-
8/13/2019 dc12-2158.full
1/7
-
8/13/2019 dc12-2158.full
2/7
review boards of the University of Califor-nia at San Francisco
and Davis and theUniversity of Michigan.
Diabetes At each study visit , diabetes classi cationwas based
on fasting glucose level$ 126mg/dL, antidiabetic medication use,
orself-reports of a physician diagnosis atany study visit prior to
dementia/CIND,death, or last study visit. Diabetic medi-cations
were recorded at every study visitby direct visual inspection of
medicationsand classi ed using the Centers for Dis-ease Control
Ambulatory Care Drug Da-tabase (http://www2.cdc.gov/drugs/ ).Given
the advanced age of the cohort,most, if not all, cases were
probably type2 diabetes (16). Hereafter, we will use theterm
diabetes to refer to type 2 diabetes.
Dementia and CINDThe classi cations of dementia and CINDwere
determined at all home visits by amultistage assessment protocol,
whichhas been described extensively elsewhere(24). In brief, at
each visit, two cognitivescreening tests wereused to determine
theneed for further neuropsychological eval-uation: the Modi ed
Mini-Mental StateExam (3MSE) (25), a global cognitivefunction test,
and a delayed word recalltrial from the Spanish English
VerbalLearning Test (SEVLT) (26), a word-listlearning and memory
test. At baseline, a
participant was referred for further evalu-ation if his or her
score on either test fellbelow the 20th percentile. At follow-up,
aparticipant was referred for a neuropsy-chological test battery
and a standardneuropsychological examination by a ger-iatrician if
his or her follow-up score de-clined from the baseline score by
morethan eight points on 3MSE or more thanthree points on SEVLT and
the score fellbelow the 20th percentile. A team of neu-rologists
and a neuropsychologist re-viewed all potential dementia and
CINDcases and classi ed participants as de-mented, CIND, or
cognitively normal.Standard diagnostic criteria were appliedfor
dementia (DSM-IV) (27), Alzheimerdisease (National Institute of
Neurologicand Communicative Disorders andStroke-Alzheimer Disease
and RelatedDisorders Association) (28), and vasculardementia
(California Alzheimers DiseaseDiagnostic and Treatment Centers)
(29).Dementia and CIND cases were referredfor magnetic resonance
imaging. For par-ticipants who died during the study pe-riod
without a previous diagnosis of
dementia or CIND, dementia diagnoseswere also ascertained from
death certi -cates based on the following causes of death listed
anywhere on the death certif-icate: dementia in Alzheimer disease,
vas-cular dementia, other dementia, orunspeci ed dementia. For this
analysis,dementia and CIND were combined intoone outcome,
dementia/CIND.
MortalityMortality ascertainment included inter-views with
family members to track par-ticipants who could not be reached
forannual study visits or interim 6-monthphone calls, online
surveillance of deathnotices, and review of the Social Security
Death Index, the National Death Index,and vital statistics data les
from the stateof California. Mortality surveillance isongoing, but
this analysis is limited todeaths that occurred during active
follow-up for dementia/CIND (1998 2007). Wehad complete or partial
social security numbers on most (80%) of the deceasedand obtained
death certi cates and causeof death for 93.1% of deceased
partici-pants. We classi ed cause of death usingICD-10.
Other variables At the baseline interview, participantsreported
their age, sex, years of education,country of birth, whether or not
they had a regular medical doctor (as a marker
of access to medical care), smokingstatus,and any alcohol use.
As a marker of physical activity, participants were askedto
classify their usual outdoor walkingpace as never walks
outdoors/unable towalk, easy pace, or brisk pace.
Depressivesymptoms were measured by the Centerfor Epidemiologic
Studies DepressionScale, a widely used scale (range 0 60)(30).
