were conducted examining the relation
between premorbid hypertension to cardiovascular (Myocardial infarction, Atrial fibrillation and
congestive heart failure) and renal comorbidities, medical (hyperglycemia, renal failure, fever
and anemia, pulmonary edema) and neurological (cerebral infarction, hydrocephalus,
intraventricular hemorrhage, rebleeding and delayed ischemic neurological deficits)
complications after SAH. This was done to provide insight into putative pathways by which
premorbid hypertension influences the outcome of patients with SAH, if any. The study imputed
3% of total data required for analysis of the effect of premorbid hypertension and 8.8% of total
data required for analysis of the effect of neurologic status.
3.3.2 Results
The average age of patients was 52.5 ± 13.4 years. The overall proportion of patients with
a premorbid history of hypertension was 37.5% (Range across studies: 31-48%). The distribution
of neurologic status was U-shaped across studies, except for the Intraoperative Hypothermia for
Aneurysm Surgery (IHAST, which excluded poor grade patients a priori); the ISAT; and data
from Heinrich-Heine University Concomitant Intraventricular Fibrinolysis and Low-Frequency
Rotation after Severe Subarachnoid Hemorrhage Trial, which partially or completely excluded
good grade patients a priori (Figure 3.5).
Patients with premorbid history of hypertension were older, and more likely to present
with poorer neurologic status than patients without premorbid hypertension (p≤ 0.001). They
experienced progressively worse crude outcome at 3-month (Table 3.2). A premorbid history of
hypertension was associated with poorer outcomes across studies; the unadjusted pooled odds
79
ratio was OR, 1.73 (95% confidence intervals [CI]: 1.50 - 2.00). There was no evidence of
between study heterogeneity in the estimates of the effect of premorbid hypertension (I2 test of
heterogeneity: 44.5%, p=0.108). Adjusting the effect of premorbid hypertension for age and
neurologic status resulted to a moderate decrease in the magnitude of the effect of premorbid
hypertension (OR, 1.37 95% CI: 1.25 - 1.53; Model B), suggesting that age and neurologic status
mediates part of the effect of premorbid hypertension on outcome. Further adjusting for CT clot
burden, aneurysm size and location (Model C) and modality of treatment (Model D) had no
further effect on the strength of the relation of hypertension with outcome (Table 3.3). In
secondary analysis, patients with premorbid history of hypertension had a significantly higher
prevalence of pre-existing cardiovascular events and renal disease, and higher rates of medical
and neurological complications than patients without a premorbid history of hypertension (Table
3.4). In adjusted analyses, premorbid hypertension was independently associated with a history
of myocardial infarction, a history of kidney disease; and higher odds of renal failure and fever,
but not with higher odds of hyperglycemia, anemia and pulmonary edema. Premorbid
hypertension was also independently associated with higher odds of neurological complications
including cerebral infarction, hydrocephalus, intraventricular hemorrhage, rebleeding, and
delayed ischemic neurologic deficits (Table 3.4). Including rebleeding in the full adjustment
model examining the relation of hypertension to GOS outcome (Model D) resulted into a slight
reduction in the odds ratio associated with the effect of premorbid hypertension from OR, 1.38 to
1.32; indicating rebleeding had further explanatory effect on the relation between premorbid
hypertension and outcome. Similar adjustment for history of myocardial infarction, history of
kidney disease, renal failure, fever, cerebral infarction, hydrocephalus, intraventricular
80
hemorrhage, and delayed ischemic neurologic deficits during admission had no effect on the
prognostic strength of premorbid hypertension.
Meta-analysis (Figure 3.7) demonstrated each increase in neurologic status on the WFNS
grade resulted to approximately doubling of the risk of poor outcome at 3 months. Between-
study heterogeneity in the estimate of the effect of neurologic status on the WFNS scale was not
significant for all grades of WFNS (grade II: I2, 17.3%, p=0.29; grade III: I
2, 44.8%, p=0.05; and
grade IV: I2:11.0%, p=0.35); except for WFNS grade V (poorest grade) patients (I2=78%; p<
0.001). Sequentially adjusting the effect of neurologic status for age, neuroimaging covariates
and modality of treatment had only slight effect on the magnitude of the effect of neurologic
status. In the full adjustment model (Model D), the Odds ratios associated with the effect of
neurologic status was WFNS II (OR, 1.85; 95% CI: 1.68-2.03); WFNS III (OR, 3.85; 95% CI:
3.32-4.47); WFNS IV (OR, 5.58; 95% CI: 4.91-6.35); WFNS V (OR, 95% CI: 14.18; 12.20-
16.49). When prognostic strength of premorbid hypertension and neurologic status was evaluated
in terms of added predictive value (Figure 3.4), premorbid hypertension added only marginal
predictive value to adjustment models already containing prognostic factors. The partial R2 in
adjustment models was less than 0.5%. In contrast, admission neurologic status had added
incremental predictive value beyond those of other covariates in the adjustment models. The
unadjusted R2 was 22.65%. The partial R
2 in adjustment models was 14.04% (Model B), 14.18%
(Model C), and 12.98% (Model D) respectively.
3.3.3 Discussion
This study addressed the inconsistencies in the literature regarding the role of premorbid
history of hypertension as prognostic factor in SAH. Consistent evidence was found in support of
a significant association between premorbid history of hypertension and poor outcome. Though
81
the association was weakened after adjusting for the effect of other prognostic factors, the effect
of history of hypertension remained independent of these factors; suggesting that a premorbid
history of hypertension is an independent prognostic factor for 3-month outcome after
aneurysmal SAH. The study however demonstrated that premorbid hypertension added marginal
prognostic information to models already containing other prognostic factors, suggesting that
premorbid hypertension is a weak prognostic factor. Many studies have been published in the
literature to support82, 137-139, 141, 144
or refute93, 94, 142-145
an association between premorbid
hypertension and outcome after SAH. The design of the present study enabled us to address a
number of reasons that could have potentially contributed to the conflicting results of prior
studies, including reasons such as variability in hypertension prevalence across studies,
differences in case mix, inadequate statistical power, lack of adjustment for important
confounders or differences in outcome measures due to dichotomization of the GOS to evaluate
prognostic effect of hypertension for mortality or risk of unfavorable outcome.
Of interest is the likely mechanism by which premorbid hypertension could increase the
risk of poor outcome after SAH. The results of previous studies and our analysis indicate a
multifactorial pathway. Juvela138
alluded to the effect of chronic hypertension on arteriolar
smooth muscle cells causing hypertrophy and premorbid narrowing of cerebral arteries, which
could predispose to a higher risk of ischemic injuries after SAH. Other researchers have shown
that cerebral infarction after SAH increases the likelihood of unfavorable outcome fivefold and is
significantly predicted by a history of hypertension. 50, 223
One study found higher risk of
atherosclerosis independently predicted poor outcome in SAH patients, which was unrelated to
the occurrence of delayed cerebral ischemia, but, was in part, related to a marked decrease in
rebleeding in SAH patients who did not have or had minor degrees of atherosclerosis.224
A recent
82
study found more severe initial bleeding, higher risk of rebleeding and higher risk of inhospital
mortality in patients with premorbid hypertension relative to those without hypertension.225
The
present analysis corroborated the findings of previous studies. It demonstrated premorbid
hypertension to be associated with more severe initial bleeding, cardiovascular and renal
comorbidities, and higher risk of medical and neurological complications. In particular, adjusting
for the effects of age, neurologic status and rebleeding decreased the magnitude of the effect of
hypertension; suggesting that these factors directly mediate, in part, the effect of premorbid
hypertension on SAH outcome. Nonetheless, the effect of hypertension was still independent of
their intermediary effects, suggesting other factors hitherto unknown may be involved also.
Though premorbid hypertension was shown to be a weak prognostic factor, the independent
relation of hypertension to neurological complications in the present study suggests the need to
consider premorbid history of hypertension as putative confounder in studies evaluating the
effect of new therapies using neurologic complications such as DIND or rebleeding as surrogate
outcome measures.
Admission neurologic status reflects the severity of brain injury at time of rupture. It is
considered the single most important predictor of outcome in SAH patients.4, 25
The present
analysis is consistent with this established finding. The study further demonstrated that
neurologic status added predictive value above the combined value of other prognostic factors.
The meta-analysis suggests that each increase in neurologic grade resulted in an almost doubling
of the risk of poor outcome at 3 months. The confidence intervals around the effect estimates in
the present study, which are narrower than those of previous studies, indicate a more precise
estimation of the prognostic strength of neurologic status as measured on the WFNS scale. The
83
high between study heterogeneity seen among WFNS grade V (poorest grade) patients could be
indicative of the preferential exclusion of grade V patients from clinical trials.
The study has some limitations. How premorbid hypertension was precisely defined in all
studies could not be ascertained, though the incidence is comparable across studies. The present
study shares with previous studies the limitation of not investigating whether the prognostic
effect of premorbid hypertension is related to the duration or severity of hypertension, or the
adequacy or otherwise of blood pressure control. The limitations of using the WFNS scale for
grading neurologic status has been reviewed elsewhere.123
Though some other scales have shown
relatively better interobserver agreement and a more graduated relationship to outcome than the
WFNS scale, they were not better than the latter in the capacity to differentiate patients by
outcome and are less popular than the WFNS scale.131-134
Some researchers have also argued
about the optimal time point for assessing neurologic status for purposes of prognostication, with
different time points proposed in the literature including clinical assessment soon after injury125
,
and after neurologic resuscitation.126
In conclusion, this study has provided a more definitive and accurate estimation of the
value of patient neurologic status measured on the WFNS scale and premorbid history of
hypertension as prognostic factors in SAH than any prior study. It demonstrated premorbid
hypertension to be an independent but weak prognostic factor in SAH. The effect of premorbid
hypertension on outcome could involve multifactorial mechanisms including an increase in the
severity of initial bleeding, the rate of comorbid events and neurological complications, among
other factors.
84
Table 3.2 – Distribution of premorbid hypertension and neurologic status by 3-month GOS
Good Moderate Severe Vegetative Dead Total
Hypertension 1037(30.4) 514(38.3) 363(44.5) 58(45.7) 508(49.6) 2480(36.9)
WFNS I 3100(61.6) 1089(46.9) 394(29.9) 73(23.0) 201(15.7) 4862(47.2)
II 1293(25.7) 645(27.8) 353(26.8) 77(24.3) 239(18.2) 2607(25.3)
III 256(5.1) 157(6.8) 157(12.0) 38(12.0) 131(10.0) 739(7.2)
IV 248(4.9) 283(12.2) 234(17.7) 67(21.1) 293(22.3) 1125(10.9)
V 134(2.7) 147(6.3) 179(13.6) 62(19.6) 445(33.9) 967(9.4)
Table 3.3 – Adjusted effects of premorbid hypertension and neurologic status
Model A Model B Model C Model D
Hypertension 1.82 (1.66 - 1.99) 1.37 (1.24 - 1.52) 1.37 (1.24 - 1.52) 1.38 (1.25 - 1.53)
WFNS I 1 1 1 1
II 2.02 (1.84 - 2.21) 1.95 (1.78 - 2.14) 1.82 (1.65 - 2.00) 1.85 (1.68 - 2.03)
III 4.65 (4.03 - 5.38) 4.19 (3.62 - 4.84) 3.86 (3.3 3 -4.47) 3.85 (3.32 - 4.47)
IV 6.62 (5.84 - 7.50) 6.12 (5.40 - 6.93) 5.56 (4.89 - 6.32) 5.58 (4.91 - 6.35)
V 17.94 (15.5 - 20.7) 18.09 (15.7 - 20.9) 15.39 (13.3 - 17.9) 14.18 (12.2 - 16.5) Model A: Predictor (hypertension or WFNS) + Study
Model B: Model A + WFNS+Age (age only in the analysis of the effect of neurologic status)
Model C: Model B + Neuroimaging data (Fisher grade+ Artery + Ruptured aneurysm size)
Model D: Model C + Repair (clipping vs. coiling vs. conservative)
85
Table 3.4 – Relation of premorbid hypertension to comorbid conditions and complications
N (Event) Premorbid hypertension Adjusted Odds ratio
(95% Confidence
intervals)
No (%) Yes (%)
Comorbidities *
Myocardial infarction 6310 (145) 38 (1.0) 107 (4.5) 3.26 (2.21 – 4.81)
Atrial fibrillation 2421 (63) 21 (1.6) 42 (3.9) 1.50 (0.85 – 2.63)
CHF 2417 (28) 2 (0.2) 26 (2.4) 8.07 (1.86 – 34.91)
Kidney 1422 (34) 8 (1.1) 26 (3.8) 3.84 (1.59 – 9.24)
Medical complications†
Hyperglycemia 1442 (730) 334 (44.9) 396 (56.7) 1.21 (0.95 – 1.54)
Renal failure 2440 (46) 9 (0.7) 37 (3.4) 4.57 (2.11 – 9.90)
Fever 5584 (2133) 1233 (35.6) 900 (42.4) 1.20 (1.05 – 1.36)
Anemia 1854 (518) 271 (26.3) 247 (30.0) 0.90 (0.71 – 1.13)
Pulmonary edema 4919 (553) 289 (9.4) 264 (14.3) 1.15 (0.95 – 1.40)
Neurological complications†
Cerebral infarction 7100 (1588) 928 (20.9) 660 (24.8) 1.17 (1.03 – 1.33)
Hydrocephalus 6758 (2718) 1557 (36.8) 1161 (46.0) 1.19 (1.07 – 1.33)
Intraventricular hemorrhage 5809 (2729) 1566 (42.7) 1163 (54.2) 1.24 (1.10 – 1.40)
Rebleeding 2089 (194) 88 (7.5) 106 (11.6) 1.46 (1.05 – 2.03)
DIND 7275 (1626) 979 (21.4) 647 (24.0) 1.31 (1.16 – 1.49) N (events) denotes the total number of patients with known status on the variable; in parenthesis is the number of
events analysed. Example of how odds ratio should be interpreted: “Patients with history of hypertension had 3.26
times higher odds of a history of myocardial infarction compared with patients without history of hypertension, on
adjusting for age and the fixed effect of study”
* adjusted for age and fixed effect of study; †adjusted for age and WFNS grade and fixed effect of study
86
Figure 3.5 – Percentage distribution of neurologic status in included studies
0
10
20
30
40
50
60
70
I: Good grade
II
III
IV
V: Poor grade
87
Figure 3.6 – Forest plot of the effect of premorbid hypertension across studies
NOTE: Weights are from random effects analysis
Overall (I-squared = 44.5%, p = 0.108)
SHOP
TIRILAZAD
D-SAT
Study
C-1
IHAST
IMASH
1.73 (1.50, 2.00)
2.18 (1.75, 2.71)
1.86 (1.63, 2.13)
Odds
1.40 (0.99, 1.99)
ratio (95% CI)
1.24 (0.83, 1.84)
1.73 (1.34, 2.23)
1.58 (1.06, 2.35)
1.73 (1.50, 2.00)
2.18 (1.75, 2.71)
1.86 (1.63, 2.13)
Odds
1.40 (0.99, 1.99)
ratio (95% CI)
1.24 (0.83, 1.84)
1.73 (1.34, 2.23)
1.58 (1.06, 2.35)
Better Poorer 1.369 1 2.71
88
Figure 3.7 – Forest plot demonstrating consistency in the effect of neurologic status across
studies
WFNS grade I was the reference category against which other categories were compared
NOTE: Weights are from random effects analysis
OverallSHOP
ISATIMASH
TIRILAZADD-SAT
Study
MASH-1
C-1
MASH-2
IHAST
2.16 (1.93, 2.42)3.25 (2.29, 4.63)
2.00 (1.65, 2.44)1.73 (0.98, 3.04)
1.92 (1.60, 2.30)2.35 (1.41, 3.91)
ratio (95% CI)
2.02 (1.18, 3.43)
1.89 (1.22, 2.93)
2.48 (1.89, 3.25)
2.28 (1.73, 3.00)
Odds
2.16 (1.93, 2.42)3.25 (2.29, 4.63)
2.00 (1.65, 2.44)1.73 (0.98, 3.04)
1.92 (1.60, 2.30)2.35 (1.41, 3.91)
ratio (95% CI)
2.02 (1.18, 3.43)
1.89 (1.22, 2.93)
2.48 (1.89, 3.25)
2.28 (1.73, 3.00)
Odds
Better Poorer 1.216 1 4.63
WFNS II
NOTE: Weights are from random effects analysis
OverallSHOPD-SAT
IMASH
TIRILAZAD
C-1
MASH-1
Study
IHAST
MASH-2
ISAT
4.21 (3.26, 5.45)2.76 (1.38, 5.52)2.30 (0.90, 5.83)
2.97 (1.42, 6.20)
5.27 (4.23, 6.57)
1.57 (0.53, 4.67)
8.67 (3.18, 23.60)
ratio (95% CI)Odds
5.28 (3.04, 9.19)
6.28 (3.81, 10.35)
3.76 (2.64, 5.34)
4.21 (3.26, 5.45)2.76 (1.38, 5.52)2.30 (0.90, 5.83)
2.97 (1.42, 6.20)
5.27 (4.23, 6.57)
1.57 (0.53, 4.67)
8.67 (3.18, 23.60)
ratio (95% CI)Odds
5.28 (3.04, 9.19)
6.28 (3.81, 10.35)
3.76 (2.64, 5.34)
Better Poorer 1.0424 1 23.6
WFNS III
NOTE: Weights are from random effects analysis
OverallSHOP
MASH-1
IMASHC-1
Study
ISAT
D-SAT
MASH-2TIRILAZAD
7.03 (6.07, 8.15)8.98 (6.53, 12.37)
4.03 (2.15, 7.56)
7.01 (3.91, 12.55)5.65 (3.49, 9.15)
ratio (95% CI)
6.55 (3.88, 11.06)
5.31 (3.19, 8.83)
7.27 (5.22, 10.13)7.67 (6.09, 9.65)
Odds
7.03 (6.07, 8.15)8.98 (6.53, 12.37)
4.03 (2.15, 7.56)
7.01 (3.91, 12.55)5.65 (3.49, 9.15)
ratio (95% CI)
6.55 (3.88, 11.06)
5.31 (3.19, 8.83)
7.27 (5.22, 10.13)7.67 (6.09, 9.65)
Odds
Better Poorer 1.0797 1 12.5
WFNS IV
NOTE: Weights are from random effects analysis
Overall
D-SAT
Study
MASH-2
ISAT
MASH-1
SHOP
TIRILAZAD
IMASH
C-1
13.60 (9.08, 20.36)
19.01 (11.08, 32.61)
Odds ratio (95% CI)
10.22 (6.92, 15.10)
11.54 (3.62, 36.77)
7.38 (3.53, 15.43)
35.40 (24.57, 51.01)
14.00 (11.23, 17.45)
8.80 (4.38, 17.68)
6.14 (0.89, 42.31)
13.60 (9.08, 20.36)
19.01 (11.08, 32.61)
Odds ratio (95% CI)
10.22 (6.92, 15.10)
11.54 (3.62, 36.77)
7.38 (3.53, 15.43)
35.40 (24.57, 51.01)
14.00 (11.23, 17.45)
8.80 (4.38, 17.68)
6.14 (0.89, 42.31)
Better Poorer 1.0196 1 51
WFNS V
89
3.4 PART C: Neuroimaging factors: Fisher CT clot burden, aneurysm
location and size
Neuroimaging characteristics at hospital admission play an important role in the accurate
diagnosis of SAH from ruptured intracranial aneurysms.25
They help identify the aneurysm that
is the source of bleeding; and delineate its anatomical configuration and that of contiguous
structures to guide optimal choice of treatment modality. Despite several studies the value of
neuroimaging parameters for prognostication of patient outcomes has not been conclusively
determined. Among these parameters, the Fisher grade of CT SAH clot burden,161
aneurysm
location and size have been mostly studied.124
Fisher grade of CT clot burden is considered a
predictor of risk of symptomatic vasospasm160
but no consensus exists as to its independent
association with clinical outcomes after SAH.83, 94, 149, 156-159
Whereas ruptured aneurysm location
is associated with the risk of periprocedural complications164
and outcome in some studies,141, 149,
165 some other studies have found no significant relationship between aneurysm location and
SAH outcomes.82, 83, 94, 157, 159, 166, 167
Studies have shown that larger diameter aneurysms
increases the risk of rebleeding,154, 155
which potentially could worsen outcome, the literature
also contain conflicting results with respect to a direct association between ruptured aneurysm
diameter and SAH outcome.141, 149, 155, 159, 167
Often, this parameter has been dichotomized
differently in different studies.94, 141, 149, 155, 159, 164, 167, 226
Advances in diagnostic and interventional neuroradiology have raised further queries
about the value of neuroimaging parameters for prognostication in SAH.51, 162
Most prognostic
studies investigating neuroimaging characteristics have scarcely accounted for differences in
method of aneurysm repair.80, 83, 141, 167
With increasing adoption of endovascular coil
embolization, accounting for differences in treatment modality may be needed to better
90
understand and more accurately evaluate the value of neuroimaging characteristics as prognostic
factors in SAH.
This part of the chapter focused on the associations between CT SAH clot burden graded
on the Fisher scale, ruptured aneurysm location and diameter and 3-month outcome on the
Glasgow Outcome Scale (GOS). The primary objective was to investigate the role of these
neuroimaging characteristics as prognostic factors for 3-month outcome on the GOS.
3.4.1 Methods
Data at hospital admission was available on ruptured aneurysm diameter and location in 7
studies and CT SAH clot burden in 8 studies of the 14 studies in the repository. Fisher CT clot
burden was estimated from the modified Fisher grade in the SHOP dataset (Table 3.5) and from
CT clot size (classified as thick or thin) and location (localized or diffused) with or without the
presence of intraventricular hemorrhage in the Tirilazad and CONSCIOUS 1 trials datasets.
Ruptured aneurysm diameter was available as a continuous variable in 5 studies, and as a
categorical variable in the Tirilazad and CONSCIOUS 1 datasets. Aneurysm diameter was
analysed primarily as a continuous predictor, and secondarily as a categorical variable so as to
include patients for whom data has been dichotomized a priori and to evaluate consistency in the
results of both analyses. For the latter, aneurysm diameter was categorized as small (1-12mm),
large (13-24mm), and giant (≥25mm). Ruptured aneurysm locations were categorized broadly
into anterior cerebral artery (ACA), internal carotid artery (ICA; including posterior
communicating region), middle cerebral artery (MCA) and posterior circulation (PCQ). The
proportion of data imputed was 1.6% for analysis of the effect of aneurysm location and effect of
aneurysm size, and 3% for analysis of the effect of Fisher grade.
91
3.4.2 Results
The study cohort consisted of 9125 patients for the analysis of the effect of ruptured
aneurysm location and size, and 9452 patients for the analysis of the effect of CT clot burden on
the Fisher grade. Median aneurysm diameter was 6mm (IQR: 4-9mm). Aneurysm diameter
significantly varied by treatment modality, with smaller aneurysms preferentially treated by
coiling (coiling: 6.5±3.4mm; clipping: 8.01±5.6mm; none: 9.32±10.2mm; ANOVA p<0.001).
Most aneurysms were located in the anterior circulation (89%). The distributions of SAH clot
burden, aneurysm location and size across studies are shown in Table 3.5. Proportion of small
aneurysms was higher in RCT studies relative to observational studies (73-96% versus 47-67%)
whereas the converse was the case in respect of large diameter aneurysms (3-23% versus 9-
51%). Posterior circulation aneurysms were relatively fewer in RCT studies (3-14%) compared
with observational studies (18-19%). Aneurysms of the middle cerebral artery were
preferentially treated by clipping while aneurysms of the posterior circulation were preferentially
treated conservatively or with endovascular coiling (Table 3.6). A greater proportion of patients
were classified as Fisher grade 3 (Table 3.5).
Distribution of studied neuroimaging factors by 3-month GOS is provided in Table 3.7. A
U-shaped relationship was noted between aneurysm diameter and GOS outcome, with best
outcomes indicated at a diameter of 5.5mm (Figure 3.8). However, the improvement in model fit
achieved by fitting age with a nonlinear (square) rather than a linear term was not statistically
significant (p=0.76). Meta-analyses of univariable associations (Figure 3.9) demonstrated poorer
outcome with increasing aneurysm diameter; and poorer outcome in patients with posterior
circulation aneurysms relative to those with anterior circulation aneurysms. The 95% confidence
intervals however included an odds ratio of 1, indicating that the associations were not
92
significant at the 5 % significance level. In multivariable analyses (Table 3.8), increasing
aneurysm diameter was associated with poor outcome, adjusting for the core prognostic factors
of age and neurologic status (Model B: OR, 1.10; 95% CI: 1.05-1.15 for 9mm vs. 4mm i.e. 75th
vs. 25th
percentile); and adjusting for other neuroimaging factors had no further explanatory
effect (Model C: OR, 1.09; 95% CI: 1.04-1.15). Further adjusting for treatment modality resulted
to loss of significant association (Model D: OR, 1.03; 95% CI: 0.98-1.09). Similarly, aneurysm
location was associated with outcome. Compared with patients with anterior cerebral artery
aneurysms, patients with posterior circulation aneurysms were at higher risk of poor outcome;
the effect was independent of age and neurologic status (OR: 1.25, 95% CI: 1.08-1.44;
p=0.0023) and of neuroimaging covariates (OR: 1.20, 95% CI: 1.04-1.39; p=0.0037). Adjusting
for treatment modality resulted to a loss of the significant relation between aneurysm location
and outcome (OR: 1.03; 95% CI: 0.98-1.09; p=0.376).
An interaction effect was noted between aneurysm diameter and patient neurologic status
(p<0.001), and method of aneurysm treatment (p=0.005) but not between aneurysm size and age
(p=0.226). An interaction was demonstrated between aneurysm location and treatment modality
(p=0.0002) but no interaction was found between aneurysm location and neurologic status
(p=0.52), or age (p=0.85). The results of analysing aneurysm diameter as a continuous predictor
were not altered when aneurysm diameter was analysed as a categorical predictor.
Increasing Fisher grade of SAH clot burden was univariably associated with poorer
outcomes across studies (Figure 3.10). However considerable between study heterogeneity was
noted in the prognostic effect of Fisher CT clot burden; particularly for Fisher grade 3 and 4. In
adjusted analysis (Table 3.8), increasing Fisher CT SAH clot burden was significantly related to
poorer outcome in a gradient manner (p<0.001), however with overlap in confidence intervals of
93
preceding categories. In the full model (Model D), the adjusted Odds ratio for Fisher grade 2 was
1.26, 95% CI: 1.04-1.53; Fisher grade 3 was OR, 1.77, 95% CI: 1.48-2.10; and Fisher grade 4
was OR, 1.86, 95% CI: 1.54-2.26.
Each neuroimaging predictor explained less than 1% of the variability in GOS outcome
in the models estimating their adjusted effects (Figure 3.4).
3.4.3 Discussion
Prognostic studies evaluating association of neuroimaging parameters with outcome in
patients with SAH have come to conclusions that are conflicting with regard to the association of
Fisher CT clot burden, aneurysm location, diameter and outcome.94, 141, 149, 155, 159, 164, 167, 226
Utilizing patient data far exceeding those of any previous studies in number, the present study
found that Fisher CT clot burden, aneurysm location and diameter are each separately associated
with 3-month outcome on the GOS following hospital admission for SAH. These neuroimaging
parameters however added only marginal incremental value beyond the effect of neurologic
status and age. The associations of ruptured aneurysm location and diameter with outcome
differed with, and were attenuated by accounting for treatment modality. This finding suggests
that results of prior studies which did not fully account for treatment modality (whether
aneurysms were treated conservatively, or by surgical clipping or endovascular coiling) may
have been confounded by treatment selection bias which may have contributed, in part, to the
conflicting results of these studies. The finding could also suggest that appropriate selection of
patients for treatment modality mitigates the prognostic relationship of aneurysm location and
diameter to clinical outcome.