Fasting blood samples were takenat annual study visits. Fasting
glucosewas measured with the Cobas MiraChemistry Analyzer (Roche
DiagnosticsCorporation, Indianapolis, IN). Fastinginsulin was
measured using a double-antibody radioimmunoassay using 125
I-labeled human insulin tracer (LincoResearch, St. Charles, MO), a
guineapig antiporcine insulin rst antibody (MichiganDiabetes
Research and TrainingCenter, Ann Arbor, MI), and a goat antiguinea
pig g -globulin (Antibodies Incor-porated, Davis, CA) and
standardizedagainst the Human Insulin InternationalReference
Preparation for Insulin. Sittingsystolic
anddiastolicbloodpressuremeas-urements were taken with an
automatic
digital blood pressure monitor twice at a10-min interval and
averaged. Hyperten-sion was based on measured systolic
bloodpressure ($ 140 mmHg), self-report of aphysician diagnosis,
and/or antihyperten-sive medication use. Waist circumferencewas
measured at the level of the umbilicusat midrespiration with the
participantstanding erect. History of stroke was basedon
self-report of a physician diagnosis andhospitalization.
Statistical analysisThe objective of the analysis was toexamine
the association of diabetes withincidence of dementia/CIND and to
ac-count for the competing risk of mortality. We rst compared
baseline descriptivecharacteristics by treated and
untreateddiabetes status (ever vs. never duringstudy) with ANOVA
for continuous var-iables and x2 tests for categorical
varia-bles.
We evaluated the association of trea-ted and untreated diabetes
with incidenceof dementia/CIND using competing riskregression
models with the method pro-posed by Fine and Gray (31).
Participantswere observed from study entry until theoccurrence of
dementia/CIND (the eventof interest), death (the competing
event),or censoring (last date of contact). Weused time-dependent
diabetes as the ex-posure variable for all models to captureall
diabetes cases that occurred prior to
diagnosis of dementia, death, or censor-ing and time-dependent
variables for co-variates to re ect changes in valuesthroughout the
study period. Because of the strong association between dementiaand
age, for all models, we used age atdiagnosis, death, or censoring
as the time-scale and adjusted for baseline age (32).
Our competing risk regression ap-proach accounts for the fact
that individ-uals who die prior to developingdementia/CIND will
never develop de-mentia/CIND. Thus, the association be-tween
diabetes and dementia/CINDdepends on the association between
di-abetes and death (33 35). Like the stan-dard Coxregression
model, this approachmeasures the association of diabetes withrisk
of dementia/CIND with a hazard ratio(HR). Since previous studies of
this issuehave not accounted for the competingrisk of death, we
also speci ed Cox re-gression models, where participants whodied
were censored at age at death, to ex-amine how taking into account
the com-peting risk of death in uenced riskestimates.
2 DIABETES CARE care.diabetesjournals.org
Diabetes and risk of dementia
http://www2.cdc.gov/drugs/http://www2.cdc.gov/drugs/
-
8/13/2019 dc12-2158.full
3/7
-
8/13/2019 dc12-2158.full
4/7
antidiabetic medications remained rela-tively constant
throughout follow-up(64.7% in year 1 up to 69% in year 7).
More participants with diabetes died
(n = 182, 26.9%) than participants with-out diabetes (n = 179,
19.0%). In a Coxmodel, treated and untreated diabeteswere
associated with increased risk of death after adjustment for sex,
education,time-dependent waist circumference, andtime-dependent
stroke (HR 2.15 [95% CI1.58 2.95]; 2.12 [1.65 2.73]). Therewere
more deaths among participantswith incident dementia/CIND (n =
63,39.6%) than among those without de-mentia/CIND (n = 298, 20.4%).
Incidentdementia/CIND was associated with anincreased risk of death
in a Cox modeladjusted for the same covariates (2.48[1.75
3.51]).