94
Because aneurysm diameter has commonly been analysed as a categorical predictor, with
different threshold values applied in different studies, including the use of 10mm155, 159, 164, 226
,
13mm94
, or the use of different multiple categories141, 167
, comparison of reported results is
difficult. In this study, aneurysm diameter was analysed as a continuous and categorical
predictor, and the former was found to be a more optimal approach, which showed that a change
point in the prognostic effect of aneurysm diameter, if present, would be around the 5mm
diameter; though the effect could be adequately described by a linear function.
The gradient association of Fisher CT clot burden with outcome in this study was also the
finding of another study83
, but not that of other previous studies that found no gradient effect158
or no significant association between Fisher CT clot burden and SAH outcome.80, 94, 149
The
considerable between-study heterogeneity in the effect of Fisher CT clot burden, seen
particularly for grade 3 and 4, may be due to a number of reasons including the subjective nature
of the Fisher scale which has been shown to have a high measurement variability51, 227
, or
because the Fisher grade was estimated in 2 studies based on a simple re-categorization of SAH
clot thickness and location. Nonetheless, in some respect, the results further underscore the need
for better methods of grading subarachnoid blood volume, density and distribution on CT scan
images.
In summary, Fisher CT clot burden, ruptured aneurysm location and diameter are weak
prognostic factors for clinical outcome of SAH, adding very small predictive value to the core
prognostic factors of age and neurologic status. Accounting for modality of treatment completely
attenuates the prognostic effect of aneurysm diameter and location.
95
Table 3.5 – Distribution of ruptured aneurysm location, diameter and Fisher clot burden
by study
CONS_1
N=433
HHU
N=60
IMASH
N=327
IHAST
N=998
ISAT
N=2143
TIRILAZAD
N=3552
D-SAT
N=439
SHOP
N=1500
Location: ACA 174(42) 20(33) - 391(40) 1085(51) 1256(36) 137(31) 395(32)
ICA 118(29) 11(18) - 309(31) 697(32) 1046(30) 130(30) 412(33)
MCA 76(18) 21(35) - 205(21) 303(14) 711(20) 88(20) 209(17)
PCQ 46(11) 8(14) - 83(8) 58(3) 474(14) 84(19) 218(18)
Diameter:
Median(IQR)
- 5(4,8) - 7(5,10) 5(4,7) - 16(3,22) 7(5,10)
Fisher: 1 3 0 2(1) 53(5) 114(6) 338(9) 68(16) 210(15)
2 68(16) 2(3) 24(7) 342(34) 360(17) 455(13) 47(11) 315(22)
3 345(80) 9(15) 262(80) 473(48) 902(42) 2315(66) 184(41) 695(48)
4 17(4) 49(82) 39(12) 130(13) 753(35) 420(12) 142(32) 218(15) C-1: CONSCIOUS 1 trial; IHAST: Intraoperative Hypothermia for Aneurysm Surgery Trial; IMASH: Intravenous
Magnesium Sulphate for Aneurysmal Subarachnoid Hemorrhage; D-SAT: Database of Subarachnoid Treatment of
the University of Washington; SHOP: Subarachnoid Hemorrhage Outcomes project of Columbia University HHU:
HHU: Heinrich-Heine University Concomitant Intraventricular Fibrinolysis and Low-Frequency Rotation after
Severe Subarachnoid Hemorrhage Trial
Table 3.6 – Distribution of aneurysm location by treatment modality
Treatment modality
Location Clipping Coiling None Total (%)
ACA 2617 (38.9) 716 (44.7) 125 (28.5) 3458 (39.5)
ICA 2096 (31.2) 501 (31.3) 126 (28.7) 2723 (31.0)
MCA 1358 (20.2) 203 (12.7) 52 (11.9) 1613 (18.4)
PCQ 653 (9.7) 182 (11.3) 136 (30.9) 971 (11.1)
Table 3.7 – Distribution of aneurysm location, diameter and Fisher clot burden by GOS
Good Moderate Severe Vegetative Dead
Location: ACA 1629 (39.4) 727 (40.9) 482 (42.1) 116 (51.1) 317 (33.7)
ICA 1330 (32.1) 544 (30.6) 316 (27.6) 58 (25.5) 286 (30.4)
MCA 749 (18.1) 327 (18.4) 218 (19.0) 31 (13.7) 176 (18.7)
PCQ 430 (10.4) 181 (10.1) 130 (11.3) 22 (9.7) 161 (17.1)
Diameter: small 3519 (83.0) 1473 (82.1) 945 (81.8) 197 (86.8) 603 (58.6)
Large 556 (13.1) 254 (14.1) 168 (14.5) 27 (11.9) 262 (25.5)
Giant 164 (3.9) 68 (3.8) 43 (3.7) 3 (1.3) 163 (15.9)
Fisher: 1 500 (11.5) 150 (8.0) 36 (3.0) 2 (0.8) 36 (3.3)
2 950 (21.8) 330 (17.7) 141 (11.6) 12 (4.5) 78 (7.3)
3 2222 (51.1) 1048 (56.1) 724 (59.8) 155 (59.2) 757 (70.3)
4 679 (15.6) 341 (18.2) 310 (25.6) 93 (35.5) 206 (19.1)
96
Table 3.8 – Adjusted effects of studied neuroimaging factors
Model A Model B Model C Model D
Location: ACA referent
ICA 0.91(0.76-1.09) 1.01(0.91-1.12) 1.00(0.90-1.11) 0.99(0.89-1.10)
MCA 1.03(0.93-1.13) 0.92(0.81-1.04) 0.89(0.78-1.00) 0.91(0.81-1.03)
PCQ 1.17(1.04-1.32) 1.25(1.08-1.44) 1.20(1.04-1.39) 1.10(0.95-1.28)
Diameter: small referent
Large 1.65(1.31-2.07) 1.44(1.27-1.64) 1.47(1.29-1.66) 1.42(1.25-1.61)
Giant 2.37(1.61-3.51) 1.95(1.61-2.36) 1.97(1.63-2.39) 1.15(0.93-1.43)
Diameter (75th
vs.
25th
percentile)
1.13(1.08-1.19) 1.10(1.05-1.15) 1.09(1.04-1.15) 1.03(0.98-1.09)
Fisher grade: 1 referent
2 1.48(1.28-1.77) 1.17(0.96-1.41) 1.21(1.00-1.47) 1.26(1.04-1.53)
3 3.29(2.79-3.87) 1.68(1.41-1.99) 1.74(1.46-2.07) 1.77(1.48-2.10)
4 3.89(3.26-4.64) 1.75(1.45-2.11) 1.79(1.48-2.17) 1.86(1.54-2.26) NB: Data on aneurysm diameter is presented when analysed as continuous and as categorical variable
Analysis was done separately for each neuroimaging characteristic
Model A: Predictor (CT clot burden or Aneurysm location or diameter) + Study
Model B: Model A + WFNS+Age
Model C: Model B + Neuroimaging data (Fisher grade+ Artery + Ruptured aneurysm size, as applicable)
Model D: Model C + Repair (clipping vs. coiling vs. conservative
Figure 3.8 – U-shaped relation of aneurysm size to GOS outcome with change point at 5.5mm
Ruptured aneurysm diameter in mm
Pro
ba
bili
ty o
f po
or
ou
tco
me
s
0.5
0.6
0.7
0.8
0 10 20 30
97
Figure 3.9 – Forest plot to examine consistency in the relation of aneurysm location and diameter
to outcome across studies
NOTE: Weights are from random effects analysis
Overall (I-squared = 64.9%, p = 0.009)
HHU
ISAT
IHAST
D-SAT
SHOP
TIRILAZAD
C-1
Study
0.93 (0.76, 1.13)
0.20 (0.04, 1.00)
1.13 (0.94, 1.36)
0.86 (0.63, 1.17)
1.48 (0.95, 2.29)
0.78 (0.59, 1.04)
0.80 (0.68, 0.94)
0.93 (0.60, 1.45)
Odds
ratio (95% CI)
0.93 (0.76, 1.13)
0.20 (0.04, 1.00)
1.13 (0.94, 1.36)
0.86 (0.63, 1.17)
1.48 (0.95, 2.29)
0.78 (0.59, 1.04)
0.80 (0.68, 0.94)
0.93 (0.60, 1.45)
Odds
ratio (95% CI)
Better Poorer 1.0405 1 24.7
Internal carotid artery
NOTE: Weights are from random effects analysis
Overall (I-squared = 40.0%, p = 0.124)
IHAST
D-SAT
HHU
Study
SHOP
ISAT
TIRILAZAD
C-1
1.05 (0.88, 1.24)
0.83 (0.59, 1.18)
1.85 (1.14, 3.00)
0.63 (0.20, 2.03)
ratio (95% CI)
1.06 (0.74, 1.53)
Odds
0.94 (0.73, 1.21)
0.99 (0.83, 1.18)
1.41 (0.86, 2.34)
1.05 (0.88, 1.24)
0.83 (0.59, 1.18)
1.85 (1.14, 3.00)
0.63 (0.20, 2.03)
ratio (95% CI)
1.06 (0.74, 1.53)
Odds
0.94 (0.73, 1.21)
0.99 (0.83, 1.18)
1.41 (0.86, 2.34)
Better Poorer 1.195 1 5.12
Middle cerebral artery
NOTE: Weights are from random effects analysis
Overall (I-squared = 27.4%, p = 0.219)
SHOP
IHAST
Study
TIRILAZAD
HHU
C-1
ISAT
D-SAT
1.21 (1.00, 1.47)
Odds
0.93 (0.67, 1.30)
1.56 (0.99, 2.47)
ratio (95% CI)
1.15 (0.94, 1.40)
0.00 (0.00, .)
1.85 (1.01, 3.39)
0.90 (0.53, 1.51)
1.63 (0.99, 2.70)
1.21 (1.00, 1.47)
Odds
0.93 (0.67, 1.30)
1.56 (0.99, 2.47)
ratio (95% CI)
1.15 (0.94, 1.40)
0.00 (0.00, .)
1.85 (1.01, 3.39)
0.90 (0.53, 1.51)
1.63 (0.99, 2.70)
Better Poorer 11.0e-08 1 1.0e+08
Posterior circulation
NOTE: Weights are from random effects analysis
Overall (I-squared = 88.1%, p = 0.000)
ISAT
SHOP
HHU
IHAST
Study
D-SAT
1.17 (0.97, 1.41)
1.19 (1.03, 1.36)
1.48 (1.31, 1.66)
1.17 (0.58, 2.38)
1.15 (1.02, 1.29)
ratio (95% CI)
0.94 (0.85, 1.04)
Odds
1.17 (0.97, 1.41)
1.19 (1.03, 1.36)
1.48 (1.31, 1.66)
1.17 (0.58, 2.38)
1.15 (1.02, 1.29)
ratio (95% CI)
0.94 (0.85, 1.04)
Odds
Better Poorer 1.421 1 2.38
Aneurysm diameter (9mm vs. 4mm)
98
Figure 3.10 – Forest plot illustrating prognostic strength of SAH clot burden across studies
NOTE: Weights are from random effects analysis
Overall (I-squared = 36.6%, p = 0.163)
D-SAT
SHOP
Study
TIRILAZAD
ISAT
IHAST
C-1
1.31 (1.00, 1.72)
1.28 (0.62, 2.64)
1.82 (1.23, 2.70)
ratio (95% CI)
1.13 (0.82, 1.57)
0.94 (0.63, 1.41)
2.14 (1.01, 4.52)
0.80 (0.08, 7.84)
Odds
1.31 (1.00, 1.72)
1.28 (0.62, 2.64)
1.82 (1.23, 2.70)
ratio (95% CI)
1.13 (0.82, 1.57)
0.94 (0.63, 1.41)
2.14 (1.01, 4.52)
0.80 (0.08, 7.84)
Odds
Better Poorer 1.0825 1 12.1
Fisher Grade 2
NOTE: Weights are from random effects analysis
Overall (I-squared = 81.5%, p = 0.000)
ISAT
Study
SHOP
TIRILAZAD
IHAST
C-1
D-SAT
3.10 (1.96, 4.89)
1.35 (0.94, 1.96)
Odds
ratio (95% CI)
4.09 (2.89, 5.80)
3.39 (2.61, 4.42)
3.19 (1.53, 6.65)
1.72 (0.18, 16.19)
5.73 (3.28, 9.98)
3.10 (1.96, 4.89)
1.35 (0.94, 1.96)
Odds
ratio (95% CI)
4.09 (2.89, 5.80)
3.39 (2.61, 4.42)
3.19 (1.53, 6.65)
1.72 (0.18, 16.19)
5.73 (3.28, 9.98)
Better Poorer 1.0618 1 16.2
Fisher Grade 3
NOTE: Weights are from random effects analysis
Overall (I-squared = 87.5%, p = 0.000)
ISAT
SHOP
C-1
IHAST
TIRILAZAD
D-SAT
Study
3.68 (2.00, 6.75)
2.16 (1.48, 3.14)
10.54 (6.92, 16.05)
1.01 (0.09, 11.30)
5.12 (2.34, 11.25)
2.43 (1.78, 3.32)
3.58 (2.03, 6.32)
Odds ratio (95% CI)
3.68 (2.00, 6.75)
2.16 (1.48, 3.14)
10.54 (6.92, 16.05)
1.01 (0.09, 11.30)
5.12 (2.34, 11.25)
2.43 (1.78, 3.32)
3.58 (2.03, 6.32)
Odds ratio (95% CI)
Better Poorer
1.0623 1 16.1
Fisher Grade 4
99
Chapter 4
SAHIT score: Novel prognostic scores for early prediction of outcome after
aneurysmal subarachnoid hemorrhage
This chapter is adapted from the following manuscript being prepared for publication as follows:
Jaja BN, Macdonald RL, Lingsma H, Thorpe KE, Saposnik G, Mohammed M, Steyerberg EW,
Schweizer TA, for the SAHIT Investigators. SAHIT score: international prognostic score for
aneurysmal subarachnoid hemorrhage.
100
4.1 Introduction
Subarachnoid hemorrhage from ruptured intracranial aneurysms has a variable clinical
course and outcome,7 hence accurate prognostication in patients who are affected by this
condition is challenging. In one series, 20% of patients with the poorest neurologic status at
hospital admission made good functional and cognitive recovery, going on to live a normal
life.228
On the other hand, patients who initially were expected to have good recovery have
deteriorated clinically, eventually dying from the condition.191
Research has shown that
physician’s prognostication in patients with SAH is not always accurate; in some cases they may
make overly pessimistic predictions of outcome, which potentially could lead to withholding
treatment that otherwise could be lifesaving.229
Clinical prediction models statistically combine a set of patient and disease
characteristics to estimate the probability of an outcome.75
They could therefore be useful as
decision support tools to reinforce clinician’s prognostication, and in the complex choices they
make regarding patient management, and to facilitate counseling of patient and family members
for informed shared decision making.75
Additionally, they could be useful in clinical research to
improve the design and analysis of randomized trials and in health services research to adjust for
case mix variations in the comparison of outcomes and performance assessments among
different care settings.75
Many prediction models and risk scores have been published to estimate
the probability of death or functional outcome in patients with SAH. In a review, we
demonstrated scope for improvement in the development of prediction models and risk scores in
SAH, particularly a need for increased study power and use of patient cohorts representing a
101
broad spectrum of settings to develop models and risk scores with greater reliability, predictive
accuracy and applicability.124
The large sample size and diversity of patient characteristics and practice patterns
represented by data in the SAHIT repository presents a unique opportunity to develop reliable
prediction models and risk scores in SAH that are applicable to a wide variety of settings. This
chapter therefore tests the hypothesis that prediction models based on the readily obtained
parameters at hospital admission of patients age, premorbid hypertension, neurologic status on
the WFNS scale, Fisher CT clot burden, aneurysm location and size, and envisioned treatment
modality will have adequate predictive accuracy in a large, heterogeneous development sample,
and potential to perform satisfactorily in new cohorts.
4.2 Methods
4.2.1 Study population
The study cohort consisted of SAH patients whose information are available in the
SAHIT repository and for whom outcomes on the Glasgow Outcome Score (GOS) was available.
Abstracted data included those of 7 randomized clinical trials (RCT) 58, 200-205
and 2 prospective
observational hospital registries.207-209
The design and characteristics of patients in the different
studies have been reviewed in chapter 2 (Table 2.1). Of note is that all studies reported
prospectively collecting patient records at hospital admission, and none of the clinical trials
found significant treatment effect for the intervention being studied, allowing for the pooling of
data of both treatment and control groups in the analysis; except for the ISAT trials58
where very
few patients were still enrolled up to 24 days after rupture, and where outcome differences were
reported between patients who were treated by surgical clipping and those who were treated by
102
endovascular coiling. To address this, treatment modality was included as a predictor in the
prediction models.
4.2.2 Variable selection
We completed a systematic review of prognostic models in aneurysmal subarachnoid
hemorrhage identifying independent predictors of poor outcome in multivariable analysis to
inform variable selection for the prediction models (Table 1.2). The review was conducted using
established methods according to the PRISMA guidelines. Eligible studies were those whose
primary endpoint was mortality or outcome on the Modified Rankin scale or Glasgow outcome
scale. Readily obtained admission parameters most consistently identified as predictors of
outcome were age, neurologic status measured on the World Federation of Neurological
Surgeons (WFNS) scale, premorbid history of hypertension, and Fisher grade of SAH volume on
CT scan images, ruptured aneurysm size and location. Their prognostic value in the SAHIT
cohort was explored in chapter 3.
4.2.3 Outcome measure
The outcome measure was the Glasgow Outcome Score (GOS) at 3 months. The GOS
consists of five ordered categories including GOS 1, dead; 2, persistent vegetative state; 3, severe
disability; 4, moderate disability; 5, good recovery. The GOS was dichotomized to develop
prediction models for mortality at 3 months (GOS 1 versus GOS 2-5) and for unfavorable
outcomes at 3 months (GOS 1, 2, 3 versus GOS 4, 5).
4.2.4 Model development
The distributions of categorical variables were examined by frequency tables; the
distributions of categorical variable across GOS were examined by box plots. The likelihood of
103
nonlinearity in the effect of age was explored using restricted cubic splines. The association
between predictor variables and GOS was analysed by fitting proportional odds logistic
regression models. Prognostic strength was quantified as odds ratios (OR) with 95% confidence
intervals. The relative importance of each predictor in the model was estimated as difference in
Nagelkerke’s R2 values when the predictor was added or removed from the model. The
difference in R2 values also called partial R
2 estimates the independent contribution of the
predictor to the variance of GOS outcome. Two sets of sequential prediction models were
developed for early prediction of mortality and unfavorable outcomes at 3 months respectively
according to the GOS. This was done by fitting binary logistic regression models. We developed:
(1) core model consisting of age, WFNS score, and history of premorbid hypertension as
predictor variables; (2) neuroimaging model consisting of the core model plus Fisher grade of
CT SAH clot burden, ruptured aneurysm lumen size and location; (3) Repair model consisting of
the neuroimaging model plus modality of aneurysm repair (whether aneurysm was treated by
clipping or coiling or conservatively).
Multiple imputations were performed for missing data according to the protocol set out in
chapter 2. In all, we imputed 4309 of the 54,680 values (7.9%) needed for the core model; 9286
of the 87,488 values (10.6%) needed for the neuroimaging model; and 9286 of the 98,424 values
(9.4%) needed for the repair model.
4.2.5 Model performance
The overall predictive performance of the models was evaluated using Nagelkerke’s R2
statistics.230
Models performance was further assessed by discrimination: the ability of the
models to differentiate between patients who had or who did not have the outcome, using c-
statistics which is equivalent to the area under the receiver operator characteristics curve (AUC).
104
Model performance was also assessed by calibration: the ability of the models to produce
unbiased estimates of the outcome probability. We did this graphically using calibration plots
(plots of observed versus predicted outcomes) and statistically by computing the following
measures of calibration: (1) a goodness of fit test of the model in the development cohort; (2)
calibration-in-the-large which estimates the difference between the average of observed
outcomes and the average of predicted outcomes, and corresponds to the intercept of the
regression model refitted with the linear predictors only. A value of zero indicates perfect
calibration. Values lesser than or greater than zero indicate average under- or over-estimation of
the outcome respectively; (3) recalibration slope, which is the slope of the refitted model; a value
of 1 indicates perfect agreement between observed and predicted outcomes and values lesser
than or greater than 1 indicate average under- or over-estimation of the outcome respectively.182
4.2.6 Model Validation
Internal validation evaluates the stability of a prediction model to changes in sample
composition.214
Internal validation was performed by the bootstrap resampling technique.
Regression models were fitted in 500 bootstrap samples, drawn with replacement from the
development sample. The model was refitted in each bootstrap sample and tested on the original
sample to estimate optimism in performance (R2 values, C-statistics). The optimism was
subtracted from the apparent performance estimates in the development sample.231
We further
assessed the performance of the models at cross validation by omission of each of the 10 studies
in turn. This procedure provided an assessment of the potential of the models to predict
accurately when applied to new cohorts from a variety of settings. Significance level was set at
5%.
105
4.3 Results
The research cohort consisted of 10 936 patients, comprising 8997 patients who were
enrolled into clinical trials in SAH and 1939 unselected hospital patients. Patients’ characteristics
are shown in Table 4.1. Their median age was 53 years (IQR: 44-62 years) and most patients
were women (71%). The proportion of patients who died within 3 months was 13% and the
proportion that experienced unfavorable outcome was 29%. There was a continuous relationship
between age and GOS outcome (Figure 4.1) which could be adequately approximated as a linear
function (chapter 3 part A). Table 4.2 shows the association between predictors and 3-month
GOS outcome. The predictor variables were all significantly associated with outcome. There was
no significant correlation between the individual predictor variables. The predictor with the
strongest prognostic value, in terms of partial R2, was neurologic status (R
2 = 12.03%), followed
by age (R2 = 1.91%), then treatment modality (R
2 = 1.25%), Fisher grade of CT clot burden (R
2 =
0.65%), premorbid history of hypertension (R2 = 0.37%), aneurysm size (R
2 = 0.12%) and lastly
aneurysm location (R2 = 0.06%).
4.3.1 Model performance
The predictors included in the models explained 23 – 31% of the variability in GOS
outcomes (Table 4.3). The models adequately discriminated between patients who had or did not
have the outcomes (Bootstrap AUC: 0.77 – 0.83). No significant evidence was found for lack of
fit in the development cohort (goodness of fit tests p = 0.15 – 0.75; Table 4.3). At cross
validation, we found varying levels of miscalibration across studies. However, agreement
between predicted and observed outcomes was generally satisfactory across studies, even in RCT
cohorts who are selected with respect to the general SAH population (Table 4.4). Predictive
performance was lowest in the CONSCIOUS I cohorts (Table 4.5 and Figure 4.2: R2: 4 – 15%;
106
AUC: 0.64 – 0.72). The CONSCIOUS I cohorts were selected to maximise the likelihood of
vasospasm, which may explain the less than satisfactory performance. There was better
discrimination (AUC: 0.75 – 0.79) and less miscalibration (Figure 4.3) in the Tirilazad cohort,
who are older cohorts treated by surgical clipping or conservatively. In the IHAST cohort, all of
whom are good grade patients, the AUC was 0.70 – 0.74. There was good agreement between
predicted and observed outcomes for the models predicting risk of mortality and the models
predicting risk of unfavorable outcome (Table 4.4 and Figure 4.4).
Cross validation plots demonstrated satisfactory agreement between predicted and
observed outcomes in patients in the D-SAT and SHOP registries (Figures 4.5 and 4.6). Table
4.4 shows that calibration-in-the-large values were closer to the ideal value of zero and
recalibration slope values were closer to the ideal value of 1 in the D-SAT and SHOP cohorts
compared with RCT cohorts with values which were farther from these ideal values; indicating
lesser miscalibration in unselected hospital cohorts than RCT cohorts. The SHOP cohorts had the
highest R2 values (41 – 46%) and AUC (0.82 – 0.86). Performance indices were slightly better
for the models predicting risk of unfavorable outcome (R2: 15 – 46%; AUC: 0.67 – 0.85) than for
the models predicting risk of mortality (R2: 5 – 42%; AUC: 0.64 – 0.86). Extending the core
models to include neuroimaging parameters and treatment choices improved AUC by 0.01 on
average, but not uniformly in all study cohorts (Table 4.4).
4.3.2 Model presentation
A simple score chart is presented in Figure 4.7 which could be used to compute the
probability of mortality within 3-months of hospital admission with 95% confidence intervals
based on the core model. The scores are derived from rounding of model coefficients (Table 4.5).
107
4.4 Discussion
Aneurysmal subarachnoid hemorrhage is a cerebrovascular emergency that could lead to
death and disability in young adults. Currently, clinicians have limited tools to estimate clinical
outcomes early after hospitalization for SAH. This chapter presents a novel set of prediction
models which were developed for early prediction of mortality or unfavorable outcome
according to the GOS at 3 months. Variables for inclusion in the models were selected after a
systematic review to identify prognostic factors that would be useful for early prediction of
outcome in SAH. The predictors most regularly shown to be associated with outcome in prior
studies were neurologic status, age, Fisher grade of CT clot burden, premorbid history of
hypertension, aneurysm size and aneurysm location. Other variables were less regularly
identified as such (Table 1.2). Consistent with the conclusions of prior studies, the present study
demonstrated that patient neurologic status accounted for the greater proportion of the combined
predictive value of studied predictor variables. Other predictors added only small predictive
information. The choice of treatment for aneurysm repair could have prognostic information. 83
In this study, it was the third strongest prognostic factor, having predictive information that was
slightly lower than that of age and greater than those of subarachnoid clot burden on CT scan,
premorbid history of hypertension, aneurysm size and aneurysm location. Treatment modality is
a decision usually made very soon after admission. Extending the prediction models to include
treatment modality may help minimize the effect of self-fulfilling prophecy due to treatment
choices on the outcome predictions derived from the models.
Many prognostic models and risk scores have been reported for early prediction of
outcomes in patients with SAH. Most used predictors similar to the ones of the present study 124
83, 94, 141-143, 146, 157-159, 167, 181; however our models differ from previous models and risk scores in
108
certain important respects. Previous studies have commonly developed prediction models and
risk scores from small patient cohorts often representing experience from a single center; these
models or risk scores may be less reliable or are applicable to a much narrower setting.124
A few
previous studies used large samples which were derived exclusively from datasets of randomized
clinical trials in SAH, including those of the cooperative aneurysm study of intravenous
nicardipine after aneurysmal SAH (NICSAH),143
the tirilazad trials in SAH,141
and the
International subarachnoid aneurysm trials (ISAT).83
The homogenous nature of randomized
clinical trial cohorts may constrain the application of prognostic models developed in trial
cohorts to unselected hospital patients. A relatively older prognostic model with greater potential
for generalizability to a wider cohort of SAH patients was developed from 3521 patients enrolled
between 1980 and 1983 into the prospective, multicenter observational International Cooperative
Study on the Timing of Aneurysm Surgery.192
This model however preceded the widespread use
of nimodipine, early aneurysm repair and increasing adoption of endovascular coiling in
appropriately selected patients; treatment changes that have improved outcomes of SAH patients
and may potentially influence the reliability of the model.
It is likely that the set of prediction models presented in this chapter have greater
reliability and potential to generalize to a broad spectrum of settings than previous models and
risk scores in SAH. The cohorts are well described and encompassed 10936 SAH patients; this
large sample size being about thrice the sample size of the most statistically powered prior
studies. The cohorts were derived from prospectively collected, multicenter randomized clinical
trials as well as unselected hospital registries patients, hence they are more reflective of a broad
spectrum of settings in terms of patient composition, practice patterns, geographic and temporal
settings. The models demonstrated adequate discrimination at bootstrap validation and at cross
109
validation. Though the latter indicated the possibility of systematic miscalibration when the
models are applied to new patients, agreement between predicted and observed outcomes were
generally satisfactory, particularly in less selected patients including those seen at tertiary
hospitals treating SAH patients. It is likely that a simple setting-specific recalibration may be
sufficient for accurate application of the SAHIT models to individual patients in different clinical
settings. Though the simpler core models with only 3 parameters have greater face validity, there
was some evidence that extending the models to include neuroimaging parameters and treatment
choice improved predictive ability, as indicated by the slight increase in R2, AUC and lesser
miscalibration with increasing model complexity.
Physicians often rely on clinical experience and intuition for prognosticating patient
outcomes and while generally correct in their assessment of prognosis, very few physicians are
certain about the accuracy of their predictions, and research indicates that patients’ relatives tend
to rely less on physicians intuition.229
Moreover, a recent randomized study revealed that
clinicians are inaccurate at estimating outcomes for ischemic stroke patients. A validated risk
score performed better, and was useful in providing accurate estimation when discussing about
prognostication with patients and their families.232
An objective clinical scoring tool, such as the
SAHIT scores, could be helpful adjuncts to clinicians’ assessment of prognosis, facilitating
evidence based discussions with patients and their relatives around treatment choices, outcomes
and rehabilitation needs.75
They may also be useful in the design and analysis of clinical trials in
SAH, by providing an objective tool to assess potential trial cohorts based on baseline prognostic
distributions so as to target for enrolment those patients who may optimally benefit from the
intervention.75
The SAHIT scores could be useful to adjust for differences in baseline prognostic
risk of trial cohorts during the analysis phase to account for differences in case mix.75
110
The limitations however must be noted. First, the development cohort was weighted
towards patients who were enrolled into RCTs than to general hospital patients, which may
account for the relatively low proportion of patients who died (13%) or who experienced
unfavorable outcome (29%) at 3 months. Second, the patients were enrolled over a wide time
period (1997-2011) during which there has been a trend towards increasing adoption of
endovascular coiling. However, the models would likely be robust to such changes over time
considering that they performed adequately on cross validation in datasets of both older and
more recent studies. Third, because patients’ clinical condition could continue to improve long
after the 3 months end point used in this study, the clinical utility of the SAHIT scores for
accurate prognostication of long term outcomes is uncertain, but could be evaluated in further
studies. Fourth, the reliability of the models may be confounded by measurement errors due to
the subjective nature of the Fisher grade of CT clot burden, the grading scheme having been
shown to be challenging to reliably reproduce in the current era of high resolution cranial CT
scan imaging.51
Fifth, though the prediction models were based on readily obtained hospital
admission parameters that have been shown to be strongly prognostic in SAH, the models’
predictions may be unreliable should complications which could markedly alter the patient’s
clinical course develop during the inpatient course, particularly vasospasm or rebleeding. It
would therefore be inappropriate to apply the models for treatment limiting decisions in
individual patients. Finally, data imputation was done to minimize loss of prognostic information
that could result had all patients with missing data been excluded from the analysis. Although the
proportion of imputed data was reasonable and the method used reliable, it would have been
preferable that data were completely available for all patients included in the study.
111
In conclusion, this chapter presents a novel set of prediction models that were developed
from a large cohort of international patients based on readily obtained reliable hospital admission
parameters for early prediction of mortality or unfavorable outcomes within 3 months of hospital
admission in patients with SAH. The prediction models showed good discrimination between
patients with or without the outcomes and prospect for satisfactory performance in a broad
spectrum of settings. In some settings, a recalibration of the models may be necessary to produce
accurate prediction at individual patient level. The SAHIT scores are herein recommend given
their potential reliability and generalizability, and usefulness in clinical practice and research.
112
Table 4.1 – Baseline distribution of variables by study cohort
Predictor Coding N (%) C-I IHAST IMASH ISAT MASH Tirilazad HHU D-SAT SHOP
Age (years) Median
(25-75 percentile)
53
(44 – 62)
51
(44 – 59)
52
(43 – 60)
57
(48 – 68)
52
(44 – 60)
56
(48 – 65)
52
(42 – 62)
60
(48 – 64)
51
(42 – 62)
55
(45 – 64)
Hypertension No 4607 (63) 300 (69) 601 (60) 198 (61) - 150 (72) 2334 (67) - 277 (63) 747 (52)
Yes 2725 (37) 133 (31) 397 (40) 129 (39) - 57 (28) 1147 (33) - 162 (37) 700 (48)
WFNS grade I (good grade) 5088 (47) 191 (44) 658 (65) 95 (29) 1335 (63) 730 (49) 1289 (36) - 165 (37) 625 (44)
II 2711 (25) 125 (28) 289 (29) 83 (25) 549 (25) 345 (23) 1045 (29) 4 (7) 82 (18) 189 (13)
III 774 (7) 12 (2) 51 (5) 31 (9) 134 (6) 63 (4) 417 (12) 9 (15) 17 (4) 40 (2)
IV 1222 (11) 102 (23) - 81 (25) 74 (3) 219 (15) 355 (10) 21 (35) 84 (19) 286 (20)
V (poor grade) 1039 (10) 3 (1) - 37 (11) 20 (1) 126 (8) 445 (12) 26 (43) 91 (20) 291 (20)
Location ACA 3469 (38) 174 (42) 391 (40) - 1085 (51) 11 (4) 1256 (36) 20 (33) 137 (31) 395 (32)
ICA 2834 (31) 118 (29) 309 (31) - 697 (33) 111 (40) 1046 (30) 11 (18) 130 (30) 412 (33)
MCA 1708 (19) 76 (18) 205 (21) - 303 (14) 95 (34) 711 (20) 21 (35) 88 (20) 209 (17)
PCQ 1033 (12) 46 (11) 83 (8) - 58 (3) 62 (22) 474 (14) 8 (14) 84 (19) 218 (18)
Aneurysm size Small (<12 mm) 7328 (79) 390 (96) 876 (88) - 2078 (97) 143 (70) 2594 (74) 56 (93) 207 (47) 984 (67)
Large (13 – 24 mm) 1337 (15 10 (2) 94 (10) - 63 (3) 14 (7) 800 (23) 4 (7) 222 (51) 130 (9)
Giant (≥ 25 mm) 566 (6) 7 (2) 24 (2) - 2 (0.09) 46 (23) 128 (3) - 9 (2) 350 (24)
Fisher gradea 1 786 (8) - 53 (5) 2 (1) 114 (5) 1 (1) 338 (9) - 15 (16) 210 (15)
2 1635 (17) 68 (16) 342 (34) 24 (7) 360 (17) 22 (11) 455 (13) 2 (3) 47 (11) 315 (22)
3 5226 (55) 345 (80) 473 (47) 262 (80) 902 (42) 43 (21) 2315 (66) 9 (15) 182 (42) 695 (48)
4 1909 (20) 17 (4) 130 (13) 39 (12) 753 (35) 141 (68) 420 (12) 49 (82) 142 (32) 218 (15)
Repair Clipping 7497 (68) 201 (46) 998
(100)
141 (43) 1070 (50) 551 (37) 3209 (90) 36 (60) 439
(100)
852 (57)
Coiling 2503 (23) 232 (54) - 150 (46) 1073 (50) 738 (50) - 24 (40) - 286 (19)
None 936 (9) - - 36 (11) - 195 (13) 343 (10) - - 362 (24)
3-month GOS
outcome
Good 5034 (49) 81 (19) 642 (64) 126 (39) 930 (47) 642 (43) 1975 (57) 26 (48) 167 (38) 445 (39)
Moderate 2328 (22) 219 (50) 213 (21) 60 (18) 528 (27) 441 (30) 462 (13) 6 (11) 127 (29) 272 (24)
Severe 1331 (13) 103 (24) 81 (8) 60 (18) 374 (19) 105 (7) 424 (12) 17 (32) 39 (9) 128 (11)
Vegetative 323 (3) 4 (1) 1 (0.1) 36 (11) 130 (6.7) 59 (4) 59 (2) 4 (7) 9 (2) 21 (2)
Dead 1317 (13) 26 (6) 61 (6) 45 (14) 6 (0.3) 234 (16) 578 (17) 1 (2) 96 (22) 270 (24) Values in brackets represent precentage, unless otherwise indicated. Figures do not include missing data. ACA: Anterior cerebral artery; ICA: Internal carotid artery; MCA:
Middle cerebral artery; PCQ: Posterior circulation. a: Fisher grade was computed from modified Fisher grade in SHOP dataset, and from CT clot size (classified as thick or thin)
and location (localized or diffused) with or without the presence of intraventricular hemorrhage in the Tirilazad and C-1 datasets. C-I: Clazosentan to overcome neurological
ischemia and infarction occurring after SAH (CONSCIOUS 1) trial; IHAST: Intraoperative Hypothermia for Aneurysm Surgery Trial; IMASH: Intravenous Magnesium Sulphate
for Aneurysmal Subarachnoid Hemorrhage; ISAT: International Subarachnoid Aneurysm Trial; MASH (MASH I and II combined: Magnesium Sulfate in Aneurysmal
Subarachnoid Hemorrhage; HHU: Heinrich-Heine University Concomitant Intraventricular Fibrinolysis and Low-Frequency Rotation After Severe Subarachnoid Hemorrhage
Trial; D-SAT: Database of Subarachnoid Treatment of the University of Washington; SHOP: Subarachnoid Hemorrhage Outcomes project of Columbia University.
113
Table 4.2 – Association between predictors and 3-month outcome
3 month outcome (%) Odds ratios with 95% confidence intervals
Coding Mortality Unfavorable Univariable Core model Neuro model Repair model Partial R2 (%)
Age (years) - - 1.82(1.72-1.92) 1.65(1.56 -1.75) 1.59(1.50-1.68) 1.55(1.44-1.64) 1.91
Hypertension:No 544(12) 1094(25) - - - - 0.37
Yes 522(21) 956(38) 1.73(1.57-1.90) 1.29(1.16-1.44) 1.31(1.17-1.46) 1.32(1.18-1.47)
WFNS grade:1 206(4) 674(14) - - - - 12.03
2 239(9) 668(26) 1.83(1.66-2.01) 1.79(1.62-1.97) 1.66(1.51-1.84) 1.69(1.53-1.87)
3 130(18) 325(44) 3.78(3.20-4.46) 3.44(2.93-4.03) 3.19(2.72-3.74) 3.18(2.71-3.72)
4 293(26) 594(53) 5.93(5.13-6.84) 5.35(4.70-6.09) 4.81(4.21-5.49) 4.73(4.14-5.41)
5 444(46) 683(71) 13.06(11.3-15.1) 12.75(10.8-15.0) 10.83(9.14-12.8) 9.81(8.29-11.6)
Location: ACA 317(10) 916(28) - - - - 0.06
ICA 313(12) 705(27) 0.98(0.88-1.08) - 0.98(0.88-1.09) 0.96(0.87-1.07)
MCA 188(12) 450(28) 1.06(0.95-1.18) - 0.83(0.74-0.94) 0.86(0.76-0.96)
PC 180(18) 339(34) 1.27(1.11-1.45) - 1.08(0.94-1.24) 0.97(0.84-1.13)
Size: Small (<12
mm)
628(9) 1801(26) - - - - 0.12
Large (13 – 24 mm) 266(21) 465(36) 1.51(1.35-1.69) - 1.26(1.12-1.41) 1.22(1.08-1.37)
Giant (≥ 25 mm) 174(36) 227(47) 2.34(1.87-2.94) - 1.75(1.39-2.19) 1.21(0.95-1.54)
Fisher grade:1 37(5) 74(10) - - - - 0.65
2 79(5) 233(15) 1.22(1.00-1.49) - 1.24 (1.01-1.52) 1.27(1.03-1.56)
3 766(16) 1649(33) 2.55(2.11-3.09) - 1.72(1.41-2.08) 1.72(1.41-2.09)
4 236(13) 676(38) 3.11(2.53-3.83) - 1.97(1.61-2.40) 2.00(1.63-2.45)
Repair: Clip 712(10) 1881(26) - - - - 1.25
Coil 166(7) 579(25) 1.16(1.06-1.26 - - 1.14(1.03-1.26)
None 439(52) 511(60) 5.09(4.19-6.17) - - 2.66(2.21-3.21)
Odds ratios were derived from proportional odds analysis
114
Table 4.3 – Performance indices of the six models at bootstrap validation
Mortality outcome Unfavorable outcome
R2 AUC GoF R
2 AUC GoF
Core 23.06 0.79 0.74 26.10 0.77 0.15
Neuro 26.41 0.81 0.29 27.14 0.77 0.35
Repair 30.92 0.83 0.20 28.33 0.78 0.72 GoF: Le Cessie - van Houwelingen - Copas - Hosmer test of goodness of fit
Figure 4.1 – Spline plot of the relationship between age and 3-month GOS outcome
115
Table 4.4 – Performance indices at leave-one-study-out cross validation
Core model Neuro model Repair model
Mortality N R2 (%) AUC CIL Slope R
2 (%) AUC CIL Slope R
2 (%) AUC CIL Slope
C-1 433 5 0.67 -0.81 0.66 4 0.65 -0.74 0.60 4 0.64 -0.66 0.60
IHAST 998 11 0.73 -0.35 1.28 10 0.72 -0.28 1.05 10 0.72 -0.15 1.15
IMASH 327 10 0.70 -0.30 0.66 11 0.72 -0.17 0.68 14 0.75 -0.15 0.65
ISAT 2143 11 0.77 -2.14 1.26 14 0.81 -1.96 1.38 14 0.80 -1.80 1.43
MASH 1484 12 0.70 0.04 0.60 12 0.71 -0.01 0.64 18 0.72 0.13 0.75
TIRILAZAD 3552 18 0.75 0.20 0.76 20 0.76 0.23 0.78 27 0.79 -0.13 0.72
HHU 60 - - -
D-SAT 439 28 0.79 0.24 1.00 27 0.79 0.15 0.97 25 0.78 0.45 1.04
SHOP 1500 42 0.86 0.20 1.41 34 0.83 0.04 1.18 41 0.84 0.11 1.15
Unfavorable
C-1 433 15 0.70 0.41 0.76 17 0.72 0.53 0.82 15 0.71 0.48 0.79
IHAST 998 16 0.74 -0.30 1.23 17 0.74 -0.23 1.16 16 0.74 -0.13 1.19
IMASH 327 31 0.78 0.28 1.18 33 0.79 0.30 1.28 34 0.80 0.31 1.27
ISAT 2143 11 0.67 0.48 0.80 11 0.67 0.55 0.79 12 0.68 0.62 0.68
MASH 1484 25 0.75 0.25 0.79 24 0.75 0.16 0.84 28 0.76 0.17 0.90
TIRILAZAD 3552 27 0.78 0.01 0.85 28 0.78 0.08 0.89 31 0.79 -0.04 0.86
HHU 60 - - -
D-SAT 439 37 0.82 -0.14 1.03 38 0.82 -0.17 1.06 37 0.82 0.00 1.11
SHOP 1500 46 0.85 -0.09 1.20 40 0.83 -0.06 1.06 42 0.84 -0.13 1.04
CIL: Calibration in the large, numbers closer to zero indicates better calibration. Slope: Recalibration slope, numbers closer to 1 indicates better calibration.
Cross validation procedure was not implemented for HHU data because we considered the data insufficiently powered to reliably estimate performance indices.
116
Figure 4.2 – Cross validation plots in CONSCIOUS I cohorts
117
Figure 4.3 – Cross validation plots in Tirilazad cohorts
118
Figure 4.4 – Cross validation plots in IHAST cohorts
119
Figure 4.5 – Cross validation plots in D-SAT cohorts
120
Figure 4.6 – Cross validation plots in SHOP cohorts
121
Figure 4.7 – Plots of predicted probabilities of mortality by sum score with 95% confidence
intervals based on the core model
Coding Age (years)
≤ 39 0
40 – 49 1
50 – 59 2
60 – 69 3
≥ 70 10
Premorbid
hypertension:
No
0
Yes 2
WFNS =1 0
WFNS=2 6
WFNS=3 10
WFNS=4 13
WFNS=5 20
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
0 5 10 15 20 25 30 35
Pre
dic
ted
pro
ba
bility o
f 3
-mo
nth
mo
rta
lity
Sum score
SAHIT score plot
122
Table 4.5 – Model coefficients to obtain linear predictors for computing outcome probabilities
Predictor Mortality Unfavorable outcome
Core Neuro Repair Core Neuro Repair
Intercept -4.918 -5.475 -5.350 -3.703 -4.175 -4.122
Age 0.032 0.030 0.027 0.034 0.032 0.031
Hypertension 0.327 0.346 0.344 0.268 0.277 0.273
WFNS 1 - - - - - -
2 0.707 0.676 0.687 0.688 0.602 0.598
3 1.393 1.352 1.273 1.448 1.360 1.321
4 1.803 1.699 1.669 1.723 1.600 1.580
5 2.786 2.578 2.404 2.565 2.399 2.300
Fisher 1 - - - - - -
2 - -0.008 0.072 - 0.310 0.349
3 - 0.470 0.497 - 0.729 0.750
4 - 0.323 0.487 - 0.854 0.931
Location:ACA - - - - -
ICA - 0.220 0.222 - -0.105 -0.109
MCA - -0.100 -0.027 - -0.266 -0.247
PCQ - 0.473 0.318 - 0.032 -0.033
Size:small - - - - -
Large - 0.658 0.481 - 0.222 0.136
Giant - 1.178 0.370 - 0.529 0.131
Repair:clip - - - - -
Coil - - -0.390 - - -0.177
None - - 1.543 - - 0.842
123
Chapter 5
Investigating socioeconomic status and race/ethnicity as latent prognostic
factors in aneurysmal subarachnoid hemorrhage
This chapter is adapted with permission from the following publications:
Jaja BN, Saposnik G, Nisenbaum R, Schweizer TA, Reddy D, Thorpe KE, Macdonald RL.
Effect of socioeconomic status on inpatient mortality and use of postacute care after
subarachnoid hemorrhage. Stroke. 2013;44(10):2842-7.
Jaja BN, Saposnik G, Nisenbaum R, Lo BW, Schweizer TA, Thorpe KE, Macdonald RL.
Racial/ethnic differences in inpatient mortality and use of institutional postacute care following
subarachnoid hemorrhage. Journal of Neurosurgery. 2013;119(6):1627-32
124
5.1 Introduction
Known prognostic factors in SAH explain only a small proportion of the variance in
outcome. Thus, there is a need to further explore other possible determinants and associative
factors for poor outcome in patients with SAH. This chapter investigates whether differences in
outcomes of patients with SAH is related to socioeconomic status on one hand and race/ethnicity
on the other hand using patients information in 2 nationally representative administrative
databases obtained from 2 countries with divergent health care systems. Improved understanding
of the prognostic effect of socioeconomic status and race/ethnicity could provide new
perspectives on the pathobiology of the condition and contribute to improved outcomes for all
populations.
5.2 Part A: Socioeconomic status and inpatient mortality risk after SAH
About a quarter of SAH patients will die within two weeks of hospital admission and a
similar proportion of survivors are discharged with functional disabilities that may require
prolonged institutional care for recovery to premorbid life style.8 Several studies have shown that
socioeconomic status (SES) has a significant impact on cerebrovascular and cardiovascular
disease outcomes.120, 121, 233
This effect appears to be irrespective of health care system or SES
indicator.234-237
Research in the United States, Canada and other countries has demonstrated
higher mortality and poorer chances of survival after ischemic stroke and acute myocardial
infarction among people living in less affluent neighborhoods, or possessing lower levels of
education, occupation or income compared with those higher in the socioeconomic hierarchy.233,
235-240 There is also evidence that socioeconomic circumstances may influence the utilization of
institutional post-acute care (iPAC) among stroke survivors in the US241
and in Canada.239
Nevertheless, not all studies have reported an association between SES and mortality after
125
stroke.242
And, to a large extent, researchers are uncertain as to what factors mediate the effect of
SES on disease outcomes.120, 121
While a substantial body of literature has accumulated for other acute life threatening
cerebrovascular and cardiovascular diseases, with at least 2 recent reviews published to
synthesize the current evidence on the association of SES and stroke,120, 121
very little or nothing
is known about the relation between SES and SAH. A literature search found only two small
studies that partially addressed this topic in SAH using data from over 10 years ago.106, 122
The goal of Part A of this chapter is to provide a better understanding of the relationship between
SES and SAH outcomes. The primary objective was to investigate the hypothesis that SES is
associated with inpatient mortality risk following hospital admission for SAH. The secondary
objectives were to investigate the hypothesis that SES is related to use of institutional post-acute
care among SAH survivors; and that any relationship between SES and the outcomes studied is
independent of the health care system under which patients received treatment.
5.2.1 Methods
The study cohort consisted of SAH patients with records in 2 nationally representative
datasets: the Canadian Discharge Abstract Database (DAD) and the US Nationwide Inpatient
Sample (NIS). Information about patients’ SES was determined on the basis of estimated median
household income of residents for patient’s ZIP or postal code. This information is available in
the NIS and DAD data sets but computed and categorized somewhat differently in the 2 data
sets. In the NIS, neighborhood income status was computed relative to the 2000 distribution of
the US population, with annual corrections to account for inflation and change in income
distribution.243
In the DAD, neighborhood income status was computed relative to the 2006
distribution of Canadian population.244
The NIS applied quartile cut points with quartile 1
126
representing the lowest income neighborhoods, and quartile 4 representing the wealthiest
neighborhoods. The DAD applied quintile cut points with quintile 1 representing the lowest
income neighborhoods, and quintile 5 representing the wealthiest neighborhoods. Both processes
of determining SES have been validated and used in previous studies.239, 245
Patient discharge disposition was categorized into a 3-level categorical outcome variable
that was similar for both data sets: (1) Routine discharge, comprising discharged home or alive,
destination unknown or signed against medical advice; (2) In-hospital mortality; (3) Discharge to
institutional care comprising transfer to short-term hospital, home healthcare, other transfers,
including skilled nursing facility, intermediate care, and another type of facility. Potential
confounding variables that were accounted for in the analysis included the following: (1)
Demographic covariates such as age, sex, race (categorized in the NIS as white, black, Hispanic,
Asian/pacific Islander, Native American or Others), and insurance status (Medicare, Medicaid,
Private, including Health Maintenance Organization, self-pay, or no charges). Information on
race was not provided in the DAD, and insurance status was not relevant in the Canadian context
because healthcare services in Canada are publicly funded. (2) Clinical covariates such as
admission type (elective versus urgent/emergency) and modified Charlson–Deyo comorbid index
score, a measure of the number and severity of patients’ comorbid illness.246
(3) Hospital
covariates such as geographic region of hospital location (Northeast, Midwest, South, or West)
as in the NIS; bed size (small, medium, or large), and teaching status (teaching or nonteaching
hospital). The hospital covariate provided in the DAD was hospital status, which was categorized
into small, medium, or large community hospitals or teaching hospitals on the basis of bed size
and academic status.
127
There were no missing data in the DAD. The proportion of missing data in the NIS was
<0.5% for each variable, except for median neighborhood income quartile (2.7% missing) and
race (23.8% missing) for which multiple imputations were performed with the imputation model
specified on the primary predictor variable, outcome variable, and explanatory covariates as well
as hospital and discharge weights provided by the NIS for variance estimation.
Considering that the NIS and DAD data sets differ in sampling and coding designs, the 2
data sets were analysed separately and comparison made at the level of aggregate results.
Descriptive statistics were computed to provide information on patient demographic, clinical,
and hospital characteristics as well as crude outcomes according to neighborhood income status.
Trends across categorical data were tested with a Mantel–Haenszel χ2 test and across continuous
variables by analysis of variance (ANOVA). Multinomial logistic regression models were
thereafter fitted to examine the association between neighborhood income status and inhospital
mortality or discharge to iPAC, with routine discharge as base outcome. Where a significant
SES-outcome association was present, sequential adjustment for demographic, clinical, and
hospital factors was performed to assess whether these factors had a mediation effect on the
association. In addition, tests of overall interaction effects were performed, including interaction
of neighborhood income status and sex, neighborhood income status and race, and neighborhood
income status and insurance status. Plots of predicted probabilities were obtained to visually
examine how any observed income–mortality associations changed with age. Given the single-
stage stratified cluster sampling design of the NIS, discharge and hospital weights provided for
variance estimation were applied to obtain robust confidence intervals. Sensitivity analysis was
performed to examine the impact of imputing data, which was found to be satisfactory.
128
5.2.2 Results
The study cohort consisted of 31 631 US patients and 16 531 Canadian patients. Baseline
characteristics according to neighborhood income levels are shown in Table 5.1 for US patients
and in Table 5.2 for Canadian patients. Average age of patients in both countries was similar, at
58 years. Average age or sex did not differ by neighborhood income status in the United States
or Canada. US patients living in the lowest income neighborhoods were less likely than those in
wealthy neighborhoods to be white (52% versus 70%) or Asian/Pacific Islanders (1.95% versus
9.0%) and more likely to be black (25% versus 7%) or Hispanic (17% versus 9%), P<0.001.
Low-income patients were more likely to be seen in hospitals in the south region (54% versus
24%; P<0.001), to be on Medicaid or opt for self-pay, or have greater comorbid burden. US
patients were more likely to receive urgent/emergency admission than Canadian patients (91%
versus 83%) and presented with greater comorbidity compared with their Canadian counterparts
(40% versus 25%). However, crude mortality rate among US patients (22%) was similar to that
of Canadian patients (21%). The percentage of patients who experienced routine discharge was
higher in Canada (55%) compared with the United States (36%) where the proportion differed by
SES (lowest income neighborhood, 35% versus highest income neighborhood, 38%).
Multivariable analysis of US data revealed a significant income–mortality association
(P=0.001) with patients with SAH in the highest neighborhood income quartile at significantly
lower risk of mortality than patients in the lowest income quartile; OR, 0.69, 95% C.I: 0.58 –
0.82; Table 5.3). This income–mortality association was not attenuated by sequential adjustment
for demographic covariates (model 2: age, sex, race and insurance status), clinical covariates
(model 3: model 2 + admission type and comorbid index score), and hospital covariates (model
4: model 3 + hospital region, bed size, teaching status). Canadian patients in the highest income
quintile experienced a marginal reduction in the risk of in-hospital mortality compared with
129
those in the lowest income quintile; OR, 0.97, 95% C.I: 0.85 – 1.12; Table 5.4). The risk
reduction, however, was not statistically significant (p = 0.51). Use of iPAC did not differ
significantly by income status in the US or Canada (Tables 5.3 and 5.4). Among US patients, the
effects of income status on inpatient mortality or discharge to iPAC varied with insurance status
(interaction; p = 0.001 for both outcomes) but not with sex (mortality, p = 0.401; institutional
outcome, p =0.414) or race (mortality, p = 0.464; institutional outcome, p = 0.184). Plots of
predicted probabilities indicated that the income–mortality association was present across the age
spectrum, and that it slightly widened with increasing age (Figure 5.1).
5.2.3 Discussion
This exploratory analysis used nationally representative data sets with comparable
variables and covering recent similar time periods to investigate association between SES and
inpatient mortality after SAH in 2 countries operating different health care systems. SES was
significantly associated with inpatient mortality in the United States, which operates a privately
funded health care system, with patients with SAH living in wealthy neighborhoods experiencing
a modest reduction in risk of mortality compared with patients living in low-income
neighborhoods. The magnitude of the income–mortality association was not influenced by non-
modifiable risk factors of age, sex, race, and comorbidity or hospital status. Unlike in the United
States, no significant effect was demonstrated in Canada which operates a publicly funded health
care system. The results agree somewhat with those of previous studies in SAH that reported
higher mortality rates among lower SES groups compared with higher SES groups.106, 122
The
Finnish contribution to the World Health Organization Multinational Monitoring of trends and
determinants of Cardiovascular disease (FINMONICA) study used personal income as a proxy
for SES and investigated 956 SAH cases for socioeconomic differences in case fatality at 7 days,
130
28 days, and 1 year post admission across different age groups and sex.122
The study reported
significant income–mortality association in only young adult males of ages 25 to 44 years. The
present study however found that income–mortality association tends to widen with increasing
age in the United States. The FINMONICA study might have been underpowered to detect
significant associations at older age, considering the small number of cases studied. A study of
annual SAH mortality across ethnic/ racial groups by household income in Los Angeles found an
inverse income–mortality association among minority populations only.106
The present more
inclusive analysis did not indicate that ethnicity/race significantly alters the effect of SES on
inpatient mortality.