Table 2 shows the HRs relating time-dependent diabetes to the
incidence of dementia/CIND from competing riskmodels and Cox
models. In both typesof models, treateddiabetes was associatedwith
higher incidence of dementia/CINDthan no diabetes. Untreated
diabetes wasonly associated with higher incidence of dementia/CIND
in Cox models. Compar-ison of dementia/CIND risk in treated
tountreated diabetes was not signi cant
(HR 1.32 [95% CI 0.80 2.19], P =0.28). In model 2, adjustment
for time-dependent waist circumference, a poten-tial confounder of
the association between
diabetes and dementia/CIND risk, in-creased the HRs for treated
diabetes anduntreateddiabetes by 10 and 15%, respec-tively,
compared with the base model(model 1), which adjusted for only
age(as timescale), sex, andyears of education.In model 3 (fully
adjusted model), addi-tion of time-dependent stroke modestly
decreased the HR for treated diabetes by 9% and untreated diabetes
by 10%, butthe association with treated diabetes re-mained strong
and statistically signi cant.Time-dependent stroke did not modify
the association between diabetes and de-mentia/CIND (interaction
between strokeand diabetes P = 0.97; data not shown).Compared with
the Cox regression model3, accounting for competing risk of deathin
model 3 attenuated the HR for treateddiabetes by 14% andfor
untreated diabetesby 18%.
Figure 2 displays the estimated cu-mulative incidence functions
for demen-tia/CIND by diabetes status from the fully
adjustedcompeting riskregression model(model 3). The graph
demonstrates thatthe incidence of dementia/CIND was
highest among those with treated diabe-tes, followed by people
untreated for di-abetes and then those without diabetes.
CONCLUSIONS d In this 10-yearpopulation-based study, we found
thatthose with treated diabetes had a twofoldincreased risk of
dementia/CIND amongolder Mexican Americans, even after ac-counting
for the competing risk of mor-tality and changes over time in risk
factorexposures. Those with treated diabetes inour sample have
higher glucose and in-sulin, hypertension, and more
comorbidcardiovascular disease and may havemore severe diabetes
than those whowere untreated. This may explain why their dementia
risk is higher, given thatthe risk of death is similar in both
groups.However, comparison of dementia/CINDrisk in treated to
untreated diabetes wasnot statistically signi cant (HR treated
vs.untreated: 1.32, P = 0.28). Standard Coxmodels that do not
incorporate adjust-ment for competing risk consistently provided a
larger estimate of effect.
Previous epidemiologic studies innon-Hispanic white populations
havealso found that diabetes is associatedwith a twofold increased
risk of dementia(2 13). Most research on the associationbetween
diabetes and dementia has beenconducted in non-Hispanic white
popu-lations, where diabetes burden is lowerthan among Mexican
Americans (16).
To our knowledge, this is the rst study to evaluate the
associationbetweendiabe-tes and dementia or CIND among olderMexican
Americans.
In addition to studying a differentethnic group, the current
study differsfrom previous studies because the analy-sis takes into
account the competing riskof death. It is well established that
di-abetes is associated with higher mortality rates (16), and
previous studies have alsoshown that cognitive decline (22,23)
isassociated withan increased riskof death.Diabetes-related
mortality differentially in uences the number of dementia/ CIND
cases that occur. Since diabetes isassociated with death, a
standard Coxmodel, which ignores the competingrisk of death,
mayresult in an overestima-tion of the effect of diabetes on
incidenceof dementia/CIND (34,37). Although theCox model treats
deaths as censored ob-servations and removes individuals fromthe
denominator for the hazard rate atdeath, the competing risk model
treatsdeath as a competing risk by retainingthose who die without
dementia in the
Table 1 d Baseline characteristics of participants by diabetes
status (ever vs. never) duringstudy ( N = 1,617)
VariableDiabetes treated
(n = 940)Diabetes untreated
(n = 155)No diabetes
(n = 522) P value
Age (years) 69.6 (6.4) 69.7 (6.3) 70.7 (7.1) 0.010Male sex 44.4
47.1 40.1 0.117
Education (years) 7.4 (5.4) 7.4 (5.5) 7.4 (5.3) 0.99U.S. born
55.8 49.7 46.3 0.0025Regular medical doctor 91.3 88.4 86.9
0.040Health insurance 92.7 89.0 90.0 0.17Smoking status 0.0013
Never smoked 41.7 45.2 48.0Former smoker 48.7 47.7 38.8Current
smoker 9.6 7.1 13.2
Waist circumference (inches). 40 male, 35 female 63.8 47.1 40.5
, 0.001
Fasting glucose (mg/dL) 151.8 (57.8) 119.3 (40.5) 92.1 (11.0) ,
0.001Fasting glucose $ 126 mg/dL 58.8 23.2 0 , 0.001Fasting insulin
(mIU/mL) 13.8 (13.4) 13.1 (10.8) 10.5 (20.8) 0.0025Hypertension
73.0 67.7 57.1 , 0.001Stroke 11.4 6.5 6.2 0.0019Myocardial
infarction 12.5 9.7 5.6 , 0.001Congestive heart failure 3.9 3.9 1.7
0.030Intermittent claudication 12.1 7.9 6.3 , 0.001Kidney disease
13.9 7.8 5.5 , 0.001
Continuous variables are displayed as mean (SD) and categorical
variables are displayed as column %.P values comparing
characteristics between those with and without diabetes are two
sided.