Much of the income–mortality gradient seen among US patients was not explained by
demographic, comorbid, and hospital factors, which suggests that these factors are not primary
mediators in the link between SES and SAH mortality. However, the analyses showed that
among US patients the effect of SES cannot be interpreted independent of patient insurance
status. Insurance status has been shown to significantly impact postoperative outcomes in
neurosurgical patients in the United States with worse outcomes more likely to occur among
patients who are inadequately insured.247
Better insured patients are more likely to live in
wealthy neighborhoods and have better access to timely, high quality specialized care, 235
which
has been suggested to be important to improved outcomes after SAH.57
The present study provides some evidence in support of the concept that access to care
inclusive of prevention and management of comorbid conditions for lower SES groups is
relatively better in a public healthcare system than in a private healthcare system.248, 249
Patients
with SAH in the United States presented with greater comorbid burden compared with Canadian
patients. The effect of SES was more pronounced in the United States than in Canada; a finding
131
that is consistent with previous comparative studies that demonstrated stronger link between SES
and overall population health117, 118
and disease-specific outcomes119
in the United States
compared with Canada. It is possible that Canada’s more inclusive publicly funded health
insurance coverage facilitated a relatively better access to treatment for comorbid conditions and
improved chances of better outcomes after SAH among lower SES groups. It is also possible that
the analysis potentially underestimated the effect of SES in Canada. For instance, when median
household income is used as SES indicator, the choice of census geographic unit and number of
subgroups has been identified as potential confounders of the strength of income–mortality
associations.250
Furthermore, because this community-level SES indicator was computed relative
to a fixed time point in the Canadian data set, it might have inadequately accounted for any
widening income disparities with time, hence less reflective of socioeconomic inequalities in
Canada.
That the use of iPAC was not influenced by SES among SAH survivors in the US or
Canada is not unexpected, considering the need for institutional care could be influenced by
multiple factors, including the severity of post-acute residual disability, availability of
rehabilitation services, differences in referral patterns and sociocultural behaviors related to
family preferences and support. In contrast, some studies in ischemic stroke have reported
differing patterns of discharge to and use of iPAC by insurance status or personal income in the
US241, 251
or by neighborhood income status in Canada.239
Of note are the multiple potential limitations of this study. It is cross sectional and
focused on the inpatient course; hence, the results are not generalizable to patients with SAH
who did not survive to hospital admission or who died after discharge from acute care. Inpatient
mortality and discharge disposition are rather crude outcome measures, though validated.
132
However, it is plausible that more nuanced outcomes would have demonstrated stronger effect of
SES in the US than in Canada.19
The limitations of using large administrative data sets for health
services research are well recognized in the literature.252, 253
Of particular relevance to this study
is the inability to adjust for case severity. However, though low SES groups have been shown to
present with more severe stroke in comparison with higher SES groups,254
there is inconsistent
evidence in support of a causal effect of stroke severity on income–mortality associations after
ischemic stroke.237, 238
The effects of SES on SAH severity, if any, and on mortality therefore
could be causally unrelated. Another limitation relates to the known disadvantages of using
ecological measures of SES.255
Finally, subtle differences in the way variables were defined and
categorized between the 2 databases, and adjusted for in the analyses requires that the results be
viewed as indicating overall trends rather than an estimation of the magnitude of SES–mortality
associations after SAH in both countries.
In summary, SES is associated with inpatient mortality risk after admission for SAH. The
extent of the association could be related to the health care system under which treatment was
provided.
133
Table 5.1 – Baseline characteristics of US patients by quartile of median household income
Q1 (Lowest)
n=8349
Q2
n=7999
Q3
n=7348
Q4 (Highest)
n=7095
Total
N=30791
p-value
Age 56.82±16.81 58.02±16.75 57.67±16.81 58.89±17.05 57.81±16.84 0.458
Sex 0.460
Female 5135(62.0) 5006(63.0) 4535(62.0) 4343(62.0) 19019(62.0)
Race <0.001
Whites 3336(52.2) 4034(67.7) 3752(68.3) 3886(69.5) 15008(64.1)
Blacks 1623(25.4) 716(12.0) 556(10.1) 407(7.3) 3302(14.1)
Hispanics 1077(16.8) 743(12.5) 690(12.6) 527(9.4) 3037(12.9)
Asian/PI 125(1.95) 189(3.2) 257(4.7) 503(9.0) 1074(4.6)
Native A. 43(0.67) 24(0.4) 16(0.3) 13(0.2) 96(0.4)
Others 191(2.9) 247(4.2) 223(4.0) 254(4.5) 915(3.9)
Payer St. <0.001
Medicare 2839(34.1) 2762(34.6) 2357(32.2) 2369(32.2) 10327(33.6)
Medicaid 1503(18.1) 989(12.4) 707(9.7) 419(5.9) 3618(11.7)
Priv/HMO 2473(29.7) 2999(37.6) 3339(45.6) 3632(51.2) 12443(40.5)
Self-pay 1039(12.5) 828(10.4) 569(7.8) 370(5.2) 2806(9.1)
No charge 95(1.1) 61(0.8) 49(0.7) 39(0.6) 244(0.8)
Others 374(4.5) 335(4.2) 304(4.1) 253(3.5) 1266(4.2)
Adm. type 0.01
Emerg./urgent 7564(91.0) 7243(91.0) 6577(90.0) 6347(90.0) 27731(91.0)
Elective 706(9.0) 700(9.0) 711(10.0) 682(10.0) 2799(9.0)
Bed size <0.001
Small 302(3.7) 347(4.4) 414(5.7) 487(6.9) 1550(5.1)
Medium 1678(20.4) 1465(18.5) 1085(14.9) 1073(15.1) 5301(17.4)
Large 6250(75.9) 6084(77.1) 5791(79.4) 5522(78.0) 23647(77.5)
Teaching status <0.001
Nonteaching 1868(23.0) 1996(25.0) 1860(25.0) 1763(25.0) 7487(25.0)
Teaching 6362(77.0) 5900(75.0) 5430(75.0) 5319(75.0) 23011(75.0)
Region <0.001
North East 1035(12.4) 1119(14.0) 1270(17.3) 1834(25.8) 5258(17.1)
Midwest 1453(17.4) 1964(24.5) 1689(23.0) 1294(18.2) 6400(20.8)
South 4479(53.7) 3094(38.7) 2177(29.6) 1699(24.0) 11451(37.2)
West 1382(16.5) 1820(22.8) 2212(30.1) 2268(32.0) 7682(24.9)
Charlson <0.001
1 4761(57.0) 4722(59.0) 4385(60.0) 4471(63.0) 18339(60.0)
≥2 3588(43.0) 3277(41.0) 2963(40.0) 2624(37.0) 12452(40.0)
Discharge status
Died in
Hospital
1880(22.5) 1841(23.0) 1541(21.0) 1438(20.3) 6700(21.8) <0.001
Institutional
care
3557(42.6) 3383(42.3) 3046(41.5) 3004(42.4) 12990(42.2)
Routine 2905(34.8) 2766(34.6) 2758(37.5) 2650(37.3) 11079(36.0)
Charlson: Charlson-Deyo Comorbidity index score. The total number of patients (30 791) includes those for whom
data are completely available for race/ethnicity. The dominators for each factor vary because of missing data.
Number in parenthesis represents percentages.
134
Table 5.2 – Baseline characteristics of Canadian patients by quintile of median household
income
Q1 (Lowest)
N=13723
Q2
N=3532
Q3
N=3274
Q4
N=3117
Q5 (Highest)
N=2885
Total
N=16531
p-value
Age 57.79±15.69 57.68±15.87 58.04±16.05 58.18±15.52 57.53±16.02 57.85±16.02 0.28
Sex 0.898
Female 2277(61.0) 2144(61.0) 2007(62.0) 1918(62.0) 1743(60.0) 10089(61.0)
Adm. type 0.637
Emerg./urgent 3112(84.0) 2923(83.0) 2707(83.0) 2562(82.0) 2383(83.0) 13687(83.0)
Elective 611(16.0) 609(17.0) 567(17.0) 555(18.0) 502(17.0) 2844(17.0)
Hospital status 0.681
Small CH 78(2.1) 80(2.3) 63(1.9) 53(1.7) 57(1.9) 331(2.0)
Medium CH 158(4.2) 147(4.1) 141(4.3) 143(4.6) 139(4.8) 728(4.4)
Large CH 954(25.6) 854(24.1) 820(25.0) 764(24.5) 740(25.6) 4132(25.1)
Teaching 2533(68.0) 2658(69.0) 2250(69.0) 2156(69.0) 1949(67.6) 11338(68.6)
Charlson 0.379
1 2831(76.0) 2658(75.0) 2466(75.0) 2322(75.0) 2181(76.0) 12458(75.0)
≥2 892(24.0) 874(25.0) 808(25.0) 795(25.0) 704(25.0) 4073(25.0)
Discharge status 0.278
Died in Hospital 771(21.0) 770(21.8) 708(21.6) 649(20.8) 569(20.0) 3467(21.0)
Institutional care 904(24.0) 810(23.0) 820(25.0) 754(24.0) 722(25.0) 4010(24.0)
Routine 2048(55.0) 1952(55.0) 1746(53.0) 1714(55.0) 1594(55.0) 9054(55.0)
CH: Community hospital. Number in parenthesis represents percentages.
Table 5.3 – Relation of neighborhood income to in-hospital mortality and discharge to iPAC for
US patients
Model 1
Unadjusted
Model 2
(model 1 + demographic)
Model 3
(model 2 + clinical)
Model 4
(model 3 + Hospital)
In-hospital mortality
Lowest Q1 referent
Q2 0.93 (0.80 – 1.09) 0.97 (0.82 – 1.13) 0.97 (0.83 – 1.13) 0.96 (0.82 – 1.13)
Q3 0.80 (0.68 – 0.94)* 0.89 (0.75 – 1.04) 0.89 (0.75 – 1.05) 0.88 (0.75 – 1.04)
Highest Q4 0.69 (0.58 – 0.82)* 0.77 (0.64 – 0.91)* 0.78 (0.65 – 0.93)* 0.77 (0.65 – 0.93)*
Institutional outcome
Lowest Q1 referent
Q2 0.97 (0.84 – 1.11) 1.02 (0.88 – 1.19) 1.03 (0.89 – 1.20) 1.02 (0.88 – 1.18)
Q3 0.92 (0.80 – 1.05) 1.02 (0.88 – 1.18) 1.02 (0.88 – 1.19) 1.00 (0.86 – 1.17)
Highest Q4 0.93 (0.85 – 1.04) 0.97 (0.78 – 1.20) 1.00 (0.80 – 1.24) 0.98 (0.79 – 1.20)
Institutional outcome: Discharged to institutional post-acute care facility
Values are odd ratios and 95% confidence intervals from multinomial logistic regressions
Demographic covariates: age, sex, race and insurance status
Clinical covariates: admission type and Comorbid index score
Hospital covariates: Hospital region, bed size, teaching status
*significant at p<0.05
135
Table 5.4 – Relation of neighborhood income status to in-hospital mortality and discharge to
iPAC for Canadian patients
Mortality Institutional outcome
Income Quintile
Lowest Q1 referent
Q2 1.05 (0.93 – 1.20) 0.93 (0.83 – 1.05)
Q3 1.10 (0.96 – 1.25) 1.07 (0.95 – 1.21)
Q4 1.03 (0.90 – 1.18) 1.03 (0.91 – 1.16)
Highest Q5 0.97 (0.85 – 1.12) 1.07 (0.95 – 1.21)
Values are odd ratios and 95% confidence intervals from multinomial logistic regressions
Adjusted for age, sex, comorbid index score, hospital status
Figure 5.1 – Plots of predicted probabilities of mortality showing the effect of neighborhood
income status in the US increases with advancing age
.1.1
5.2
.25
.3.3
5
Pro
bab
ility
of in
-ho
spital m
ort
alit
y
20 25 30 35 40 45 50 55 60 65 70 75 80Age in years at admission
Lowest Quartile Low medium Quartile
High medium Quartile Highest Quartile
136
5.3 Part B: Race/ethnicity and inpatient mortality risk after SAH
The mortality rates of SAH differ among race/ethnic groups, with many reports
indicating higher mortality rates in non-white populations compared with white populations in
industrialized nations. 100, 102-109
However, it is uncertain whether differences in mortality rates
are only due to differences in the prevalence of SAH among race/ethnic groups or are also
related to differences in case fatality. A few studies have examined differences in outcomes
among racial/ethnic groups, while simultaneously accounting for factors that could potentially
confound the race/ethnicity association with mortality114
. These studies have done so using
patients participating in trials114
or patients in a small geographic cohort.115, 116
and reported
varying results. A post hoc analysis predominantly comparing African-Americans to Caucasians
who were recruited into a large clinical trial between 1991 and 1997 found no significant
race/ethnic differences in 3-month functional outcome.114
Another study using the New York
database of hospital discharges in 2003 to investigate the relationship between race/ethnicity and
SAH outcome reported that white patients in New York with SAH had better functional
outcomes than non-white patients.115
It is presently unknown whether the use of institutional care
among survivors of SAH differs across racial/ethnic groups. Hence, part B of this chapter
investigated the hypothesis that patients admitted into hospital for SAH differ in their risk of
inpatient mortality when compared by their self-reported racial/ethnic groups. It secondarily
examined whether racial/ethnic differences are present in the use of institutional post-acute care
(iPAC) following hospitalization for SAH.
5.3.1 Methods
The study was based on SAH patient data in the United States Nationwide Inpatient
Sample (NIS). The independent variable was patient self-identified race/ethnicity categorized in
137
the NIS as white, black, Asian/Pacific islander (API), Native American or others. A number of
covariates that could help explain or confound the effect of race/ethnicity on the outcomes of
interest were accounted for in the analysis. These were patient age, sex, neighborhood median
income per patient ZIP code expressed in quartiles (a measure of socioeconomic status at the
community level), insurance status (categorized as Medicare, Medicaid, Private including HMO,
Self-pay or no charges), admission type (elective, urgent or emergency), hospital region
(Northeast, Midwest, South or West), bed size (small, medium or large), and hospital teaching
status (nonteaching or teaching). In order to account for the effect of comorbid conditions, each
patient’s modified Charlson-Deyo comorbidity index score (CCI) was computed and included in
the analysis. CCI was categorized into 4 groups including CCI of 1, 2, 3, or ≥4. The primary
outcome was in-hospital mortality whereas the secondary outcome was use of iPAC, defined as
transfer to short-term hospital, other transfer including skilled nursing facility, intermediate care,
and another type of facility, and home health care. The proportion of missing data was less than
0.5% for most variables except for admission type (23.8%), neighborhood median income
quartile (2.4%) and race (23.9%). Multiple imputations were performed to fill in plausible values
for all missing data as described in study methodology (Chapter 2).
Patient demographic, clinical and hospital characteristics were summarized by
race/ethnicity using descriptive statistics with continuous variables expressed as means with
standard deviations and categorical variables expressed as frequencies (percentage). Differences
among race/ethnic groups for categorical variables were tested with a Mantel Haenszel chi-
square test, and for continuous variables differences were tested by analysis of variance
(ANOVA). The associations between race/ethnicity and patient discharge disposition were
examined by fitting multinomial logistic regression models, adjusting for the aforementioned
138
explanatory variables. All analyses accounted for the single-stage stratified cluster sampling
design of the NIS by including discharge and hospital weights provided in the NIS for accurate
variance estimation. This also allowed for extrapolation of results to the entire US population.
The level of statistical significance was set at p ≤ 0.05. All analyses were repeated using
complete cases only and gave similar results.
5.3.2 Results
During the study time period, 31,631 discharges were recorded in the NIS for SAH.
Table 5.5 shows baseline characteristics of the study cohort by race/ethnicity. Whites and API
patients were significantly older than patients of other ethnic groups (p ≤0.001). Patients in the
Hispanic group were less likely to be females compared with other groups. Black patients were
more likely to live in low income neighborhoods, whereas API patients were more likely to live
in high income neighborhoods. The crude in-hospital mortality rate was 22%, and the proportion
of patients discharged to institutional care was 42%.
In multivariable analyses, race/ethnicity was a significant predictor of in-hospital
mortality (p=0.003) and of discharge to institutional care (p ≤0.001). Black patients were similar
to white patients in the risk of in-hospital mortality (OR, 1.04; 95% CI: 0.93-1.16), but were
more likely to be discharged to post-acute institutional care than white patients (OR, 1.27; 95%
CI: 1.14-1.40) (Table 5.6). Hispanic patients were at significantly lower risk of in-hospital
mortality than were white patients (OR, 0.84; 95% CI: 0.72-0.97); however, both groups were
statistically similar in likelihood of discharge to institutional care (OR, 0.97; 95% CI: 0.86-1.09).
Compared with white patients, API patients were at higher risk of death (OR, 1.34; 95% CI:
1.13-1.59) and in greater need for institutional care (OR, 1.17; 95% CI: 0.99-1.37). Similarly,
Native American patients were at higher risk of death (OR, 1.10; 95% CI: 0.90-1.34) and
139
institutional care (OR, 1.16; 95% CI: 0.98-1.38), although the differences were not statistically
significant (Table 5.6). Figure 5.2 shows a plot of race/ethnic differences in the risk of in-
hospital mortality by age, expressed as predicted probabilities. Hispanic patients were the least
likely to die during hospitalization whereas patients of API nativity were the most likely to die
during hospitalization for SAH.
5.3.3 Discussion
Among patients with SAH admitted to acute care hospitals in the US, race/ethnicity was
found to be a significant predictor of inpatient mortality and discharge to institutional care.
Racial/ethnic differences were present after accounting for age, sex, SES and insurance status,
and other clinical and hospital characteristics that could potentially influence outcomes. The
results of this study agree to some extent with those of a study of the New York Statewide
Planning and Research Cooperative System (SPARC) database that suggested that white patients
with SAH had better functional outcomes than non-white patients.115
Our findings, however,
differ from those of other hospital-102, 109, 114
and population-based studies116
256
in the US102, 109,
114, 116and other countries
256 that demonstrated no racial/ethnic differences in case fatality rate,
102,
109 functional disability,
114 or other outcomes.
116, 256 The small number of events observed in
previous studies,109
especially with respect to groups other than “Whites” or “Blacks,”114, 256
may
well have contributed to the negative findings of these studies. It is also plausible that differences
in the categorization of race/ethnicity in the different studies contributed to the inconsistency
between the findings of the present study and those of prior studies.
The present study showed that black and white patients had a similar risk of death.
Previous studies also showed that, after adjustment for important determinants of outcome, these
groups do not differ significantly in survival rates, case fatality rate,102
time to treatment,116
or 3-
140
month functional outcome.114
Taken together, these findings indicate that black and non-
Hispanic white patients have a comparable short-term prognosis after SAH. Although in the
present study black patients were more likely to be discharged to other facilities rather than to
home, whether this result indicates greater residual disability in black patients or is due to the
presence of other factors related to referral to or receipt of iPAC can only be a matter of
speculation.
The finding of significantly lower risk of in-hospital mortality in patients of Hispanic
ethnicity compared with non-Hispanic white patients agrees with a growing body of
epidemiological literature in the US; this literature shows that Hispanic ethnicity is associated
with lower risk of all-cause and disease-specific death.257-261
The estimated relative risk
reduction of 16% in this study is comparable to the 7%–13% reported for stroke patients of
Hispanic ethnicity in the population-based Brain Attack Surveillance in the Corpus Christi
Project in Texas.257
In the Northern Manhattan Study (NOMAS),258
Hispanic ethnicity was
associated with much greater risk reduction for coronary death (64%) and vascular death (34%).
Hispanic populations in the US on average have a lower socioeconomic status, greater burden of
risk factors, lower insurance coverage, and lower access to the processes of care than do non-
Hispanic white patients.262
By experiencing more favorable mortality outcomes than non-
Hispanic white patients, Hispanic patients present an “epidemiologic paradox,” the cause of
which, though the subject of ongoing investigations, remains uncertain.258, 261
Of the racial/ethnic groups compared in this study, API patients had the highest risk of
death and need for post-acute care. A somewhat different result was reported by a similar study
that used only the 1997 NIS data and found that API patients were similar to non- Hispanic white
patients in the risk of in-hospital mortality but had more need for institutional care than did non-
141
Hispanic white patients.109
However, among patients who received thrombolysis for acute
ischemic stroke in the US, API patients had the highest risk of death and intracerebral
hemorrhage.263, 264
The prospective, population-based Auckland Regional Community Stroke III
study in New Zealand also found that API patients had the worst 6-month functional outcomes
after acute ischemic stroke compared with other racial/ethnic groups.265
A number of reasons may explain the worse prognosis in API patients. It is possible these
patients have more severe SAH. Secondly, some studies have reported that API patients have a
relatively low likelihood of admission to high quality hospitals after stroke, compared with other
racial/ethnic groups in the US.266
Timely, quality, specialized care has been suggested to be
important for improved outcomes after SAH.57
Furthermore, researchers have reported cultural,
language, and other barriers among API that could have an impact on their capacity to access
health care and experience favorable health outcomes.267
Nevertheless, these explanations fail to
account for the fact that API patients in the US constitute a heterogeneous group reflecting to a
larger extent than other racial/ethnic groups both extremes of socioeconomic status. For instance,
API patients in the present study lived in wealthier neighborhoods relative to other racial/ethnic
groups.
This study has a number of limitations that could impact its conclusions. The
categorization of race/ethnicity, as it is done in the NIS, is a social construct that is not based on
any unique genetic composition; hence, the results of this study may not necessarily be due to
genetic differences among studied groups. Second, because the NIS has no variable to measure
case severity, no adjustment was made for this important prognostic factor and hence it is
difficult to ascertain what effect adjustment for SAH severity would have had on the results of
this study.
142
In summary, patients with SAH who survived to receive treatment in hospitals differed
significantly in their risk of in-hospital mortality and use of institutional post-acute care when
compared according to their self-reported racial/ethnic groups. Hispanic patients had the best
outcomes and API patients had the worst outcomes during the inpatient course.
143
Table 5.5 – Baseline distribution of variables according to race/ethnicity status
White
15376
Black
3412
Hispanic
3133
Asian/PI
1098
NA/Others
1069
Total
24088
Age (mean ± sd) 60.27±16.57 53.00±15.42 52.46±17.15 59.80±17.79 54.85±17.19 57.96±16.9
Sex
Female 9489(62) 2227(65) 1822(58) 717(65) 634(59) 14899 (62)
Income Quartile
Lowest Q1 3336(22) 1623(49) 1077(36) 125(12) 234(23) 6395(27)
Q2 4034(27) 716(22) 743(25) 189(18) 271(27) 5953(25)
Q3 3752(25) 556(17) 690(23) 257(24) 239(24) 5494(24)
Highest Q4 3886(26) 407(12) 527(17) 503(49) 267(26) 5590(24)
Insurance status
Medicare 5935(39) 865(25) 666(21) 347(32) 290(27) 8103(34)
Medicaid 1220(8) 672(20) 770(25) 151(14) 185(17) 2998(12)
Priv/HMO 6437(42) 1199(35) 949(30) 458(42) 413(39) 9456(39)
Self-pay 1140(7) 443(13) 444(14) 90(8) 121(11) 2238(9)
No charge/
Others
599(4) 215(6) 301(10) 52(4) 56(5) 1223(5)
Adm. type
Emergent 10400(68) 2584(76) 2098(67) 744(68) 690((65) 16516 (69)
Urgent 3534(23) 557(16) 629(20) 205(19) 273(25) 5198(22)
Elective 1275(8) 239(7) 389(13) 147(13) 104(10) 2154(9)
Bed size
Small 854(6) 126(4) 179(6) 86(8) 60(6) 1305(5)
Medium 2856(19) 656(20) 406(13) 188(17) 158(15) 4264(18)
Large 11518(75) 2568(76) 2541(81) 820(75) 838(79) 18285(76)
Teaching Status
Nonteaching 4279(28) 601(18) 835(27) 285(26) 221(21) 6220(26)
Teaching 10950(72) 2749(82) 2291(73) 809(74) 835(79) 17634(74)
Region
North East 3494(23) 760(22) 438(14) 167(15) 287(27) 5146(21)
Midwest 2458(16) 451(13) 84(3) 37(3) 154(14) 3184(13)
South 5940(39) 1826(53) 1139(36) 164(15) 329(31) 9398(39)
West 3484(22) 375(11) 1472(47) 730(66) 299(28) 6360(26)
Charlson
1 9013(58) 1960(57) 1891(60) 646(59) 669(62) 14179(59)
2 3136(20) 692(20) 592(19) 225(20) 203(19) 4848(20)
3 1964(13) 409(12) 399(13) 139(13) 116(11) 3027(13)
≥4 1263(8) 351(10) 251(8) 88(8) 81(7) 2034(8)
Discharge status
In-hospital death 3530 (23) 686 (20) 575 (18) 294 (27) 231 (22) 5316(22)
Institutional care 6526 (42) 1528 (45) 1213 (39) 465 (42) 460 (43) 10192 (42)
Routine
discharge
5307 (35) 1193 (35) 1344 (43) 339 (31) 377 (35) 8560 (36)
NA: Native Americans. The total number of patients (24,088) includes those for whom data are completely available
for race/ethnicity. The dominators for each factor vary because of missing data. Number in parenthesis represents
percentages.
144
Table 5.6 – Results of multivariable analysis of the relation of race/ethnicity to in-hospital
mortality
In-Hospital mortality Institutional care
OR (95% C.I.) OR (95% C.I.)
Race: whites 1 1
Blacks 1.04 (0.93-1.16) 1.27 (1.14-1.40)
Hispanics 0.84 (0.72-0.97) 0.98 (0.87-1.09)
Asian/PIs 1.34 (1.13-1.59) 1.17 (0.99-1.37)
NA/Others 1.10 (0.90-1.34) 1.16 (0.98-1.38)
Age 1.04 (1.03-1.04) 1.03 (1.02-1.03)
Female 1.19 (1.11-1.27) 1.21 (1.14-1.28)
Median income: Q1 1 1
Q2 1.02 (0.93-1.12) 0.99 (0.90-1.08)
Q3 0.87 (0.79-0.97) 0.89 (0.81-0.98)
Q4 0.79 (0.70-0.90) 0.88 (0.74-1.06)
Payor: Medicare 1 1
Medicaid 0.95 (0.83-1.09) 0.83 (0.75-0.93)
Private/HMO 0.64 (0.58-0.70) 0.65 (0.60-0.71)
Self-pay 1.03 (0.90-1.19) 0.54 (0.47-0.62)
No charge/others 0.75 (0.62-0.90) 0.48 (0.41-0.56)
Admission: Emergency 1 1
Urgent 0.72 (0.65-0.82) 1.02 (0.91-1.14)
Elective 0.57 (0.49-0.67) 0.80 (0.70-0.91)
Bed size: Small 1 1
Medium 0.89 (0.67-1.18) 0.87 (0.69-1.09)
Large 0.72 (0.55-0.94) 0.72 (0.58-0.89)
Teaching Hospital: No 1 1
Yes 0.60 (0.53-0.67) 0.64 (0.56-0.72)
Region: Northeast 1 1
Midwest 0.76 (0.64-0.91) 0.85 (0.72-0.99)
South 0.72 (0.61-0.83) 0.63 (0.52-0.78)
West 0.78 (0.65-0.94) 0.72 (0.61-0.86)
Charlson : 1 1 1
2 1.21 (1.12-1.31) 1.31 (1.21-1.41)
3 1.90 (1.67-2.17) 2.84 (2.51-3.21)
4 2.23 (1.93-2.58) 2.59 (2.26-2.97)
Estimates are multinomial odds ratios with 95% confidence intervals
145
Figure 5.2 – Plots of race/ethnicity differences in risk of mortality by age, expressed as predicted
probabilities (y axis)
The uppermost curve represents API patients and the lowermost represent Hispanic patients; the other racial/ethnic
groups are clustered in the middle
.1.2
.3.4
Pre
dic
ted
pro
bab
ility
of d
ea
th
20 25 30 35 40 45 50 55 60 65 70 75 80Age in years at admission
Whites Blacks
Hispanics Asian/Pacific Islanders
Native Americans/Others
146
Chapter 6
General discussion and conclusion
147
6.1 Introduction
Aneurysmal subarachnoid hemorrhage is a complex, acute cerebrovascular condition of
variable clinical course. Though uncommon in the general population, the condition is associated
with poor prognosis. This dissertation reports on series of studies to provide a better
understanding of prognostic associations in SAH and to develop prediction models for
translating the knowledge into clinical applications and research use. The research commenced
with a systematic review to identify prognostic factors that are easily measured at hospital
admission and that are useful for early prediction of outcome following SAH. The review also
critically appraised available prognostic models in SAH with respect to methodological validity
and potential for predictive accuracy in different settings in order to guide further attempts at
model development. The results of the review informed the choice of 7 conventional prognostic
factors. Each of these factors was explored extensively in chapter 3 to provide more precise
insight into their prognostic relevance in SAH. A set of prediction models was then developed in
chapter 4 for predicting mortality and unfavorable outcome within 3 months of SAH. Finally,
attention was shifted in chapter 5 towards elucidating the role of socioeconomic status and
race/ethnicity as possible prognostic factors in SAH.
6.2 Strength of the research
The research presented in this thesis has 3 major strengths: (1) strong statistical power;
(2) greater representativeness of the study populations; and (3) conformity to high levels of
statistical standard. In planning the research, these three strengths were considered fundamental
148
to providing higher level evidence than is currently available in the literature on the nature of
prognostic associations in SAH. With respect to statistical power, the research used large,
prospectively collected clinical data on patients from a broad spectrum of settings. The size of
the data supported the extensive analyses that were performed and assured more precise
estimation of the magnitude of prognostic associations. With respect to representativeness, the
diverse case mix of the patient populations studied enabled more optimal determination of the
average effects of studied prognostic factors. With regard to statistical standards, the
heterogeneity and complexity of the datasets necessitated the adoption of analytical approaches
that are relatively novel in the context of prognostic studies in SAH. These include the use of
spline functions to investigate the shape of the association between continuous predictors and
outcome. Spline functions are used to more optimally explore potential change points in
prognostic effect while avoiding the common practice of arbitrarily categorizing predictor
variables using data driven threshold values. The relative advantage of individual participant data
(patient-level) meta-analysis over meta-analysis of aggregated data from published primary
studies was exploited in chapter 3 to summarize univariable associations with respect to studied
conventional prognostic factors. One advantage was the opportunity to standardize analyses
across multiple primary studies in the SAHIT repository, including identical missing data
analysis. Furthermore, the random effects model used to pool summary estimates derived from
primary studies was useful to synthesize univariable prognostic effects across studies where
patient populations may differ from each other in ways that could impact on prognostic
associations. Another advantage was the opportunity to perform multivariable risk adjustments
accounting for a similar set of adjustment factors, including differences in treatment modality.
The utilization of proportional odds analysis enabled us to take advantage of the ordinal nature of
149
the GOS in computing prognostic associations. Previous studies have commonly estimated
prognostic associations over the dichotomized GOS or modified Rankin score to assess the risk
of mortality83, 187
or unfavorable outcome.80-82, 268, 269
The analytical approaches adopted in this
research are well established and have been applied previously in other conditions to investigate
prognostic associations.212, 213
For a research of this scope, it is almost invariable that the
problematic issue of missing data will be encountered. Among the many methods available to
deal with this near-ubiquitous problem in clinical research, the technique of multiple imputations
is considered the least biased method.216
This advanced technique was adopted in a consistent
manner in all studies pertaining to the research thereby reducing bias from missing data to a
minimum.
6.3 Summary of research findings and contributions to knowledge
Much is known about prognostic associations and prognostic factors in SAH. However,
no meta-analysis has been reported to synthesize evidence on prognostic association as relates to
studied conventional prognostic factors. In a systematic review, we found that neurologic status
and age were consistently identified to be associated with outcome in most primary studies
reporting prognostic associations in SAH.124
We found that neurologic status is the strongest
predictor of outcome after SAH, which is consistent with opinion and findings in some prior
studies. In chapter 3, a series of studies were presented which, to some extent, confirmed what is
already known about the prognostic effect of the studied prognostic factors. However, the higher
precision of the analyses and wider generalizability of the results has elevated the level of
evidence in support of the effect of the studied factors. Additionally, some areas of uncertainty in
the literature were addressed. One of such is the change point in the prognostic effect of age.
Prognosis would be poorer in elderly patients with SAH compared with younger patients –
150
considering suboptimal response of the aging brain to injury, greater prevalence of extra-cerebral
comorbidities in the elderly and elevated risks of peri-procedural complications during aneurysm
repair in the elderly. In contrast to prior studies suggesting threshold values around which
prognosis worsens considerably with age, the present research found no clear evidence in support
of a specific threshold value in the prognostic effect of age. Therefore, we argued that the effect
of age on outcome is continuous and linear. Where the primary research focus is on the elderly, it
is probably more meaningful, from a prognostic standpoint, to define elderly subgroup of SAH
patients using the value of 65 years as prognosis seemed to become increasingly poorer around
this age, although no breakpoint was apparent. Another area of controversy pertains to the
prognostic role of premorbid hypertension. We found consistent evidence across studies in
support of an independent association between premorbid hypertension and outcome. The
finding may have provided added rationale for active blood pressure control in hypertensive
individuals who may harbor intracranial aneurysms or who are at high risk of rupture or who
may already be undergoing treatment for SAH. Though of interest may be to understand the
likely mechanism(s) through which premorbid hypertension may influence outcome. Our
analysis indicated a multifactorial pathway including severer primary injury and higher risk of
complications such as rebleeding in patients with premorbid hypertension.
Early prediction of outcome after SAH is very challenging given the variable course of
the condition. In chapter 4, a set of prediction models were presented that, in terms of
methodological validity, are arguably the most robust set of prediction models available for early
prediction of outcome in patients with SAH. The development sample reflects diverse patient
background clinically, temporally, therapeutically and geographically. The predictive
performances of the models were evaluated with greater emphasis on calibration behavior than is
151
seen in the literature on SAH, and using currently recommended parameters, including
calibration in the large and recalibration slope. Calibration is very relevant to model application
to individual patients. The good discrimination performance of the prognostic models implies
that they are potentially useful for risk stratification of SAH cohorts. The scores could therefore
serve either as tools for screening potential trial subjects based on the concept of prognostic
targeting or as tools to stratify trial subjects into more homogenous prognostic groups based on
their baseline prognostic risk to study treatment effect on different subgroups of patients.
Prognostic targeting and stratification randomization are particularly valuable in small trials
where small imbalances in baseline prognosis may impact on treatment effect. Most SAH trials
are small trials. Theoretically these strategies could reduce trial sample size without loss of study
power.75
As researchers involved in the conduct of clinical trials in SAH increasingly advocate
for a rethink of the strategies for improving the conduct and analysis of RCTs in SAH,270
it is
likely that we have provided them a reliable tool to evaluate and probably adopt novel strategies
to advance the conduct and analysis of RCTs in SAH. The practical relevance of the SAHIT
scores is better appreciated against the background that among 50 or more RCTs conducted to
date to assess novel therapies in SAH, only two have demonstrated robust beneficial effects: the
British Aneurysm Nimodipine Trial (BRANT) to determine the efficacy of oral nimodipine to
reduce cerebral infarction and poor outcome271
, and the International Subarachnoid Aneurysm
Trial (ISAT) to compare the safety and efficacy of endovascular coiling with microsurgical
clipping in patients at equipoise to receive either treatment.58
The SAHIT scores may have
further applications in comparative effectiveness and health services research where they could
serve as objective tools to adjust for differences in case mix between practice settings and assess
the possibility of health services provider profiling.
152
Very rarely, if ever, do prediction models estimate an outcome with absolute certainty.
Systematic miscalibration is common when models are transported to new settings. This is so
because prediction models are developed from a population of patients and commonly give
outcome probabilities for an average patient, given the values of the set of parameters included in
the models. In new patients, insight into the extent of uncertainty may be as relevant as the
predicted probabilities. Clinicians may find particularly valuable models or risk scores indicating
the margins of error around the predicted outcome probabilities. Very few prognostic models/
risk scores provide this insight. Unlike available models/ risk scores in SAH, the SAHIT scores
were designed to give estimates of the error margin associated with the predicted probabilities.
Clinicians involved in the management of patients with SAH currently have limited tools for
early estimation of patient outcome. They rely mostly on clinical experience and intuition for
prognosticating patient outcome. This approach could be unreliable, as shown in a previous
study.229
The SAHIT scores could be supportive aids providing clinicians managing SAH
patients the empirical evidence to reflect on their clinical intuition and to engage with patients
and significant others about treatment choices and rational expectations for the immediate
foreseeable future.
For model presentation, we adopted score charts with the option of formatting the models
as a software application. This was done to optimize face validity and predictive accuracy and
maximize uptake into clinical practice. The uptake of prognostic models and risk scores among
clinicians may depend on the balance between face validity and predictive accuracy. A
prediction model that is presented as a regression formula gives the highest predictive accuracy
possible with the model but it may be cumbersome to apply which may reduce its uptake. In
contrast, a model that is presented in more parsimonious formats including nomograms or score
153
charts is relatively less accurate because of rounding of linear predictors, but it is easier to use,
hence its uptake is probably better guaranteed. With increasing utilization of computer decision
support systems especially in the management of acute life threatening conditions such as SAH,
more models and risk scores are now presented as software programs or applications for
integration into clinical decision support systems or for online usage or download into mobile
devices. These formats are more compatible with the busy environment of clinical practice.
Among the several factors that interact to influence health and health outcomes,
socioeconomic status stands out having been recognized as the single most important contributor
to health disparities.244, 255, 272
Race/ethnicity differences have been investigated in specific
disease conditions where insight has been gained into sociocultural factors and biologic
mechanisms that underlie increased vulnerability of certain populations to some disease
conditions.273
Better understanding of the role of socioeconomic factors and race/ethnicity as
etiologic or prognostic risk factors for a given condition may be crucial to designing effective
interventions to ensure improved outcomes for all populations. In chapter 5, we demonstrated
proof-of-concepts that socioeconomic status and race/ethnicity are significantly associated with
the outcomes of SAH. The findings are noteworthy. For the first time, nationally representative
evidence has been presented linking socioeconomic differences and race/ethnicity to clinical
outcomes of SAH. The inverse relationship between SES and in-patient mortality that was
apparent in SAH patients who received treatment in the US which was less well established in
Canada was attributed to the Canadian social safety net, though the hypothesis requires
validation. While epidemiological data regarding race/ethnicity differences in the incidence and
outcomes of SAH is growing, the focus has often been on differences between white and non-
white populations. This dissertation presented data indicating that the relationship of
154
race/ethnicity to SAH outcomes transcends a simple comparison of white versus non-white
populations, as was done often in the literature. The data suggested that the Hispanic mortality
paradox which has been reported in other conditions also applies to SAH. The data further
indicated worse outcomes in patients of Asian/Pacific Island ethnicity, for reasons that are
unclear.
6.4 Limitations of the research
The findings of this research should be considered in the context of certain limitations, some of
which has been earlier highlighted in the relevant section on each individual study. Of note is
that none of the studies presented in this thesis is population-based. Some may argue that
population based studies should provide more robust estimates of the average effect of
prognostic factors given the potential to account for the approximately 12% of patients who
would have died prior to hospital admission. However, a population based study is unlikely to be
similarly powered. Furthermore, the cost of a population based study that would be similarly
powered to this research will be prohibitive and the relative benefit of such a study may be
minimal. Second, some of the included studies in the SAHIT repository are relatively old, having
preceded aneurysm repair by endovascular coiling. Nonetheless, no evidence was found as to a
significant difference in prognostic associations between relatively older and more recent studies
(Chapter 3, part A). That the SAHIT repository data were weighted towards patients who were
enrolled into RCTs compared with patients who were enrolled into observational studies raises
the possibility of selection bias. Nevertheless, the effect of this skewed distribution was found to
be relevant only with respect to understanding prognostic associations in patients with the
poorest grade of neurologic status (Chapter 3, part B). Some may argue from a methodological
viewpoint that the meta-analyses presented in chapter 3 may be confounded by availability bias
155
as primary studies in the SAHIT repository represent only a fraction of all possible previous
primary studies that are includable in the analyses. On the other hand, one may also argue from a
conceptual viewpoint that the primary studies in the SAHIT repository adequately reflect the
diverse case mix and prevalent practice patterns seen in SAH; hence the datasets of these studies
could be considered appropriate and adequate for drawing reliable conclusions about the average
prognostic value of studied prognostic factors. Moreover, the primary purpose of the analysed
studies was not to investigate prognostic associations in SAH.
The primary endpoints for prognostic analysis were the GOS and mortality. Both
measures are relatively insensitive. None was developed specifically for patients with SAH.
Nevertheless, in the absence of reliable outcome measure that is specific to SAH, the GOS
remains a preferred outcome measure for observational studies and randomized clinical trials in
SAH. The time point for prognostic analysis was 3 months. Prognosis could be expected to
improve beyond this time point. Hence, the long term implications of the research findings are
uncertain. However, the 3-month time point is generally considered adequate to assess clinical
outcomes of patients with SAH and it is commonly used in randomized clinical trials in SAH.
Large administrative databases such as the NIS and DAD present the opportunity to test
hypothesis with very large patient numbers that represent a large proportion of the population of
interest, at relatively low cost. The datasets reflect “real world” practice, and enable greater
generalizability of study findings. However, these databases were not primarily designed for
outcomes research. Their cross sectional design implies that analysis could not be performed
with reference to a specific time point or to assess changes with time, and no causal inferences
could be made. Moreover, because of their lack of specificity data was unavailable for measures
indicative of case severity, such as the WFNS or other measures of neurologic status. Patients in
156
both databases are de-identified; as a result, data linkage is impossible with the NIS and
particularly difficult and time consuming with the DAD. The NIS and DAD are not without
coding and sampling errors. Studies suggest diagnosis code error rates of 16.5% in the NIS274
and 12.5% in the DAD.275
The NIS also had the limitation of a significant proportion of missing
data on race/ethnicity which necessitated data imputation to minimise bias. A significant
advantage of the databases, however, is that both are nationally representative and therefore
useful for examining national trends in disparities in access to healthcare and in health outcomes.
In the NIS and DAD, the surrogate measure of SES was neighborhood income status. Our study
was therefore limited to this measure of SES which we analysed as a patient level variable
though it is a group level variable. The study on SES (Chapter 5, part A) may therefore be
subject to the “ecological fallacy” as our use of neighborhood income leaves room for alternative
interpretations of the study findings. For instance, factors associated with self-selection into
certain neighborhoods rather than SES may explain the study findings. One of such factors is
race/ethnicity, which was accounted for in the analysis. Individual measures of SES such as
personal income identify a patient’s SES more directly than ecological measures of SES such as
neighborhood income status. Individual measures of SES are less subject to misclassification
bias and probably estimate the SES-outcome association more appropriately. However, they may
be less effective than ecological measures of SES at reflecting the contributions of the social and
economic factors that influence all individuals who share a particular social environment.255
157
6.5 Directions for future research
6.5.1 Confirmatory study of the prognostic value of other factors
The present research focused on 7 baseline demographic, clinical and neuroimaging
factors whose prognostic effect has been examined previously. Confirmatory analyses of the
prognostic value of the factors were performed using the technique of individual participant data
(patient level) meta-analysis and multivariable analysis for evidence synthesis. A plethora of
other factors with potential prognostic value have been reported in the literature – particularly
admission laboratory parameters or biomarkers – but whether they have added incremental
predictive value beyond those of established prognostic factors has not been sufficiently
examined.168, 169
The current research can be extended to synthesize evidence on the added value
of admission laboratory parameters for prognostication in patients with SAH. Of some interest
are admission glucose, hemoglobin, and markers of systemic inflammatory response after SAH,
among others. Available evidence indicates serum glucose level, hemoglobin level and fever are
independent prognosticators of outcome in patients with SAH.276-278
Their average effect and
added value could be more precisely determined using study design similar to that of this thesis.
Research in traumatic brain injury demonstrates the prospect of improving predictive ability
when biomarkers are added to other known factors in prognostic models.279
Furthermore, the
present research suggested SES has prognostic relevance in SAH. Given the findings and
limitations of the study, there is the need for more nuanced analysis to better understand the
relation between SES and SAH outcomes, and what motivating or mitigating factors relate to the
observed effect of SES on SAH. We demonstrated that the effect of SES was independent of
patient’s age, sex, race/ethnicity, and comorbid status or hospital factors including hospital
location, bed size or teaching status. Nevertheless, an intermediary role could be hypothesized
158
for systematic differences in access to care, provision of the processes of care, and classic
behavioral vascular risk factors, which have been implicated in other disease conditions.272, 280
Future research exploring the relation of SES to SAH and what factors intermediate the
relationship should preferably utilize prospective study design with data on individual measures
of SES, account for case severity and center location and assess outcomes at fixed time points or
in a longitudinal study.
6.5.2 Validation of Risk Scores in Subarachnoid Aneurysm (VISA)
External validation of prognostic models is essential, but has seldom been performed in
SAH. Although the SAHIT scores demonstrated good discrimination and reasonable calibration
on cross validation, suggesting that they are robust to changes in settings, further research is still
needed to examine whether the scores yield comparable performance in new patients from a
variety of settings. Different types and levels of validation have been proposed by different
experts. McGinn et al. 281
proposed 4 levels of validation. Level 4 validation aims to provide
preliminary information about the stability of the prognostic model/score for a limited, well
defined population. Level 3 validation aims to determine in prospective cohorts whether the
proposed model have predictive accuracy in different but similar samples. Level 2 validation
assesses whether the proposed model or score yields comparable results for a variety of patients,
which would recommend it for application to a variety of settings. Level 1 validation is aimed at
determining whether the proposed model changes clinical behavior and improves overall clinical
practice and outcomes of patients. Reilly and Evans282
proposed a conceptually similar hierarchy
of model validation. Their 5 level validation scheme include: (1) Validation in the development
cohort; (2) Narrow validation of the proposed model or score using prospective data from one
setting; (3) Broad validation of the proposed model or score using prospective data from varied
159
settings with a broad case mix; (4) Narrow impact analysis of the proposed model to evaluate its
utility as a decision tool; and (5) Broad impact analysis using prospective data from a variety of
settings to confirm that the proposed model or score improves clinical decision making in a wide
spectrum of settings. A different scheme of validation reported in a review by Toll et al.283
propose 3 levels of validation. Temporal validation evaluates the generalizability of the model
over time using one or a few institutions, same physicians or investigators. Geographic
validation assesses the generalizability of the proposed model or score in “a patient population
that is similarly defined as the development cohort, though in hospitals or institutions of other
geographic areas.”283
The authors stated that in domain validation, the generalizability of the
model or score is evaluated across different domains. The proposed model would be assessed in
patients from different settings – primary, secondary or tertiary care; academic versus general
hospital settings; patients of different age categories, among other relevant domains. It is obvious
therefore from the different validation schemes that model validation is a continuous process,
and a long period of time may elapse before model translation into practice can be fully
achieved. The collaborative nature of the SAHIT and the continuous accruing of data from
different settings into the SAHIT repository can be exploited to evaluate the performance of the
SAHIT score at the different levels of validation. The size and heterogeneity of data in the
SAHIT repository could also be exploited for purposes of validating previously reported
prognostic models and risk scores. The relative performance of prognostic models in SAH could
be compared so as to facilitate their timely translation into clinical practice, enabling clinicians to
make informed choices about the best tools to aid their clinical judgement.
6.5.3 Center variability in outcomes of SAH
Some evidence exists in the literature to indicate that the outcomes of patients with SAH
160
vary by center; however no systematic empirical investigation has been carried out pertaining to
this observation. The International Cooperative Study on Timing of Aneurysm Surgery that
accrued data on patients with SAH between 1980 and 1983 from 68 centers in 14 countries found
considerable variations in outcome between geographic regions and individual centers
participating in the study.139
The range of mortality rate between centers was 0% to 66%. The
study reported that in 5 centers patient outcomes were particularly poor and each of these centers
contributed a significant number of cases into the study. Because the 5 centers were clustered in
the same geographic region the study speculated that the effect could be related to some
management practice that was common to the centers. A study of 350 patients with SAH from 22
Neurosurgical centers in Italy reported considerable heterogeneity in the case mix and outcomes
of patients by centers.284
In a meta-analysis of 33 studies involving 8739 patients with SAH to
assess changes in case fatality over time, Nieuwkamp et al. found marked variability in case
fatality rate among the included studies.113
The case fatality rates were in the range 8.3% to
66.7%. Scarcely any insight is available as to what factors may be responsible for between-
centers variability in the outcomes of patients with SAH. It is likely the resultant effect of a
multiplicity of interacting factors including such factors as pathophysiologic heterogeneity due to
hitherto poorly understood genetic or epigenetic differences in the cases seen in different centers,
or more plausibly differences in the management strategies adopted by different centers. In their
meta-analysis, Nieuwkamp and coworkers adduced similar explanations for the variations in case
fatality rates among the different centers. They attributed the between-centers variability in case
fatality to methodological differences between studies, differences in proportion of patients who
died before admission and management differences, which could include differences in timing of
aneurysm repair, use of nimodipine, stroke units, endovascular coiling and variations in intensive
161
care management. Some centers treating patients with SAH administer antifibrinolytic drugs to
reduce rebleeding before the aneurysm is repaired, other centers do not. Rescue therapies such as
induced hypertension and angioplasty vary by centers. Similarly, medical management such as
use of anticonvulsants has been shown to vary by centers.285
Should between-centers variation be
due to modifiable management strategies, then a standardized protocol for the management of
patients with SAH could be developed and meaningful improvements in outcome may be
achieved.286
Researchers have reported better outcomes in centers handling high volumes of
cases of SAH than centers handling low volumes of cases; a finding which may be related to the
concentration of specialized resources in high-volume centers, and is supportive of a policy of
regionalization of care as a management strategy to improve outcomes in SAH.
Additionally, a better understanding of the role of center variability in SAH outcomes
could have implications for the conduct and analysis of multicenter randomized clinical trials in
SAH, particularly the use of stratification randomization during patient enrolment or covariate
adjustment for center effect in the analysis phase. The current practice is to consider between-
centers variability as unexplained variance in outcome for apparently similar patients treated in
different centers. Where between-center variability is considerable and systematic, it could mask
drug effect leading to a negative trial. The high prevalence of negative trials in SAH gives added
impetus for greater insight into how center variability relates to outcomes of patients with SAH.
6.5.4 Type and timing of outcome assessment
We noted as study limitations the rather crude nature of the outcome measures used in the
studies reported in this thesis and the short follow up duration. We rationalized that prognosis
would improve with time; however, at what time point the trajectory of outcome plateaus is
unknown. This issue could be explored in a longitudinal study to determine the optimal timing of
162
outcomes after SAH, which has implications for the conduct of randomized clinical trials and
observational studies in SAH, and clinical practice as well. Navi and coworkers reported
outcome at discharge and 6 months on the modified Rankin scale in 52 patients with SAH and
found improvement on the modified Rankin scale over this time period.287
Wilson and colleagues
stated that most studies assess outcome 3 to 6 months after SAH.287
They reviewed 88 patients
who were Hunt and Hess grades 4 or 5 and who were entered into a clinical trial comparing
clipping and coiling. Outcome was assessed at hospital discharge and then 6, 12 and 36 months
after SAH. The main improvement on the modified Rankin scale was between discharge and 6
months, although there was improvement of at least one grade on the modified Rankin scale in
about 1 in 5 patients between 6 and 12 and 12 and 36 months. It is likely that the optimal time
point for outcome assessment may differ with outcome measure. Day to day functioning and
cognitive recover may take several months. A longitudinal study which assessed the cognitive
function of 42 SAH patients at 3, 9 and 18 months reported significant improvement in delayed
verbal memory between 3 and 9 months and between 9 and 18 months post-ictus. In contrast, no
significant improvement occurred in immediate verbal memory.7, 288, 289
The study indicated that
although improvement occurred in delayed verbal memory over 18 months, 14% of patients with
SAH experienced significant delayed verbal memory impairment at 18-month follow-up.
Variations in time to recovery have also been documented in other cognitive domains and
functional outcomes. It is likely that location of brain lesion plays a role in the variation in time
to recovery, though other unknown factors may be involved also. Research is therefore needed to
investigate the optimal time point for assessment of different outcomes in SAH and the reasons
for the variations in time to recovery.
163
6.6 Conclusion
In conclusion, the overall aim of the research presented in this thesis was to provide a
higher level of evidence on the nature and shape of prognostic associations in SAH. Extensive
analyses were performed in large cohorts of patients with SAH to more precisely determine the
prognostic value of 7 conventional prognostic factors in SAH. Prognostic scores were then
developed combining the information for early prediction of risk of mortality and unfavorable
outcome in patients with SAH. The scores were shown to have satisfactory predictive ability in a
variety of settings, hence could potentially be useful as decision support aids and tools to
advance interventional and health services research in SAH. Insight was further provided on the
relation between socioeconomic status, race/ethnicity and outcomes of SAH. The research
suggested a gradient effect of SES on inpatient mortality, the extent of which could have a
bearing to national health care system. Further evidence was found suggesting that race/ethnicity
is a predictor of inpatient mortality risk in the United States. A Hispanic mortality advantage was
found, and outcome was worse among SAH patients of Asian/ pacific island nativity.
164
Reference List
(1) de Rooij NK, Linn FH, van der Plas JA, Algra A, Rinkel GJ. Incidence of subarachnoid
haemorrhage: a systematic review with emphasis on region, age, gender and time
trends. J Neurol Neurosurg Psychiatry 2007 December;78(12):1365-72.
(2) Ingall T, Asplund K, Mahonen M, Bonita R. A multinational comparison of
subarachnoid hemorrhage epidemiology in the WHO MONICA stroke study. Stroke
2000 May;31(5):1054-61.
(3) Rinkel GJ, Djibuti M, Algra A, van GJ. Prevalence and risk of rupture of intracranial
aneurysms: a systematic review. Stroke 1998 January;29(1):251-6.
(4) Steiner T, Juvela S, Unterberg A, Jung C, Forsting M, Rinkel G. European stroke
organization guidelines for the management of intracranial aneurysms and
subarachnoid haemorrhage. Cerebrovasc Dis 2013;35(2):93-112.
(5) Pakarinen S. Incidence, aetiology, and prognosis of primary subarachnoid
haemorrhage. A study based on 589 cases diagnosed in a defined urban population
during a defined period. Acta Neurol Scand 1967;43:Suppl 29:1-28.:Suppl-28.
(6) Rinkel GJ, Algra A. Long-term outcomes of patients with aneurysmal subarachnoid
haemorrhage. Lancet Neurol 2011 April;10(4):349-56.
(7) Al-Khindi T, Macdonald RL, Schweizer TA. Cognitive and functional outcome after
aneurysmal subarachnoid hemorrhage. Stroke 2010 August;41(8):e519-e536.
(8) Ferro JM, Canhao P, Peralta R. Update on subarachnoid haemorrhage. Journal of
Neurology 2008;255(4):465-79.
(9) Dupont SA, Wijdicks EF, Lanzino G, Rabinstein AA. Aneurysmal subarachnoid
hemorrhage: an overview for the practicing neurologist. Semin Neurol 2010
November;30(5):545-54.
(10) van GJ, Kerr RS, Rinkel GJ. Subarachnoid haemorrhage. Lancet 2007 January
27;369(9558):306-18.
(11) Mccormick WF, Nofzinger JD. Saccular intracranial aneurysms: an autopsy study. J
Neurosurg 1965 February;22:155-9.:155-9.
(12) Heiskanen O. Multiple intracranial arterial aneurysms. Acta Neurol Scand
1966;41(4):356-62.
165
(13) Yoshimoto T, Kayama T, Kodama N, Suzuki J. Distribution of intracranial aneurysm.
Tohoku J Exp Med 1978 October;126(2):125-32.
(14) Feigin VL, Rinkel GJE, Lawes CMM, Algra A, Bennett DA, van Gijn J, Anderson CS.
Risk factors for subarachnoid hemorrhage - An updated systematic review of
epidemiological studies. Stroke 2005;36(12):2773-80.
(15) Ballard J, Kreiter KT, Claassen J, Kowalski RG, Connolly ES, Mayer SA. Risk factors
for continued cigarette use after subarachnoid hemorrhage. Stroke 2003
August;34(8):1859-63.
(16) Lindekleiv H, Sandvei MS, Njolstad I, Lochen ML, Romundstad PR, Vatten L,
Ingebrigtsen T, Vik A, Mathiesen EB. Sex differences in risk factors for aneurysmal
subarachnoid hemorrhage: a cohort study. Neurology 2011 February 15;76(7):637-43.
(17) Clarke M. Systematic review of reviews of risk factors for intracranial aneurysms.
Neuroradiology 2008 August;50(8):653-64.
(18) Ruigrok YM, Buskens E, Rinkel GJ. Attributable risk of common and rare determinants
of subarachnoid hemorrhage. Stroke 2001 May;32(5):1173-5.
(19) Lall RR, Eddleman CS, Bendok BR, Batjer HH. Unruptured intracranial aneurysms and
the assessment of rupture risk based on anatomical and morphological factors: sifting
through the sands of data. Neurosurg Focus 2009 May;26(5):E2.
(20) Morita A, Kirino T, Hashi K, Aoki N, Fukuhara S, Hashimoto N, Nakayama T, Sakai
M, Teramoto A, Tominari S, Yoshimoto T. The natural course of unruptured cerebral
aneurysms in a Japanese cohort. N Engl J Med 2012 June 28;366(26):2474-82.
(21) Morita A, Fujiwara S, Hashi K, Ohtsu H, Kirino T. Risk of rupture associated with
intact cerebral aneurysms in the Japanese population: a systematic review of the
literature from Japan. J Neurosurg 2005 April;102(4):601-6.
(22) Wiebers DO, Whisnant JP, Huston J, III, Meissner I, Brown RD, Jr., Piepgras DG,
Forbes GS, Thielen K, Nichols D, O'Fallon WM, Peacock J, Jaeger L, Kassell NF,
Kongable-Beckman GL, Torner JC. Unruptured intracranial aneurysms: natural history,
clinical outcome, and risks of surgical and endovascular treatment. Lancet 2003 July
12;362(9378):103-10.
(23) Etminan N, Beseoglu K, Steiger HJ, Hanggi D. The impact of hypertension and
nicotine on the size of ruptured intracranial aneurysms. J Neurol Neurosurg Psychiatry
2011 January;82(1):4-7.
(24) Lindner SH, Bor AS, Rinkel GJ. Differences in risk factors according to the site of
intracranial aneurysms. J Neurol Neurosurg Psychiatry 2010 January;81(1):116-8.
(25) Connolly ES, Jr., Rabinstein AA, Carhuapoma JR, Derdeyn CP, Dion J, Higashida RT,
Hoh BL, Kirkness CJ, Naidech AM, Ogilvy CS, Patel AB, Thompson BG, Vespa P.
166
Guidelines for the management of aneurysmal subarachnoid hemorrhage: a guideline
for healthcare professionals from the American Heart Association/american Stroke
Association. Stroke 2012 June;43(6):1711-37.
(26) Ronkainen A, Miettinen H, Karkola K, Papinaho S, Vanninen R, Puranen M,
Hernesniemi J. Risk of harboring an unruptured intracranial aneurysm. Stroke 1998
February;29(2):359-62.
(27) Ruigrok YM, Rinkel GJ, Wijmenga C. Genetics of intracranial aneurysms. Lancet
Neurol 2005 March;4(3):179-89.
(28) Vlak MH, Rinkel GJ, Greebe P, Greving JP, Algra A. Lifetime risks for aneurysmal
subarachnoid haemorrhage: multivariable risk stratification. J Neurol Neurosurg
Psychiatry 2013 June;84(6):619-23.
(29) Tiel Groenestege AT, Rinkel GJ, van der Bom JG, Algra A, Klijn CJ. The risk of
aneurysmal subarachnoid hemorrhage during pregnancy, delivery, and the puerperium
in the Utrecht population: case-crossover study and standardized incidence ratio
estimation. Stroke 2009 April;40(4):1148-51.
(30) Larsson SC, Mannisto S, Virtanen MJ, Kontto J, Albanes D, Virtamo J. Dietary fiber
and fiber-rich food intake in relation to risk of stroke in male smokers. European
Journal of Clinical Nutrition 2009;63(8):1016-24.
(31) Gill RS, Hambridge HL, Schneider EB, Hanff T, Tamargo RJ, Nyquist P. Falling
Temperature and Colder Weather Are Associated with an Increased Risk of
Aneurysmal Subarachnod Hemorrhage. World Neurosurg 2012 June 23.
(32) Macdonald RL. Whether subarachnoid hemorrhage depends on the weather? World
Neurosurg 2013 January;79(1):64-5.
(33) McDonald RJ, McDonald JS, Bida JP, Kallmes DF, Cloft HJ. Subarachnoid
hemorrhage incidence in the United States does not vary with season or temperature.
AJNR Am J Neuroradiol 2012 October;33(9):1663-8.
(34) Cowperthwaite MC, Burnett MG. The association between weather and spontaneous
subarachnoid hemorrhage: an analysis of 155 US hospitals. Neurosurgery 2011
January;68(1):132-8.
(35) Caner B, Hou J, Altay O, Fuj M, Zhang JH. Transition of research focus from
vasospasm to early brain injury after subarachnoid hemorrhage. J Neurochem 2012
November;123 Suppl 2:12-21. doi: 10.1111/j.1471-4159.2012.07939.x.:12-21.
(36) Ayer R, Zhang J. Connecting the early brain injury of aneurysmal subarachnoid
hemorrhage to clinical practice. Turk Neurosurg 2010 April;20(2):159-66.
(37) Cahill J, Calvert JW, Zhang JH. Mechanisms of early brain injury after subarachnoid
hemorrhage. J Cereb Blood Flow Metab 2006 November;26(11):1341-53.
167
(38) Sehba FA, Pluta RM, Zhang JH. Metamorphosis of subarachnoid hemorrhage research:
from delayed vasospasm to early brain injury. Mol Neurobiol 2011 February;43(1):27-
40.
(39) Sehba FA, Hou J, Pluta RM, Zhang JH. The importance of early brain injury after
subarachnoid hemorrhage. Prog Neurobiol 2012 April;97(1):14-37.
(40) Budohoski KP, Czosnyka M, Kirkpatrick PJ, Smielewski P, Steiner LA, Pickard JD.
Clinical relevance of cerebral autoregulation following subarachnoid haemorrhage. Nat
Rev Neurol 2013 March;9(3):152-63.
(41) Sehba FA, Bederson JB. Mechanisms of acute brain injury after subarachnoid
hemorrhage. Neurol Res 2006 June;28(4):381-98.
(42) Bassi P, Bandera R, Loiero M, Tognoni G, Mangoni A. Warning signs in subarachnoid
hemorrhage: a cooperative study. Acta Neurol Scand 1991 October;84(4):277-81.
(43) Brilstra EH, Rinkel GJ, Algra A, van GJ. Rebleeding, secondary ischemia, and timing
of operation in patients with subarachnoid hemorrhage. Neurology 2000 December
12;55(11):1656-60.
(44) Linn FH, Rinkel GJ, Algra A, van GJ. Headache characteristics in subarachnoid
haemorrhage and benign thunderclap headache. J Neurol Neurosurg Psychiatry 1998
November;65(5):791-3.
(45) Beck J, Raabe A, Szelenyi A, Berkefeld J, Gerlach R, Setzer M, Seifert V. Sentinel
headache and the risk of rebleeding after aneurysmal subarachnoid hemorrhage. Stroke
2006 November;37(11):2733-7.
(46) Kowalski RG, Claassen J, Kreiter KT, Bates JE, Ostapkovich ND, Connolly ES, Mayer
SA. Initial misdiagnosis and outcome after subarachnoid hemorrhage. JAMA 2004
February 18;291(7):866-9.
(47) Cortnum S, Sorensen P, Jorgensen J. Determining the sensitivity of computed
tomography scanning in early detection of subarachnoid hemorrhage. Neurosurgery
2010 May;66(5):900-2.
(48) Macdonald RL, Pluta RM, Zhang JH. Cerebral vasospasm after subarachnoid
hemorrhage: the emerging revolution. Nat Clin Pract Neurol 2007 May;3(5):256-63.
(49) Fisher CM, Roberson GH, Ojemann RG. Cerebral vasospasm with ruptured saccular
aneurysm--the clinical manifestations. Neurosurgery 1977 November;1(3):245-8.
(50) Fergusen S, MacDonald RL. Predictors of cerebral infarction in patients with
aneurysmal subarachnoid hemorrhage. Neurosurgery 2007;60(4):658-67.
168
(51) Klimo P, Jr., Schmidt RH. Computed tomography grading schemes used to predict
cerebral vasospasm after aneurysmal subarachnoid hemorrhage: a historical review.
Neurosurg Focus 2006 September 15;21(3):E5.
(52) Rinkel GJ, Wijdicks EF, Vermeulen M, Tans JT, Hasan D, van GJ. Acute
hydrocephalus in nonaneurysmal perimesencephalic hemorrhage: evidence of CSF
block at the tentorial hiatus. Neurology 1992 September;42(9):1805-7.
(53) Hasan D, Vermeulen M, Wijdicks EF, Hijdra A, van GJ. Management problems in
acute hydrocephalus after subarachnoid hemorrhage. Stroke 1989 June;20(6):747-53.
(54) Sundaram MB, Chow F. Seizures associated with spontaneous subarachnoid
hemorrhage. Can J Neurol Sci 1986 August;13(3):229-31.
(55) Solenski NJ, Haley EC, Jr., Kassell NF, Kongable G, Germanson T, Truskowski L,
Torner JC. Medical complications of aneurysmal subarachnoid hemorrhage: a report of
the multicenter, cooperative aneurysm study. Participants of the Multicenter
Cooperative Aneurysm Study. Crit Care Med 1995 June;23(6):1007-17.
(56) Wartenberg KE, Mayer SA. Medical complications after subarachnoid hemorrhage:
new strategies for prevention and management. Current Opinion in Critical Care
2006;12(2):78-84.
(57) Vespa P, Diringer MN. High-volume centers. Neurocrit Care 2011
September;15(2):369-72.
(58) Molyneux A, Kerr R, Stratton I, Sandercock P, Clarke M, Shrimpton J, Holman R.
International Subarachnoid Aneurysm Trial (ISAT) of neurosurgical clipping versus
endovascular coiling in 2143 patients with ruptured intracranial aneurysms: a
randomised trial. Lancet 2002 October 26;360(9342):1267-74.
(59) van den Berg R, Foumani M, Schroder RD, Peerdeman SM, Horn J, Bipat S, Vandertop
WP. Predictors of outcome in World Federation of Neurologic Surgeons grade V
aneurysmal subarachnoid hemorrhage patients. Crit Care Med 2011
December;39(12):2722-7.
(60) Wostrack M, Sandow N, Vajkoczy P, Schatlo B, Bijlenga P, Schaller K, Kehl V,
Harmening K, Ringel F, Ryang YM, Friedrich B, Stoffel M, Meyer B. Subarachnoid
haemorrhage WFNS grade V: is maximal treatment worthwhile? Acta Neurochir (Wien
) 2013 April;155(4):579-86.
(61) Taylor CJ, Robertson F, Brealey D, O'shea F, Stephen T, Brew S, Grieve JP, Smith M,
Appleby I. Outcome in poor grade subarachnoid hemorrhage patients treated with acute
endovascular coiling of aneurysms and aggressive intensive care. Neurocrit Care 2011
June;14(3):341-7.
169
(62) Komotar RJ, Schmidt JM, Starke RM, Claassen J, Wartenberg KE, Lee K, Badjatia N,
Connolly ES, Jr., Mayer SA. Resuscitation and critical care of poor-grade subarachnoid
hemorrhage. Neurosurgery 2009 March;64(3):397-410.
(63) Sedat J, Dib M, Rasendrarijao D, Fontaine D, Lonjon M, Paquis P. Ruptured
intracranial aneurysms in the elderly - Epidemiology, diagnosis, and management.
Neurocritical Care 2005;2(2):119-23.
(64) Wilson TJ, Davis MC, Stetler WR, Giles DA, Chaudhary N, Gemmete JJ, Thompson
BG, Pandey AS. Endovascular treatment for aneurysmal subarachnoid hemorrhage in
the ninth decade of life and beyond. J Neurointerv Surg 2013 March 27.
(65) Scholler K, Massmann M, Markl G, Kunz M, Fesl G, Bruckmann H, Pfefferkorn T,
Tonn JC, Schichor C. Aneurysmal subarachnoid hemorrhage in elderly patients: long-
term outcome and prognostic factors in an interdisciplinary treatment approach. J
Neurol 2013 April;260(4):1052-60.
(66) Huang J, van Gelder JM. The probability of sudden death from rupture of intracranial
aneurysms: a meta-analysis. Neurosurgery 2002 November;51(5):1101-5.
(67) Macdonald RL, Cusimano MD, Etminan N, Hanggi D, Hasan D, Ilodigwe D, Jaja B,
Lantigua H, Le RP, Lo B, Louffat-Olivares A, Mayer S, Molyneux A, Quinn A,
Schweizer TA, Schenk T, Spears J, Todd M, Torner J, Vergouwen MD, Wong GK.
Subarachnoid Hemorrhage International Trialists data repository (SAHIT). World
Neurosurg 2013 March;79(3-4):418-22.
(68) Johnston SC, Selvin S, Gress DR. The burden, trends, and demographics of mortality
from subarachnoid hemorrhage. Neurology 1998 May;50(5):1413-8.
(69) Nieuwkamp DJ, Setz LE, Algra A, Linn FH, de Rooij NK, Rinkel GJ. Changes in case
fatality of aneurysmal subarachnoid haemorrhage over time, according to age, sex, and
region: a meta-analysis. Lancet Neurol 2009 July;8(7):635-42.
(70) Hop JW, Rinkel GJE, Algra A, van Gun J. Changes in functional outcome and quality
of life in patients and caregivers after aneurysmal subarachnoid hemorrhage. J
Neurosurg 2001;95(6):957-63.
(71) Covey J, Noble AJ, Schenk T. Family and friends' fears of recurrence: impact on the
patient's recovery after subarachnoid hemorrhage. J Neurosurg 2013
October;119(4):948-54.
(72) Rivero-Arias O, Gray A, Wolstenholme J. Burden of disease and costs of aneurysmal
subarachnoid haemorrhage (aSAH) in the United Kingdom. Cost Eff Resour Alloc 2010
April 27;8:6.:6.
(73) Qureshi AI, Suri MFK, Abu N, Kirmani JF, Ezzeddine MA, Divani AA, Giles WH.
Changes in cost and outcome among US patients with stroke hospitalized in 1990 to
1991 and those hospitalized in 2000 to 2001. Stroke 2007;38(7):2180-4.
170
(74) Riley RD, Hayden JA, Steyerberg EW, Moons KG, Abrams K, Kyzas PA, Malats N,
Briggs A, Schroter S, Altman DG, Hemingway H. Prognosis Research Strategy
(PROGRESS) 2: prognostic factor research. PLoS Med 2013;10(2):e1001380.
(75) Steyerberg EW, Moons KG, van der Windt DA, Hayden JA, Perel P, Schroter S, Riley
RD, Hemingway H, Altman DG. Prognosis Research Strategy (PROGRESS) 3:
prognostic model research. PLoS Med 2013;10(2):e1001381.
(76) Hingorani AD, Windt DA, Riley RD, Abrams K, Moons KG, Steyerberg EW, Schroter
S, Sauerbrei W, Altman DG, Hemingway H. Prognosis research strategy (PROGRESS)
4: stratified medicine research. BMJ 2013 February 5;346:e5793. doi:
10.1136/bmj.e5793.:e5793.
(77) Hemingway H, Croft P, Perel P, Hayden JA, Abrams K, Timmis A, Briggs A,
Udumyan R, Moons KG, Steyerberg EW, Roberts I, Schroter S, Altman DG, Riley RD.
Prognosis research strategy (PROGRESS) 1: a framework for researching clinical
outcomes. BMJ 2013 February 5;346:e5595. doi: 10.1136/bmj.e5595.:e5595.
(78) Lanzino G, Kassell NF, Germanson TP, Kongable GL, Truskowski LL, Torner JC, Jane
JA, Spetzler RF, Zabramski J, Culicchia F, Carter LP, Feinberg W, Urbina C, Lopez L,
Brown D, Tallman D, Selman WR, Harrington F, Warf B, Barnett GH, Little J, Palmer
J, Campbell RL, Shapiro S, Farlow M, Kay S, Horner T, Leipzig T, Redelman K,
Muizelaar JP, Turner R, Kamsheh W, Bouma G, delosReyes R, Rosenbaum D,
Klonowski E, Mohr G, Bojanowski M, Bernier G, Duquette P, LaPlante P, Murali R,
Crisafulli C, Merkel C, Rosner MJ, Mahaley MS, Aronin PA, Morawetz RB, Rosner S,
BrockSillers R, Lamond RG, Marshall L, Rothrock J, Benedict B, Moore S, Young B,
Dempsey R, Tibbs P, Phillips R, Salcman M, Rigamonti D, Kulick S, Powers S, Baker
ML, Flamm ES, Reivich M, Kushner M, Farber R, Parrott S, Simard JM, Eisenberg
HM, Kapler B, Kent T, Aldrich F, Turner B, Grotta J, Kopanicky D, Yatsur F, DeGrapa
T, Bratina P, Heilbrun HP, Call G, Hood R, Goodman S, Sunderland P, Haley EC,
Parks K, Winn HR, Grady S, Mayberg M, Burnal C, Grubb RL, Coxe W, Landou W,
Rich K, Levin A, Adams HP, Drake CG, Dyken ML, Frankowski RF, Kurtzke JF, Weir
BKA, Haley EC, Cail WS, Elder LJ, Lightfoot AL, Marron JA, Polin AS. Age and
outcome after aneurysmal subarachnoid hemorrhage: Why do older patients fare
worse? J Neurosurg 1996;85(3):410-8.
(79) Deruty R, Pelissouguyotat I, Mottolese C, Amat D, Bognar L. Level of Consciousness
and Age As Prognostic Factors in Aneurysmal Sah. Acta Neurochirurgica 1995;132(1-
3):1-8.
(80) Lagares A, Gomez PA, Lobato RD, Alen JF, Alday R, Campollo J. Prognostic factors
on hospital admission after spontaneous subarachnoid haemorrhage. Acta Neurochir
(Wien ) 2001;143(7):665-72.
(81) Rabinstein AA, Friedman JA, Nichols DA, Pichelmann MA, McClelland RL, Manno
EM, Atkinson JL, Wijdicks EF. Predictors of outcome after endovascular treatment of
cerebral vasospasm. AJNR Am J Neuroradiol 2004 November;25(10):1778-82.
171
(82) Rosengart AJ, Schultheiss KE, Tolentino J, MacDonald RL. Prognostic factors for
outcome in patients with aneurysmal subarachnoid hemorrhage. Stroke 2007
August;38(8):2315-21.
(83) Risselada R, Lingsma HF, Bauer-Mehren A, Friedrich CM, Molyneux AJ, Kerr RS,
Yarnold J, Sneade M, Steyerberg EW, Sturkenboom MC. Prediction of 60 day case-
fatality after aneurysmal subarachnoid haemorrhage: results from the International
Subarachnoid Aneurysm Trial (ISAT). Eur J Epidemiol 2010 April;25(4):261-6.
(84) Lanzino G, Kassell NF, Germanson TP, Kongable GL, Truskowski LL, Torner JC, Jane
JA. Age and outcome after aneurysmal subarachnoid hemorrhage: why do older
patients fare worse? J Neurosurg 1996 September;85(3):410-8.
(85) Chung RY, Carter BS, Norbash A, Budzik R, Putnam C, Ogilvy CS. Management
outcomes for ruptured and unruptured aneurysms in the elderly. Neurosurgery 2000
October;47(4):827-32.
(86) Fortuny LA, Adams CB, Briggs M. Surgical mortality in an aneurysm population:
effects of age, blood pressure and preoperative neurological state. J Neurol Neurosurg
Psychiatry 1980 October;43(10):879-82.
(87) O'sullivan MG, Dorward N, Whittle IR, Steers AJ, Miller JD. Management and long-
term outcome following subarachnoid haemorrhage and intracranial aneurysm surgery
in elderly patients: an audit of 199 consecutive cases. Br J Neurosurg 1994;8(1):23-30.
(88) Ryttlefors M, Enblad P, Kerr RS, Molyneux AJ. International subarachnoid aneurysm
trial of neurosurgical clipping versus endovascular coiling: subgroup analysis of 278
elderly patients. Stroke 2008 October;39(10):2720-6.
(89) Awe OO, Gonzalez LF, Hasan D, Maltenfort M, Rossenwasser R, Jabbour P.
Treatment outcome of aneurysmal subarachnoid hemorrhage in patients aged 70 years
and older. Neurosurgery 2011 March;68(3):753-8.
(90) Horiuchi T, Tanaka Y, Hongo K. Surgical treatment for aneurysmal subarachnoid
hemorrhage in the 8th and 9th decades of life. Neurosurgery 2005 March;56(3):469-75.
(91) Kazumata K, Kamiyama H, Ishikawa T. Reference table predicting the outcome of
subarachnoid hemorrhage in the elderly, stratified by age. Journal of Stroke and
Cerebrovascular Diseases 2006 January;15(1):14-7.
(92) Nieuwkamp DJ, Rinkel GJ, Silva R, Greebe P, Schokking DA, Ferro JM. Subarachnoid
haemorrhage in patients > or = 75 years: clinical course, treatment and outcome. J
Neurol Neurosurg Psychiatry 2006 August;77(8):933-7.
(93) Keller AZ. Hypertension, age and residence in the survival with subarachnoid
hemorrhage. Am J Epidemiol 1970 February;91(2):139-47.
172
(94) Mocco J, Ransom ER, Komotar RJ, Schmidt JM, Sciacca RR, Mayer SA, Connolly ES,
Jr. Preoperative prediction of long-term outcome in poor-grade aneurysmal
subarachnoid hemorrhage. Neurosurgery 2006 September;59(3):529-38.
(95) Johansson M, Cesarini KG, Contant CF, Persson L, Enblad P. Changes in intervention
and outcome in elderly patients with subarachnoid hemorrhage. Stroke 2001 December
1;32(12):2845-949.
(96) Rowe JG, Molyneux AJ, Byrne JV, Renowden S, Aziz TZ. Endovascular treatment of
intracranial aneurysms: a minimally invasive approach with advantages for elderly
patients. Age Ageing 1996 September;25(5):372-6.
(97) Laidlaw JD, Siu KH. Poor-grade aneurysmal subarachnoid hemorrhage: Outcome after
treatment with urgent surgery. Neurosurgery 2003;53(6):1275-80.
(98) Inagawa T. Management outcome in the elderly patient following subarachnoid
hemorrhage. J Neurosurg 1993 April;78(4):554-61.
(99) Schuiling WJ, de Weerd AW, Dennesen PJW, Algra A, Rinkel GJE. The simplified
acute physiology score to predict outcome in patients with subarachnoid hemorrhage.
Neurosurgery 2005;57(2):230-5.
(100) Eden SV, Meurer WJ, Sanchez BN, Lisabeth LD, Smith MA, Brown DL, Morgenstern
LB. Gender and ethnic differences in subarachnoid hemorrhage. Neurology 2008
September 2;71(10):731-5.
(101) Nieuwkamp DJ, Vaartjes I, Algra A, Bots ML, Rinkel GJ. Age- and gender-specific
time trend in risk of death of patients admitted with aneurysmal subarachnoid
hemorrhage in the Netherlands. Int J Stroke 2013 March 12;10.
(102) Kissela B, Schneider A, Kleindorfer D, Khoury J, Miller R, Alwell K, Woo D,
Szaflarski J, Gebel J, Moomaw C, Pancioli A, Jauch E, Shukla R, Broderick J. Stroke
in a biracial population: the excess burden of stroke among blacks. Stroke 2004
February;35(2):426-31.
(103) Labovitz DL, Halim AX, Brent B, Boden-Albala B, Hauser WA, Sacco RL.
Subarachnoid hemorrhage incidence among Whites, Blacks and Caribbean Hispanics:
the Northern Manhattan Study. Neuroepidemiology 2006;26(3):147-50.
(104) Kennedy BS, Kasl SV, Brass LM, Vaccarino V. Trends in hospitalized stroke for
blacks and whites in the United States, 1980-1999. Neuroepidemiology 2002
May;21(3):131-41.
(105) Ayala C, Greenlund KJ, Croft JB, Keenan NL, Donehoo RS, Giles WH, Kittner SJ,
Marks JS. Racial/ethnic disparities in mortality by stroke subtype in the United States,
1995-1998. Am J Epidemiol 2001 December 1;154(11):1057-63.
173
(106) Schievink WI, Riedinger M, Jhutty TK, Simon P. Racial disparities in subarachnoid
hemorrhage mortality: Los Angeles County, California, 1985-1998. Neuroepidemiology
2004 November;23(6):299-305.
(107) Broderick JP, Brott T, Tomsick T, Huster G, Miller R. The risk of subarachnoid and
intracerebral hemorrhages in blacks as compared with whites. N Engl J Med 1992
March 12;326(11):733-6.
(108) Bruno A, Carter S, Qualls C, Nolte KB. Incidence of spontaneous subarachnoid
hemorrhage among Hispanics and non-Hispanic whites in New Mexico. Ethn Dis
1997;7(1):27-33.
(109) Nguyen-Huynh MN, Johnston SC. Regional variation in hospitalization for stroke
among Asians/Pacific Islanders in the United States: a nationwide retrospective cohort
study. BMC Neurol 2005 November 10;5:21.:21.
(110) Kissela B, Schneider A, Kleindorfer D, Khoury J, Miller R, Alwell K, Woo D,
Szaflarski J, Gebel J, Moomaw C, Pancioli A, Jauch E, Shukla R, Broderick J. Stroke
in a biracial population: the excess burden of stroke among blacks. Stroke 2004
February;35(2):426-31.
(111) Eden SV, Morgenstern LB, Sekar P, Moomaw CJ, Haverbusch M, Flaherty ML,
Broderick JP, Woo D. The role of race in time to treatment after subarachnoid
hemorrhage. Neurosurgery 2007 May;60(5):837-42.
(112) Wong GK, Teoh J, Chan EK, Ng SC, Poon WS. Intracranial aneurysm size responsible
for spontaneous subarachnoid haemorrhage. Br J Neurosurg 2013 February;27(1):34-9.
(113) Nieuwkamp DJ, Setz LE, Algra A, Linn FH, de Rooij NK, Rinkel GJ. Changes in case
fatality of aneurysmal subarachnoid haemorrhage over time, according to age, sex, and
region: a meta-analysis. Lancet Neurol 2009 July;8(7):635-42.
(114) Rosen D, Novakovic R, Goldenberg FD, Huo D, Baldwin ME, Frank JI, Rosengart AJ,
Macdonald RL. Racial differences in demographics, acute complications, and outcomes
in patients with subarachnoid hemorrhage: a large patient series. J Neurosurg 2005
July;103(1):18-24.
(115) Zacharia BE, Grobelny BT, Komotar RJ, Sander CE, Mocco J. The influence of race on
outcome following subarachnoid hemorrhage. J Clin Neurosci 2010 January;17(1):34-
7.
(116) Eden SV, Morgenstern LB, Sekar P, Moomaw CJ, Haverbusch M, Flaherty ML,
Broderick JP, Woo D. The role of race in time to treatment after subarachnoid
hemorrhage. Neurosurgery 2007 May;60(5):837-43.
(117) Huguet N, Kaplan MS, Feeny D. Socioeconomic status and health-related quality of life
among elderly people: results from the Joint Canada/United States Survey of Health.
Soc Sci Med 2008 February;66(4):803-10.
174
(118) McGrail KM, van DE, Ross NA, Sanmartin C. Income-related health inequalities in
Canada and the United States: a decomposition analysis. Am J Public Health 2009
October;99(10):1856-63.
(119) Reinier K, Thomas E, Andrusiek DL, Aufderheide TP, Brooks SC, Callaway CW, Pepe
PE, Rea TD, Schmicker RH, Vaillancourt C, Chugh SS. Socioeconomic status and
incidence of sudden cardiac arrest. CMAJ 2011 October 18;183(15):1705-12.
(120) Addo J, Ayerbe L, Mohan KM, Crichton S, Sheldenkar A, Chen R, Wolfe CD,
McKevitt C. Socioeconomic status and stroke: an updated review. Stroke 2012
April;43(4):1186-91.
(121) Cox AM, McKevitt C, Rudd AG, Wolfe CD. Socioeconomic status and stroke. Lancet
Neurol 2006 February;5(2):181-8.
(122) Jakovljevic D, Sivenius J, Sarti C, Torppa J, Mahonen M, Immonen-Raiha P, Kaarsalo
E, Alhainen K, Tuomilehto J, Puska P, Salomaa V. Socioeconomic inequalities in the
incidence, mortality and prognosis of subarachnoid hemorrhage: The FINMONICA
Stroke Register. Cerebrovascular Diseases 2001;12(1):7-13.
(123) Rosen DS, MacDonald RL. Subarachnoid hemorrhage grading scales - A systematic
review. Neurocritical Care 2005;2(2):110-8.
(124) Jaja BN, Cusimano MD, Etminan N, Hanggi D, Hasan D, Ilodigwe D, Lantigua H, Le
RP, Lo B, Louffat-Olivares A, Mayer S, Molyneux A, Quinn A, Schweizer TA, Schenk
T, Spears J, Todd M, Torner J, Vergouwen MD, Wong GK, Singh J, Macdonald RL.
Clinical prediction models for aneurysmal subarachnoid hemorrhage: a systematic
review. Neurocrit Care 2013 February;18(1):143-53.
(125) Chiang VL, Claus EB, Awad IA. Toward more rational prediction of outcome in
patients with high-grade subarachnoid hemorrhage. Neurosurgery 2000
January;46(1):28-35.
(126) Giraldo EA, Mandrekar JN, Rubin MN, Dupont SA, Zhang Y, Lanzino G, Wijdicks
EF, Rabinstein AA. Timing of clinical grade assessment and poor outcome in patients
with aneurysmal subarachnoid hemorrhage. J Neurosurg 2012 April 27.
(127) Nishioka H. Report on the cooperative study of intracranial aneurysms and
subarachnoid hemorrhage. Section VII. I. Evaluation of the conservative management
of ruptured intracranial aneurysms. J Neurosurg 1966 November;25(5):574-92.
(128) Hunt WE, Hess RM. Surgical risk as related to time of intervention in the repair of
intracranial aneurysms. J Neurosurg 1968 January;28(1):14-20.
(129) Hunt WE, Kosnik EJ. Timing and perioperative care in intracranial aneurysm surgery.
Clin Neurosurg 1974;21:79-89.:79-89.
175
(130) Teasdale GM, Drake CG, Hunt W, Kassell N, Sano K, Pertuiset B, de Villiers JC. A
universal subarachnoid hemorrhage scale: report of a committee of the World
Federation of Neurosurgical Societies. J Neurol Neurosurg Psychiatry 1988
November;51(11):1457.
(131) van Heuven AW, Mees SMD, Algra A, Rinkel GJE. Validation of a prognostic
subarachnoid hemorrhage grading scale derived directly from the Glasgow Coma Scale.
Stroke 2008;39(4):1347-8.
(132) Julien JS, Bandeen-Roche K, Tamargo RJ, Barrow DL, Lawton MT, Fernandez-
Miranda JC, Dumont AS, Kassell NE, Solomon RA. Validation of an aneurysmal
subarachnoid hemorrhage grading scale in 1532 consecutive patients. Neurosurgery
2008;63(2):204-11.
(133) Degen LA, Dorhout Mees SM, Algra A, Rinkel GJ. Interobserver variability of grading
scales for aneurysmal subarachnoid hemorrhage. Stroke 2011 June;42(6):1546-9.
(134) Starke RM, Komotar RJ, Kim GH, Kellner CP, Otten ML, Hahn DK, Schmidt JM,
Sciacca RR, Mayer SA, Connolly ES. Evaluation of a revised Glasgow Coma Score
scale in predicting long-term outcome of poor grade aneurysmal subarachnoid
hemorrhage patients. Journal of Clinical Neuroscience 2009;16(7):894-9.
(135) Takagi K, Tamura A, Nakagomi T, Nakayama H, Gotoh O, Kawai K, Taneda M, Yasui
N, Hadeishi H, Sano K. How should a subarachnoid hemorrhage grading scale be
determined? A combinatorial approach based solely on the Glasgow Coma Scale. J
Neurosurg 1999 April;90(4):680-7.
(136) Taylor CL, Yuan Z, Selman WR, Ratcheson RA, Rimm AA. Cerebral Arterial
Aneurysm Formation and Rupture in 20,767 Elderly Patients - Hypertension and Other
Risk-Factors. J Neurosurg 1995;83(5):812-9.
(137) Toftdahl DB, Torp-Pedersen CT, Engel UH, Strandgaard S, Jepersen B. [Hypertension
and left ventricular hypertrophy in patients with spontaneous subarachnoid
hemorrhage]. Ugeskr Laeger 1997 March 3;159(10):1434-7.
(138) Juvela S. Prehemorrhage risk factors for fatal intracranial aneurysm rupture. Stroke
2003 August;34(8):1852-7.
(139) Kassell NF, Torner JC, Haley EC, Jr., Jane JA, Adams HP, Kongable GL. The
International Cooperative Study on the Timing of Aneurysm Surgery. Part 1: Overall
management results. J Neurosurg 1990 July;73(1):18-36.
(140) Pobereskin LH. Influence of premorbid factors on survival following subarachnoid
hemorrhage. J Neurosurg 2001 October;95(4):555-9.
(141) Rosen DS, MacDonald RL. Grading of subarachnoid hemorrhage: modification of the
world World Federation of Neurosurgical Societies scale on the basis of data for a large
series of patients. Neurosurgery 2004 March;54(3):566-75.
176
(142) Le Roux PD, Elliott JP, Newell DW, Grady MS, Winn HR. Predicting outcome in
poor-grade patients with subarachnoid hemorrhage: a retrospective review of 159
aggressively managed cases. J Neurosurg 1996 July;85(1):39-49.
(143) Germanson TP, Lanzino G, Kongable GL, Torner JC, Kassell NF. Risk classification
after aneurysmal subarachnoid hemorrhage. Surg Neurol 1998 February;49(2):155-63.
(144) Eskesen V, Rosenorn J, Schmidt K, Ronde F. Pre-existing arterial hypertension in
subarachnoid haemorrhage: an unfavourable prognostic factor. Br J Neurosurg
1987;1(4):455-61.
(145) Pobereskin LH. Incidence and outcome of subarachnoid haemorrhage: a retrospective
population based study. J Neurol Neurosurg Psychiatry 2001 March;70(3):340-3.
(146) Claassen J, Vu A, Kreiter KT, Kowalski RG, Du EY, Ostapkovich N, Fitzsimmons BF,
Connolly ES, Mayer SA. Effect of acute physiologic derangements on outcome after
subarachnoid hemorrhage. Crit Care Med 2004 March;32(3):832-8.
(147) Inagawa T, Tokuda Y, Ohbayashi N, Takaya M, Moritake K. Study of aneurysmal
subarachnoid hemorrhage in Izumo City, Japan. Stroke 1995 May;26(5):761-6.
(148) Roos YB, Beenen LF, Groen RJ, Albrecht KW, Vermeulen M. Timing of surgery in
patients with aneurysmal subarachnoid haemorrhage: rebleeding is still the major cause
of poor outcome in neurosurgical units that aim at early surgery. J Neurol Neurosurg
Psychiatry 1997 October;63(4):490-3.
(149) Taki W, Sakai N, Suzuki H. Determinants of poor outcome after aneurysmal
subarachnoid hemorrhage when both clipping and coiling are available: Prospective
Registry of Subarachnoid Aneurysms Treatment (PRESAT) in Japan. World Neurosurg
2011 November;76(5):437-45.
(150) Zacharia BE, Ducruet AF, Hickman ZL, Grobelny BT, Fernandez L, Schmidt JM,
Narula R, Ko LN, Cohen ME, Mayer SA, Connolly ES. Renal Dysfunction as an
Independent Predictor of Outcome After Aneurysmal Subarachnoid Hemorrhage A
Single-Center Cohort Study. Stroke 2009;40(7):2375-81.
(151) Niskanen MM, Hernesniemi JA, Vapalahti MP, Kari A. One-year outcome in early
aneurysm surgery: Prediction of outcome. Acta neurochirurgica 1993;123(1-2):25-32.
(152) Gruber A, Reinprecht A, Illievich UM, Fitzgerald R, Dietrich W, Czech T, Richling B.
Extracerebral organ dysfunction and neurologic outcome after aneurysmal
subarachnoid hemorrhage. Crit Care Med 1999 March;27(3):505-14.
(153) Bruder N, Rabinstein A. Cardiovascular and pulmonary complications of aneurysmal
subarachnoid hemorrhage. Neurocrit Care 2011 September;15(2):257-69.
177
(154) Machiel PC, Algra A, Velthuis BK, Rinkel GJ. Relation between size of aneurysms and
risk of rebleeding in patients with subarachnoid haemorrhage. Acta Neurochir (Wien )
2006 December;148(12):1277-9.
(155) Guo LM, Zhou HY, Xu JW, Wang Y, Qiu YM, Jiang JY. Risk factors related to
aneurysmal rebleeding. World Neurosurg 2011 September;76(3-4):292-8.
(156) Lindvall P, Runnerstam M, Birgander R, Koskinen LO. The Fisher grading correlated
to outcome in patients with subarachnoid haemorrhage. Br J Neurosurg 2009
April;23(2):188-92.
(157) Turck N, Vutskits L, Sanchez-Pena P, Robin X, Hainard A, Gex-Fabry M, Fouda C,
Bassem H, Mueller M, Lisacek F, Puybasset L, Sanchez JC. A multiparameter panel
method for outcome prediction following aneurysmal subarachnoid hemorrhage.
Intensive Care Medicine 2010;36(1):107-15.
(158) Toledo P, Rios PM, Ledezma A, Sanchis A, Alen JF, Lagares A. Predicting the
Outcome of Patients With Subarachnoid Hemorrhage Using Machine Learning
Techniques. Ieee Transactions on Information Technology in Biomedicine
2009;13(5):794-801.
(159) Lagares A, Gomez PA, Alen JF, Lobato RD, Rivas JJ, Alday R, Campollo J, de la
Camara AG. A comparison of different grading scales for predicting outcome after
subarachnoid haemorrhage. Acta Neurochir (Wien ) 2005 January;147(1):5-16.
(160) Frontera JA, Claassen J, Schmidt JM, Wartenberg KE, Temes R, Connolly ES,
MacDonald RL, Mayer SA. Prediction of symptomatic vasospasm after subarachnoid
hemorrhage: The modified Fisher scale. Neurosurgery 2006;59(1):21-6.
(161) Fisher CM, Kistler JP, Davis JM. Relation of cerebral vasospasm to subarachnoid
hemorrhage visualized by computerized tomographic scanning. Neurosurgery 1980
January;6(1):1-9.
(162) Wilson DA, Nakaji P, Abla AA, Uschold TD, Fusco DJ, Oppenlander ME,
Albuquerque FC, McDougall CG, Zabramski JM, Spetzler RF. A simple and
quantitative method to predict symptomatic vasospasm after subarachnoid hemorrhage
based on computed tomography: beyond the Fisher scale. Neurosurgery 2012
October;71(4):869-75.
(163) Katada K, Kanno T, Sano H, Shibata T, Shah MY. [Computed tomography of ruptured
intracranial aneurysm in acute stage (author's transl)]. No Shinkei Geka 1977
August;5(9):955-63.
(164) Pierot L, Cognard C, Anxionnat R, Ricolfi F. Ruptured intracranial aneurysms: factors
affecting the rate and outcome of endovascular treatment complications in a series of
782 patients (CLARITY study). Radiology 2010 September;256(3):916-23.
178
(165) Ogilvy CS, Cheung AC, Mitha AP, Hoh BL, Carter BS. Outcomes for surgical and
endovascular management of intracranial aneurysms using a comprehensive grading
system. Neurosurgery 2006 November;59(5):1037-42.
(166) Shiue I, Arima H, Hankey GJ, Anderson CS. Location and size of ruptured intracranial
aneurysm and serious clinical outcomes early after subarachnoid hemorrhage: a
population-based study in Australasia. Cerebrovasc Dis 2011;31(6):573-9.
(167) Salary M, Quigley MR, Wilberger JE, Jr. Relation among aneurysm size, amount of
subarachnoid blood, and clinical outcome. J Neurosurg 2007 July;107(1):13-7.
(168) van der Bilt IAC, Hasan D, Vandertop WP, Wilde AAM, Algra A, Visser FC, Rinkel
GJE. Impact of cardiac complications on outcome after aneurysmal subarachnoid
hemorrhage A meta-analysis. Neurology 2009;72(7):635-42.
(169) Lanterna LA, Ruigrok Y, Alexander S, Tang J, Biroli F, Dunn LT, Poon S. Meta-
analysis of APOE genotype and subarachnoid hemorrhage - Clinical outcome and
delayed ischemia. Neurology 2007;69(8):766-75.
(170) Kaneda K, Fujita M, Yamashita S, Kaneko T, Kawamura Y, Izumi T, Tsuruta R,
Kasaoka S, Maekawa T. Prognostic value of biochemical markers of brain damage and
oxidative stress in post-surgical aneurysmal subarachnoid hemorrhage patients. Brain
Res Bull 2010 January 15;81(1):173-7.
(171) Wertli M, Bachmann LM, Weiner SS, Brunner F. Prognostic factors in complex
regional pain syndrome 1: a systematic review. J Rehabil Med 2013 March
6;45(3):225-31.
(172) Veerbeek JM, Kwakkel G, van Wegen EE, Ket JC, Heymans MW. Early prediction of
outcome of activities of daily living after stroke: a systematic review. Stroke 2011
May;42(5):1482-8.
(173) Whiteley W, Chong WL, Sengupta A, Sandercock P. Blood markers for the prognosis
of ischemic stroke: a systematic review. Stroke 2009 May;40(5):e380-e389.
(174) Mallett S, Timmer A, Sauerbrei W, Altman DG. Reporting of prognostic studies of
tumour markers: a review of published articles in relation to REMARK guidelines. Br J
Cancer 2010 January 5;102(1):173-80.
(175) Counsell C, Dennis M. Systematic review of prognostic models in patients with acute
stroke. Cerebrovasc Dis 2001;12(3):159-70.
(176) Jacob M, Lewsey JD, Sharpin C, Gimson A, Rela M, van der Meulen JH. Systematic
review and validation of prognostic models in liver transplantation. Liver Transpl 2005
July;11(7):814-25.
(177) Perel P, Edwards P, Wentz R, Roberts I. Systematic review of prognostic models in
traumatic brain injury. BMC Med Inform Decis Mak 2006 November 14;6:38.:38.
179
(178) Mushkudiani NA, Hukkelhoven CW, Hernandez AV, Murray GD, Choi SC, Maas AI,
Steyerberg EW. A systematic review finds methodological improvements necessary for
prognostic models in determining traumatic brain injury outcomes. J Clin Epidemiol
2008 April;61(4):331-43.
(179) Lingsma HF, Roozenbeek B, Steyerberg EW, Murray GD, Maas AI. Early prognosis in
traumatic brain injury: from prophecies to predictions. Lancet Neurol 2010
May;9(5):543-54.
(180) Ingui BJ, Rogers MA. Searching for clinical prediction rules in MEDLINE. J Am Med
Inform Assoc 2001 July;8(4):391-7.
(181) Ogilvy CS, Carter BS. A proposed comprehensive grading system to predict outcome
for surgical management of intracranial aneurysms. Neurosurgery 1998
May;42(5):959-68.
(182) Ewout W.Steyerberg. Clinical Prediction models A practical approach to
Development, Validation and Updating. Springer; 2010.
(183) Groenwold RH, Donders AR, Roes KC, Harrell FE, Jr., Moons KG. Dealing with
missing outcome data in randomized trials and observational studies. Am J Epidemiol
2012 February 1;175(3):210-7.
(184) Babyak MA. What you see may not be what you get: a brief, nontechnical introduction
to overfitting in regression-type models. Psychosom Med 2004 May;66(3):411-21.
(185) Naval NS, Kowalski RG, Chang TR, Caserta F, Carhuapoma JR, Tamargo RJ. The
SAH Score: A Comprehensive Communication Tool. J Stroke Cerebrovasc Dis 2013
October 5;(13):10.
(186) Degos V, Apfel CC, Sanchez P, Colonne C, Renuit I, Clarencon F, Nouet A, Boch AL,
Pourmohamad T, Kim H, Gourraud PA, Young WL, Puybasset L. An admission
bioclinical score to predict 1-year outcomes in patients undergoing aneurysm coiling.
Stroke 2012 May;43(5):1253-9.
(187) Karamanakos PN, von Und Zu FM, Bendel S, Huttunen T, Kurki M, Hernesniemi J,
Ronkainen A, Rinne J, Jaaskelainen JE, Koivisto T. Risk factors for three phases of 12-
month mortality in 1657 patients from a defined population after acute aneurysmal
subarachnoid hemorrhage. World Neurosurg 2012 December;78(6):631-9.
(188) O'kelly CJ, Kulkarni AV, Austin PC, Wallace MC, Urbach D. The impact of
therapeutic modality on outcomes following repair of ruptured intracranial aneurysms:
an administrative data analysis. Clinical article. J Neurosurg 2010 October;113(4):795-
801.
(189) Coghlan LA, Hindman BJ, Bayman EO, Banki NM, Gelb AW, Todd MM, Zaroff JG.
Independent Associations Between Electrocardiographic Abnormalities and Outcomes
180
in Patients With Aneurysmal Subarachnoid Hemorrhage Findings From the
Intraoperative Hypothermia Aneurysm Surgery Trial. Stroke 2009;40(2):412-8.
(190) Langham J, Reeves BC, Lindsay KW, van der Meulen JH, Kirkpatrick PJ, Gholkar AR,
Molyneux AJ, Shaw DM, Copley L, Browne JP. Variation in outcome after
subarachnoid hemorrhage: a study of neurosurgical units in UK and Ireland. Stroke
2009 January;40(1):111-8.
(191) Guresir E, Beck J, Vatter H, Setzer M, Gerlach R, Seifert V, Raabe A. Subarachnoid
Hemorrhage and Intracerebral Hematoma: Incidence, Prognostic Factors, and Outcome.
Neurosurgery 2008;63(6):1088-93.
(192) Jagger J, Torner JC, Kassell NF. Neurologic assessment of subarachnoid hemorrhage in
a large patient series. Surg Neurol 1989 November;32(5):327-33.
(193) Cooper H, Patall EA. The relative benefits of meta-analysis conducted with individual
participant data versus aggregated data. Psychol Methods 2009 June;14(2):165-76.
(194) Abo-Zaid G, Sauerbrei W, Riley RD. Individual participant data meta-analysis of
prognostic factor studies: state of the art? BMC Med Res Methodol 2012 April
24;12:56. doi: 10.1186/1471-2288-12-56.:56-12.
(195) Riley RD, Lambert PC, Abo-Zaid G. Meta-analysis of individual participant data:
rationale, conduct, and reporting. BMJ 2010 February 5;340:c221. doi:
10.1136/bmj.c221.:c221.
(196) Roozenbeek B, Lingsma HF, Lecky FE, Lu J, Weir J, Butcher I, McHugh GS, Murray
GD, Perel P, Maas AI, Steyerberg EW. Prediction of outcome after moderate and
severe traumatic brain injury: external validation of the International Mission on
Prognosis and Analysis of Clinical Trials (IMPACT) and Corticoid Randomisation
After Significant Head injury (CRASH) prognostic models. Crit Care Med 2012
May;40(5):1609-17.
(197) Maas AI, Marmarou A, Murray GD, Teasdale SG, Steyerberg EW. Prognosis and
clinical trial design in traumatic brain injury: the IMPACT study. J Neurotrauma 2007
February;24(2):232-8.
(198) Maas AI, Steyerberg EW, Marmarou A, McHugh GS, Lingsma HF, Butcher I, Lu J,
Weir J, Roozenbeek B, Murray GD. IMPACT recommendations for improving the
design and analysis of clinical trials in moderate to severe traumatic brain injury.
Neurotherapeutics 2010 January;7(1):127-34.
(199) Ali M, Bath PM, Curram J, Davis SM, Diener HC, Donnan GA, Fisher M, Gregson
BA, Grotta J, Hacke W, Hennerici MG, Hommel M, Kaste M, Marler JR, Sacco RL,
Teal P, Wahlgren NG, Warach S, Weir CJ, Lees KR. The Virtual International Stroke
Trials Archive. Stroke 2007 June;38(6):1905-10.
181
(200) van den Bergh WM, Algra A, van KF, Dirven CM, van GJ, Vermeulen M, Rinkel GJ.
Magnesium sulfate in aneurysmal subarachnoid hemorrhage: a randomized controlled
trial. Stroke 2005 May;36(5):1011-5.
(201) Wong GKC, Poon WS, Chan MTV, Boet R, Gin T, Ng SCP, Zee BCY. Intravenous
Magnesium Sulphate for Aneurysmal Subarachnoid Hemorrhage (IMASH) A
Randomized, Double-Blinded, Placebo-Controlled, Multicenter Phase III Trial. Stroke
2010;41(5):921-6.
(202) Macdonald RL, Kassell NF, Mayer S, Ruefenacht D, Schmiedek P, Weidauer S, Frey
A, Roux S, Pasqualin A. Clazosentan to overcome neurological ischemia and infarction
occurring after subarachnoid hemorrhage (CONSCIOUS-1): randomized, double-blind,
placebo-controlled phase 2 dose-finding trial. Stroke 2008 November;39(11):3015-21.
(203) Todd MM, Hindman BJ, Clarke WR, Torner JC. Mild intraoperative hypothermia
during surgery for intracranial aneurysm. N Engl J Med 2005 January 13;352(2):135-
45.
(204) Johnston SC, McDougall CG, Gholkar A. The MAPS trial—Matrix and Platinum
Science-clinical trials. Colorado Spring, Colorado 2011.
(205) Etminan N, Beseoglu K, Eicker SO, Turowski B, Steiger HJ, Hanggi D. Prospective,
Randomized, Open-Label Phase II Trial on Concomitant Intraventricular Fibrinolysis
and Low-Frequency Rotation After Severe Subarachnoid Hemorrhage. Stroke 2013
June 4.
(206) Noble AJ, Schenk T. Which Variables Help Explain the Poor Health-Related Quality of
Life After Subarachnoid Hemorrhage? A Meta-analysis. Neurosurgery 2010;66(4):772-
83.
(207) Helbok R, Kurtz P, Vibbert M, Schmidt MJ, Fernandez L, Lantigua H, Ostapkovich
ND, Connolly SE, Lee K, Claassen J, Mayer SA, Badjatia N. Early neurological
deterioration after subarachnoid haemorrhage: risk factors and impact on outcome. J
Neurol Neurosurg Psychiatry 2013 March;84(3):266-70.
(208) Smith ML, Abrahams JM, Chandela S, Smith MJ, Hurst RW, Le Roux PD.
Subarachnoid hemorrhage on computed tomography scanning and the development of
cerebral vasospasm: the Fisher grade revisited. Surgical Neurology 2005;63(3):229-35.
(209) Reilly C, Amidei C, Tolentino J, Jahromi BS, MacDonald RL. Clot volume and
clearance rate as independent predictors of vasospasm after aneurysmal subarachnoid
hemorrhage. J Neurosurg 2004 August;101(2):255-61.
(210) Tirschwell DL, Longstreth WT. Validating administrative data in stroke research.
Stroke 2002;33(10):2465-70.
(211) Kokotailo RA, Hill MD. Coding of stroke and stroke risk factors using international
classification of diseases, revisions 9 and 10. Stroke 2005 August;36(8):1776-81.
182
(212) McHugh GS, Butcher I, Steyerberg EW, Lu J, Mushkudiani N, Marmarou A, Maas AI,
Murray GD. Statistical approaches to the univariate prognostic analysis of the IMPACT
database on traumatic brain injury. J Neurotrauma 2007 February;24(2):251-8.
(213) Mushkudiani NA, Engel DC, Steyerberg EW, Butcher I, Lu J, Marmarou A, Slieker F,
McHugh GS, Murray GD, Maas AI. Prognostic value of demographic characteristics in
traumatic brain injury: results from the IMPACT study. J Neurotrauma 2007
February;24(2):259-69.
(214) Frank E Harrell. Regression Modeling strategies: with application to linear, logistic and
survival analysis. 2001. New York, Springer.
(215) Murray GD, Butcher I, McHugh GS, Lu J, Mushkudiani NA, Maas AI, Marmarou A,
Steyerberg EW. Multivariable prognostic analysis in traumatic brain injury: results
from the IMPACT study. J Neurotrauma 2007 February;24(2):329-37.
(216) Statacorp. Stata 12 Base Reference Manual. College Station, Texas: Stata Press; 2011.
(217) Rosenlrn J, Eskesen V, Schmidt K. Clinical features and outcome in females and males
with ruptured intracranial saccular aneurysms. Br J Neurosurg 1993;7(3):287-90.
(218) Kongable GL, Lanzino G, Germanson TP, Truskowski LL, Alves WM, Torner JC,
Kassell NF, Spetzler RF, Zabramski J, Culicchia F, Carter LP, Feinberg W, Urbina C,
Lopez L, Brown D, Tallman D, Selman WR, Harrington F, Warf B, Barnett GJ, Little J,
Palmer J, Campbell RL, Shapiro S, Farlow M, Kay S, Horner T, Leipzig T, Redelman
K, Muizelaar JP, Turner R, Kamesh W, Bouma G, Reyes RD, Rosenbaum D,
Klonowski E, Mohr G, Bojanowski M, Bernier G, Duquette P, LaPlante P, Murali R,
Crisafulli C, Merkel C, Rosner MK, Mahaley MS, Aronin PA, Morawetz RB, Rosner
S, BrockSillers RN, Lamond RG, Marshall L, Rothrock J, Benedict B, Moore S, Young
B, Dempsey R, Phillips R, Salcman M, Rigamonti D, Kulick S, Powers S, Baker ML,
Flamm ES, Reivich M, Kushner M, Farber R, Parrott S, Grotta J, Kopanicky D, Yatsur
F, DeGrapa T, Bratina P, Heilbrun HP, Call R, Hood R, Goodman S, Sunderland P,
Haley EC, Parks K, Winn HR, Grady S, Mayberg M, Burnal C, Grubb RL, Coxe W,
Landou W, Rich K, Levin A, Hobbs V, Adams HP, Drake CG, Dyken ML, Frankowski
RF, Kurtzke JF, Weir BKA, Hayley EC, Cail WS, Elder LJ, Henry ML, Truskowski
LL, Marron JA, Polin AS, Lightfoot AL. Gender-related differences in aneurysmal
subarachnoid hemorrhage. J Neurosurg 1996;84(1):43-8.
(219) Horiuchi T, Tanaka Y, Hongo K. Sex-related differences in patients treated surgically
for aneurysmal subarachnoid hemorrhage. Neurol Med Chir (Tokyo) 2006
July;46(7):328-32.
(220) Caranci F, Briganti F, Cirillo L, Leonardi M, Muto M. Epidemiology and genetics of
intracranial aneurysms. Eur J Radiol 2013 February 8;(13):10.
(221) Sadasivan C, Fiorella DJ, Woo HH, Lieber BB. Physical factors effecting cerebral
aneurysm pathophysiology. Ann Biomed Eng 2013 July;41(7):1347-65.
183
(222) Mitchell P, Birchall D, Mendelow AD. Blood pressure, fatigue, and the pathogenesis of
aneurysmal subarachnoid hemorrhage. Surg Neurol 2006 December;66(6):574-80.
(223) Juvela S, Siironen J, Kuhmonen J. Hyperglycemia, excess weight, and history of
hypertension as risk factors for poor outcome and cerebral infarction after aneurysmal
subarachnoid hemorrhage. J Neurosurg 2005 June;102(6):998-1003.
(224) van den Bergh WM, Algra A, Elias R, Rinkel GJ. Extent of atherosclerosis and
prognosis of patients with aneurysmal subarachnoid haemorrhage. Acta Neurochir
(Wien ) 2006 October;148(10):1065-9.
(225) De Marchis GM, Lantigua H, Schmidt JM, Lord AS, Velander AJ, Fernandez A, Falo
MC, Agarwal S, Connolly ES, Jr., Claassen J, Mayer SA. Impact of premorbid
hypertension on haemorrhage severity and aneurysm rebleeding risk after subarachnoid
haemorrhage. J Neurol Neurosurg Psychiatry 2014 January;85(1):56-9.
(226) Roos EJ, Rinkel GJE, Velthuis BK, Algra A. The relation between aneurysm size and
outcome in patients with subarachnoid hemorrhage. Neurology 2000;54(12):2334-6.
(227) van Norden AG, van Dijk GW, van Huizen MD, Algra A, Rinkel GJ. Interobserver
agreement and predictive value for outcome of two rating scales for the amount of
extravasated blood after aneurysmal subarachnoid haemorrhage. J Neurol 2006
September;253(9):1217-20.
(228) Haug T, Sorteberg A, Finset A, Lindegaard KF, Lundar T, Sorteberg W. Cognitive
Functioning and Health-Related Quality of Life 1 Year After Aneurysmal
Subarachnoid Hemorrhage in Preoperative Comatose Patients (Hunt and Hess Grade V
Patients). Neurosurgery 2010;66(3):475-84.
(229) Navi BB, Kamel H, McCulloch CE, Nakagawa K, Naravetla B, Moheet AM, Wong C,
Johnston SC, Hemphill JC, III, Smith WS. Accuracy of neurovascular fellows'
prognostication of outcome after subarachnoid hemorrhage. Stroke 2012
March;43(3):702-7.
(230) Steyerberg EW, Pencina MJ, Lingsma HF, Kattan MW, Vickers AJ, Van CB.
Assessing the incremental value of diagnostic and prognostic markers: a review and
illustration. Eur J Clin Invest 2012 February;42(2):216-28.
(231) Steyerberg EW, Harrell FE, Jr., Borsboom GJ, Eijkemans MJ, Vergouwe Y, Habbema
JD. Internal validation of predictive models: efficiency of some procedures for logistic
regression analysis. J Clin Epidemiol 2001 August;54(8):774-81.
(232) Saposnik G, Cote R, Mamdani M, Raptis S, Thorpe KE, Fang J, Redelmeier DA,
Goldstein LB. JURaSSiC: accuracy of clinician vs risk score prediction of ischemic
stroke outcomes. Neurology 2013 July 30;81(5):448-55.
184
(233) Alter DA, Chong A, Austin PC, Mustard C, Iron K, Williams JI, Morgan CD, Tu JV,
Irvine J, Naylor CD. Socioeconomic status and mortality after acute myocardial
infarction. Ann Intern Med 2006 January 17;144(2):82-93.
(234) Shi L, Macinko J, Starfield B, Xu J, Politzer R. Primary care, income inequality, and
stroke mortality in the United States: a longitudinal analysis, 1985-1995. Stroke 2003
August;34(8):1958-64.
(235) Saposnik G, Jeerakathil T, Selchen D, Baibergenova A, Hachinski V, Kapral MK.
Socioeconomic status, hospital volume, and stroke fatality in Canada. Stroke 2008
December;39(12):3360-6.
(236) Cesaroni G, Agabiti N, Forastiere F, Perucci CA. Socioeconomic differences in stroke
incidence and prognosis under a universal healthcare system. Stroke 2009
August;40(8):2812-9.
(237) Arrich J, Lalouschek W, Mullner M. Influence of socioeconomic status on mortality
after stroke: retrospective cohort study. Stroke 2005 February;36(2):310-4.
(238) Kapral MK, Fang J, Chan C, Alter DA, Bronskill SE, Hill MD, Manuel DG, Tu JV,
Anderson GM. Neighborhood income and stroke care and outcomes. Neurology 2012
September 18;79(12):1200-7.
(239) Kapral MK, Wang H, Mamdani M, Tu JV. Effect of socioeconomic status on treatment
and mortality after stroke. Stroke 2002 January;33(1):268-73.
(240) Heeley EL, Wei JW, Carter K, Islam MS, Thrift AG, Hankey GJ, Cass A, Anderson
CS. Socioeconomic disparities in stroke rates and outcome: pooled analysis of stroke
incidence studies in Australia and New Zealand. Med J Aust 2011 July 4;195(1):10-4.
(241) Skolarus LE, Meurer WJ, Burke JF, Prvu BJ, Lisabeth LD. Effect of insurance status on
postacute care among working age stroke survivors. Neurology 2012 May
15;78(20):1590-5.
(242) Wong KY, Wong SY, Fraser HW, Ersoy Y, Ogston S, Wolfson D, Malek M, Struthers
AD, Tunstall-Pedoe H, MacWalter RS. Effect of social deprivation on mortality and the
duration of hospital stay after a stroke. Cerebrovasc Dis 2006;22(4):251-7.
(243) Claritas. Demographic update methodology. Claritas Inc; 2004.
(244) Canadian Institute for Health Information. Reducing gaps in health: A focus on
socioeconomic status in urban Canada. Canadian Institute for Health Information;
2008.
(245) Moy E, Barrett M, Ho K. Potentially preventable hospitalizations - United States, 2004-
2007. MMWR Surveill Summ 2011 January 14;60 Suppl:80-3.:80-3.
185
(246) Quan H, Sundararajan V, Halfon P, Fong A, Burnand B, Luthi JC, Saunders LD, Beck
CA, Feasby TE, Ghali WA. Coding algorithms for defining comorbidities in ICD-9-
CM and ICD-10 administrative data. Med Care 2005 November;43(11):1130-9.
(247) El-Sayed AM, Ziewacz JE, Davis MC, Lau D, Siddiqi HK, Zamora-Berridi GJ,
Sullivan SE. Insurance status and inequalities in outcomes after neurosurgery. World
Neurosurg 2011 November;76(5):459-66.
(248) Lasser KE, Himmelstein DU, Woolhandler S. Access to care, health status, and health
disparities in the United States and Canada: results of a cross-national population-based
survey. Am J Public Health 2006 July;96(7):1300-7.
(249) Gorey KM. Breast cancer survival in Canada and the USA: meta-analytic evidence of a
Canadian advantage in low-income areas. Int J Epidemiol 2009 December;38(6):1543-
51.
(250) Zhang-Salomons J, Qian H, Holowaty E, Mackillop WJ. Associations between
socioeconomic status and cancer survival: choice of SES indicator may affect results.
Ann Epidemiol 2006 July;16(7):521-8.
(251) Chan L, Wang H, Terdiman J, Hoffman J, Ciol MA, Lattimore BF, Sidney S,
Quesenberry C, Lu Q, Sandel ME. Disparities in outpatient and home health service
utilization following stroke: results of a 9-year cohort study in Northern California. PM
R 2009 November;1(11):997-1003.
(252) Nguyen LL, Barshes NR. Analysis of large databases in vascular surgery. J Vasc Surg
2010 September;52(3):768-74.
(253) Woodworth GF, Baird CJ, Garces-Ambrossi G, Tonascia J, Tamargo RJ. Inaccuracy of
the administrative database: comparative analysis of two databases for the diagnosis
and treatment of intracranial aneurysms. Neurosurgery 2009 August;65(2):251-6.
(254) Kleindorfer D, Lindsell C, Alwell KA, Moomaw CJ, Woo D, Flaherty ML, Khatri P,
Adeoye O, Ferioli S, Kissela BM. Patients living in impoverished areas have more
severe ischemic strokes. Stroke 2012 August;43(8):2055-9.
(255) Shavers VL. Measurement of socioeconomic status in health disparities research. J Natl
Med Assoc 2007 September;99(9):1013-23.
(256) Wolfe CD, Rudd AG, Howard R, Coshall C, Stewart J, Lawrence E, Hajat C, Hillen T.
Incidence and case fatality rates of stroke subtypes in a multiethnic population: the
South London Stroke Register. J Neurol Neurosurg Psychiatry 2002
February;72(2):211-6.
(257) Lisabeth LD, Risser JM, Brown DL, Al-Senani F, Uchino K, Smith MA, Garcia N,
Longwell PJ, McFarling DA, Al-Wabil A, Akuwumi O, Moye LA, Morgenstern LB.
Stroke burden in Mexican Americans: the impact of mortality following stroke. Ann
Epidemiol 2006 January;16(1):33-40.
186
(258) Willey JZ, Rodriguez CJ, Moon YP, Paik MC, Di Tullio MR, Homma S, Sacco RL,
Elkind MS. Coronary death and myocardial infarction among Hispanics in the Northern
Manhattan Study: exploring the Hispanic paradox. Ann Epidemiol 2012
May;22(5):303-9.
(259) Hummer RA, Powers DA, Pullum SG, Gossman GL, Frisbie WP. Paradox found
(again): infant mortality among the Mexican-origin population in the United States.
Demography 2007 August;44(3):441-57.
(260) Brown HL, Chireau MV, Jallah Y, Howard D. The "Hispanic paradox": an
investigation of racial disparity in pregnancy outcomes at a tertiary care medical center.
Am J Obstet Gynecol 2007 August;197(2):197.
(261) Borrell LN, Lancet EA. Race/ethnicity and all-cause mortality in US adults: revisiting
the Hispanic paradox. Am J Public Health 2012 May;102(5):836-43.
(262) Morales LS, Lara M, Kington RS, Valdez RO, Escarce JJ. Socioeconomic, cultural, and
behavioral factors affecting Hispanic health outcomes. J Health Care Poor
Underserved 2002 November;13(4):477-503.
(263) Bateman BT, Schumacher HC, Boden-Albala B, Berman MF, Mohr JP, Sacco RL,
Pile-Spellman J. Factors associated with in-hospital mortality after administration of
thrombolysis in acute ischemic stroke patients: an analysis of the nationwide inpatient
sample 1999 to 2002. Stroke 2006 February;37(2):440-6.
(264) Nasr DM, Brinjikji W, Cloft HJ, Rabinstein AA. Racial and Ethnic Disparities in the
Use of Intravenous Recombinant Tissue Plasminogen Activator and Outcomes for
Acute Ischemic Stroke. J Stroke Cerebrovasc Dis 2011 December 9.
(265) McNaughton H, Feigin V, Kerse N, Barber PA, Weatherall M, Bennett D, Carter K,
Hackett M, Anderson C. Ethnicity and functional outcome after stroke. Stroke 2011
April;42(4):960-4.
(266) Shen JJ, Lu M. Changes in patterns of racial disparities in attending low-mortality
hospitals and outcomes among patients with stroke. Ethn Dis 2011;21(2):135-41.
(267) Paek MS, Lim JW. Factors associated with health care access and outcome. Soc Work
Health Care 2012;51(6):506-30.
(268) Tam AKH, Ilodigwe D, Mocco J, Mayer S, Kassell N, Ruefenacht D, Schmiedek P,
Weidauer S, Pasqualin A, MacDonald RL. Impact of Systemic Inflammatory Response
Syndrome on Vasospasm, Cerebral Infarction, and Outcome After Subarachnoid
Hemorrhage: Exploratory Analysis of CONSCIOUS-1 Database. Neurocritical Care
2010;13(2):182-9.
(269) Pereira AR, Sanchez-Pena P, Biondi A, Sourour N, Boch AL, Colonne C, Lejean L,
Abdennour L, Puybasset L. Predictors of 1-year outcome after coiling for poor-grade
subarachnoid aneurysmal hemorrhage. Neurocritical Care 2007;7(1):18-26.
187
(270) Macdonald RL, Jaja B, Cusimano MD, Etminan N, Hanggi D, Hasan D, Ilodigwe D,
Lantigua H, Le Roux P, Lo B, Louffat-Olivares A, Mayer S, Molyneux A, Quinn A,
Schweizer TA, Schenk T, Spears J, Todd M, Torner J, Vergouwen MDI, Wong GKC,
Singh J. SAHIT Investigators-on the Outcome of Some Subarachnoid Hemorrhage
Clinical Trials. Translational Stroke Research 2013;4(3):286-96.
(271) Pickard JD, Murray GD, Illingworth R, Shaw MD, Teasdale GM, Foy PM, Humphrey
PR, Lang DA, Nelson R, Richards P, . Effect of oral nimodipine on cerebral infarction
and outcome after subarachnoid haemorrhage: British aneurysm nimodipine trial. BMJ
1989 March 11;298(6674):636-42.
(272) Adler NE, Boyce WT, Chesney MA, Folkman S, Syme SL. Socioeconomic inequalities
in health. No easy solution. JAMA 1993 June 23;269(24):3140-5.
(273) Cruz-Flores S, Rabinstein A, Biller J, Elkind MS, Griffith P, Gorelick PB, Howard G,
Leira EC, Morgenstern LB, Ovbiagele B, Peterson E, Rosamond W, Trimble B,
Valderrama AL. Racial-ethnic disparities in stroke care: the American experience: a
statement for healthcare professionals from the American Heart Association/American
Stroke Association. Stroke 2011 July;42(7):2091-116.
(274) Zhan C, Miller MR. Administrative data based patient safety research: a critical review.
Qual Saf Health Care 2003 December;12 Suppl 2:ii58-63.:ii58-ii63.
(275) Julie Richards, Ann Brown, Craig Horman. The data quality study of the Canadian
Discharge Abstract database. 2001.
(276) Schmutzhard E, Rabinstein AA. Spontaneous subarachnoid hemorrhage and glucose
management. Neurocrit Care 2011 September;15(2):281-6.
(277) Scaravilli V, Tinchero G, Citerio G. Fever management in SAH. Neurocrit Care 2011
September;15(2):287-94.
(278) Naidech AM, Jovanovic B, Wartenberg KE, Parra A, Ostapkovich N, Connolly ES,
Mayer SA, Commichau C. Higher hemoglobin is associated with improved outcome
after subarachnoid hemorrhage. Crit Care Med 2007 October;35(10):2383-9.
(279) Steyerberg EW, Mushkudiani N, Perel P, Butcher I, Lu J, Mchugh GS, Murray GD,
Marmarou A, Roberts I, Habbema JDF, Maas AIR. Predicting outcome after traumatic
brain injury: Development and international validation of prognostic scores based on
admission characteristics. Plos Medicine 2008;5(8):1251-61.
(280) Langagergaard V, Palnum KH, Mehnert F, Ingeman A, Krogh BR, Bartels P, Johnsen
SP. Socioeconomic differences in quality of care and clinical outcome after stroke: a
nationwide population-based study. Stroke 2011 October;42(10):2896-902.
(281) McGinn TG, Guyatt GH, Wyer PC, Naylor CD, Stiell IG, Richardson WS. Users'
guides to the medical literature: XXII: how to use articles about clinical decision rules.
Evidence-Based Medicine Working Group. JAMA 2000 July 5;284(1):79-84.
188
(282) Reilly BM, Evans AT. Translating clinical research into clinical practice: impact of
using prediction rules to make decisions. Ann Intern Med 2006 February 7;144(3):201-
9.
(283) Toll DB, Janssen KJ, Vergouwe Y, Moons KG. Validation, updating and impact of
clinical prediction rules: a review. J Clin Epidemiol 2008 November;61(11):1085-94.
(284) Citerio G, Gaini SM, Tomei G, Stocchetti N. Management of 350 aneurysmal
subarachnoid hemorrhages in 22 Italian neurosurgical centers. Intensive Care Medicine
2007;33(9):1580-6.
(285) Rosengart AJ, Huo DZ, Tolentino J, Novakovic RL, Frank JI, Goldenberg FD,
MacDonald RL. Outcome in patients with subarachnoid hemorrhage treated with
antiepileptic drugs. J Neurosurg 2007;107(2):253-60.
(286) Maas AI, Menon DK, Lingsma HF, Pineda JA, Sandel ME, Manley GT. Re-orientation
of clinical research in traumatic brain injury: report of an international workshop on
comparative effectiveness research. J Neurotrauma 2012 January 1;29(1):32-46.
(287) Wilson DA, Nakaji P, Albuquerque FC, McDougall CG, Zabramski JM, Spetzler RF.
Time course of recovery following poor-grade SAH: the incidence of delayed
improvement and implications for SAH outcome study design. J Neurosurg 2013
September;119(3):606-12.
(288) Powell J, Kitchen N, Heslin J, Greenwood R. Psychosocial outcomes at three and nine
months after good neurological recovery from aneurysmal subarachnoid haemorrhage:
predictors and prognosis. J Neurol Neurosurg Psychiatry 2002 June;72(6):772-81.
(289) Powell J, Kitchen N, Heslin J, Greenwood R. Psychosocial outcomes at 18 months after
good neurological recovery from aneurysmal subarachnoid haemorrhage. J Neurol
Neurosurg Psychiatry 2004 August;75(8):1119-24.
189
Appendices
Appendix A: PUBMED search history for systematic review of prognostic models and studies
identifying independent predictors of poor outcome in SAH
Search Add to
builder Query
Items
found Time
#40 Add Search (((((((((((((subarachnoid hemorrhage) OR
subarachnoid he*morrhage) OR subarachnoid hemorrhag*)
OR subarachnoid haemorrhag*) OR (((subarachnoid
hemorrhage[All Fields] OR "subarachnoid
hemorrhage"[MeSH Terms]) OR "subarachnoid
hemorrhage"[MeSH Terms]))) OR subarachnoid bleed*) OR
subarachnoid blood*) OR intracranial aneurysm*) OR
(("intracranial aneurysm"[MeSH Terms] OR intracranial
aneurysm[Text Word]))) OR intracranial bleed*)) AND
((((((((decision tree*) OR "Decision Trees"[Mesh]) OR
recursive partition*) OR classification tree*) OR regression
tree*) OR multiparameter panel*) OR CART) OR "Models,
Statistical"[Mesh]))) OR ((((((((((((subarachnoid hemorrhage)
OR subarachnoid he*morrhage) OR subarachnoid
hemorrhag*) OR subarachnoid haemorrhag*) OR
(((subarachnoid hemorrhage[All Fields] OR "subarachnoid
hemorrhage"[MeSH Terms]) OR "subarachnoid
hemorrhage"[MeSH Terms]))) OR subarachnoid bleed*) OR
subarachnoid blood*) OR intracranial aneurysm*) OR
(("intracranial aneurysm"[MeSH Terms] OR intracranial
aneurysm[Text Word]))) OR intracranial bleed*)) AND
(((((((((((((((nomogram*) OR "Nomograms"[Mesh]) OR
grad* scal*) OR grad* system*) OR grad* scheme*) OR
scor*) OR risk*) OR "Risk Factors"[Mesh]) OR prognos*)
OR prognos* model*) OR predict* model*) OR prognos*
index) OR predict*) OR validat*) OR "Decision Support
Techniques"[Mesh]))
3300 09:41:49
#39 Add Search (((((((((((subarachnoid hemorrhage) OR subarachnoid
he*morrhage) OR subarachnoid hemorrhag*) OR
subarachnoid haemorrhag*) OR (((subarachnoid
hemorrhage[All Fields] OR "subarachnoid
hemorrhage"[MeSH Terms]) OR "subarachnoid
hemorrhage"[MeSH Terms]))) OR subarachnoid bleed*) OR
subarachnoid blood*) OR intracranial aneurysm*) OR
(("intracranial aneurysm"[MeSH Terms] OR intracranial
aneurysm[Text Word]))) OR intracranial bleed*)) AND
(((((((((((((((nomogram*) OR "Nomograms"[Mesh]) OR
grad* scal*) OR grad* system*) OR grad* scheme*) OR
3048 09:41:28
190
Search Add to
builder Query
Items
found Time
scor*) OR risk*) OR "Risk Factors"[Mesh]) OR prognos*)
OR prognos* model*) OR predict* model*) OR prognos*
index) OR predict*) OR validat*) OR "Decision Support
Techniques"[Mesh])
#38 Add Search (((((((((((subarachnoid hemorrhage) OR subarachnoid
he*morrhage) OR subarachnoid hemorrhag*) OR
subarachnoid haemorrhag*) OR (((subarachnoid
hemorrhage[All Fields] OR "subarachnoid
hemorrhage"[MeSH Terms]) OR "subarachnoid
hemorrhage"[MeSH Terms]))) OR subarachnoid bleed*) OR
subarachnoid blood*) OR intracranial aneurysm*) OR
(("intracranial aneurysm"[MeSH Terms] OR intracranial
aneurysm[Text Word]))) OR intracranial bleed*)) AND
((((((((decision tree*) OR "Decision Trees"[Mesh]) OR
recursive partition*) OR classification tree*) OR regression
tree*) OR multiparameter panel*) OR CART) OR "Models,
Statistical"[Mesh])
512 09:41:01
#37 Add Search ((((((((((((((nomogram*) OR "Nomograms"[Mesh])
OR grad* scal*) OR grad* system*) OR grad* scheme*) OR
scor*) OR risk*) OR "Risk Factors"[Mesh]) OR prognos*)
OR prognos* model*) OR predict* model*) OR prognos*
index) OR predict*) OR validat*) OR "Decision Support
Techniques"[Mesh]
1267319 09:40:33
#36 Add Search "Decision Support Techniques"[Mesh] 56659 09:39:30
#35 Add Search validat* 280137 09:39:15
#34 Add Search predict* 959037 09:38:57
#33 Add Search prognos* index 48972 09:38:18
#32 Add Search predict* model* 198745 09:38:04
#31 Add Search prognos* model* 31288 09:37:52
#30 Add Search prognos* 545231 09:37:39
#29 Add Search "Risk Factors"[Mesh] 529255 09:34:19
#28 Add Search risk* 1516099 09:34:03
#26 Add Search scor* 529082 09:33:40
#25 Add Search grad* scheme* 4071 09:33:19
#24 Add Search grad* system* 90962 09:32:29
#23 Add Search grad* scal* 27593 09:31:56
#22 Add Search "Nomograms"[Mesh] 1176 09:31:35
#21 Add Search nomogram* 4844 09:30:59
#20 Add Search (((((((decision tree*) OR "Decision Trees"[Mesh]) OR
recursive partition*) OR classification tree*) OR regression
267630 09:30:28
191
Search Add to
builder Query
Items
found Time
tree*) OR multiparameter panel*) OR CART) OR "Models,
Statistical"[Mesh]
#19 Add Search "Models, Statistical"[Mesh] 252407 09:29:12
#18 Add Search CART 3707 09:28:48
#17 Add Search multiparameter panel* 99 09:28:36
#16 Add Search regression tree* 1454 09:28:22
#15 Add Search classification tree* 840 09:28:07
#14 Add Search recursive partition* 1168 09:27:54
#13 Add Search "Decision Trees"[Mesh] 8480 09:27:41
#12 Add Search decision tree* 10797 09:27:27
#11 Add Search (((((((((subarachnoid hemorrhage) OR subarachnoid
he*morrhage) OR subarachnoid hemorrhag*) OR
subarachnoid haemorrhag*) OR (((subarachnoid
hemorrhage[All Fields] OR "subarachnoid
hemorrhage"[MeSH Terms]) OR "subarachnoid
hemorrhage"[MeSH Terms]))) OR subarachnoid bleed*) OR
subarachnoid blood*) OR intracranial aneurysm*) OR
(("intracranial aneurysm"[MeSH Terms] OR intracranial
aneurysm[Text Word]))) OR intracranial bleed*
47074 09:26:39
#10 Add Search intracranial bleed* 1214 09:22:23
#9 Add Search ("intracranial aneurysm"[MeSH Terms] OR
intracranial aneurysm[Text Word])
21374 09:22:01
#8 Add Search intracranial aneurysm* 22202 09:21:29
#7 Add Search subarachnoid blood* 327 09:21:11
#6 Add Search subarachnoid bleed* 239 09:20:54
#5 Add Search ((subarachnoid hemorrhage[All Fields] OR
"subarachnoid hemorrhage"[MeSH Terms]) OR
"subarachnoid hemorrhage"[MeSH Terms])
22351 09:20:14
#4 Add Search subarachnoid haemorrhag* 3500 09:19:48
#3 Add Search subarachnoid hemorrhag* 20832 09:19:27
#2 Add Search subarachnoid he*morrhage 31248 09:19:01
#1 Add Search subarachnoid hemorrhage 22351 09:18:34
192
Appendix B: Copyright licenses
WOLTERS KLUWER HEALTH LICENSE
TERMS AND CONDITIONS
Oct 09, 2013
This is a License Agreement between Blessing Jaja ("You") and Wolters Kluwer Health
("Wolters Kluwer Health") provided by Copyright Clearance Center ("CCC"). The license
consists of your order details, the terms and conditions provided by Wolters Kluwer Health,
and the payment terms and conditions.
All payments must be made in full to CCC. For payment instructions, please see information listed at the bottom of
this form.
License Number 3237701304553
License date Sep 28, 2013
Licensed content publisher Wolters Kluwer Health
Licensed content publication Stroke
Licensed content title Effect of Socioeconomic Status on Inpatient Mortality and Use of Postacute Care
After Subarachnoid Hemorrhage
Licensed content author Blessing N.R. Jaja, Gustavo Saposnik, Rosane Nisenbaum, Tom A. Schweizer, Deven
Reddy, Kelvin E. Thorpe, R. Loch Macdonald
Licensed content date Oct 1, 2013
Volume Number 44
Issue Number 10
Type of Use Dissertation/Thesis
Requestor type Individual
Author of this Wolters Kluwer
article
Yes
Title of your thesis / dissertation Prognostic factors in Aneurysmal Subarachnoid hemorrhage
Expected completion date Aug 2014
Estimated size(pages) 150
Billing Type Invoice
Billing address Division of Neurosurgery,
St. Michael's Hospital
Toronto, ON M5B 1W8
Canada
Total 0.00 USD
193
SPRINGER ORDER
TERMS AND CONDITIONS
Oct 09, 2013
This is an Agreement between ("You") and Springer ("Springer"). It consists of your order details, the terms and conditions
provided by Springer, and the payment terms and conditions.
Order Number 500795462
Order date Sep 28, 2013
Licensed content publisher Springer
Licensed content publication Neurocritical Care
Licensed content title Clinical Prediction Models for Aneurysmal Subarachnoid Hemorrhage: A Systematic
Review
Licensed content author Blessing N. R. Jaja
Licensed content date Jan 1, 2012
Volume number 18
Issue number 1
Type of Use Thesis/Dissertation
Portion Full text
Number of copies 1
Author of this Springer article Yes and you are a contributor of the new work
Order reference number
Title of your thesis / dissertation Prognostic factors in Aneurysmal Subarachnoid hemorrhage
Expected completion date Aug 2014
Estimated size(pages) 150
Total 0.00 CAD
194
Request to reprint the following
Racial/ethnic differences in inpatient mortality and use of institutional postacute care following subarachnoid hemorrhage
Authors: Blessing N. R. Jaja, M.B.B.S., M.Sc., Gustavo Saposnik, M.D., M.Sc., F.R.C.P.C., Rosane Nisenbaum, Ph.D.,
Benjamin W. Y. Lo, M.D., M.Sc., F.R.C.S.C., Tom A. Schweizer, Ph.D., Kevin E. Thorpe, M.Math., Ph.D., and R. Loch
Macdonald, M.D., Ph.D., F.R.C.S.
Publication: Journal of Neurosurgery
Month: Dec
Year: 2013
Volume: 119
Issue: 6
Pages: 1627-32
Permission Category:
Original Author Seeking Permission to Use Content in an AANS Publication
Reprint Material Specifics:
ALL
Media Type: Electronic
Media: Annotation/animation
Language: English (as originally published)
Title and Edition of New Work: Prognostic factors in Aneurysmal subarachnoid hemorrhage
Anticipated Republication Date: 2014
Quantity: 1
Publisher of New Work: Jaja BN (Primary author of the article)
Permission to reproduce the requested material is granted to the original author without charge by the copyright owner, AANS,
provided that full acknowledgment is given to Journal of Neurosurgery.
Best of luck with your thesis!
Gillian