4 DIABETES CARE care.diabetesjournals.org
Diabetes and risk of dementia
-
8/13/2019 dc12-2158.full
5/7
denominator for the hazard rate. In effect,this approach models
these individuals asno longer at risk for dementia/CIND.
Thisresults in a lower HR than the standardCox model. The competing
risk regres-
sionmodel produces estimates that re ectthe actual incidence of
dementia/CINDamong people with diabetes comparedwith those without
diabetes. This ap-proach may be useful for clinical predic-tions
and for predictions of futuredementia/CIND incidence in the
popula-tion. In this population, accounting forthe competing risk
of death altered the
fully adjusted risk estimates for treatedand untreated diabetes
by 14 and 18%,respectively. This change is large enough($ 10%) to
be considered an importantconfounder by conventional confounder
identi cation criteria (38).These ndings have important
impli-cations for prediction of future dementiaincidence among
Mexican Americans andother racial/ethnic groups. It is importantto
consider how future changes in mor-tality rates among people with
diabetesmay affect dementia rates. A recent pub-lication from the
National Health Inter-view Survey reported that mortality
ratesamong adults with diabetes declined by 23% between 1997 and
2006 (20). Thistrend was observed in the overall popula-tion as
well as across age and racial/ethnicsubgroups; the decline was
greatestamong Hispanics (38%). The authorspoint out that the change
in death ratesis likely due to multiple factors, includingimproved
diabetes medical care, treat-ments, and self-management
behaviors.In fact, results from the National Healthand Nutrition
Examination Survey 1999 2008 (39) suggest that blood
pressure,glycemic control, LDL cholesterol, andHBA1c have improved
in Mexican Amer-icans as well as other groups. These im-provements
may propel the decline in
mortality rates observed over the sameperiod. These changes have
importantpublic health implications for future inci-dence of
dementia/CIND in the popula-tion. The potential impact on dementia/
CIND incidence among people with dia-betes will depend on the
factors causingthe decline in the mortality rate. If mor-tality
rates among those with diabetesdecline due to earlier screening and
im-proved management and therefore de-creases in disease
progression andseverity, dementia/CIND rates amongpeople with
diabetes might also decrease.If mortality rates decline among
thosewith diabetes without reducing diseaseseverity, and if more
severe diabetes in-
uences dementia risk, dementia/CINDrates among people with
diabetes couldpotentially increase.
SALSA is a population-based study.The sample is representative
of olderLatinos residing in the Sacramento areain California in
1998 1999 (24). The riskestimates from this study are
generaliz-able to populations with similar charac-teristics,
including similar mortality rates.In this population, diabetes was
associ-ated with over a twofold increased riskof death. Mortality
associated with diabe-tes is slightly higher in our populationthan
in two recent nationally representa-tive studies: the National
Health Inter-view Survey (20) and the CancerPrevention Study-II
(21). In both of these
studies, diabetes was associated with anearly twofold increased
risk of death. A major strength of this study is the
large sample and population-based lon-gitudinal design, which
enabled us tostudy incident dementia/CIND over along time period in
an understudiedethnic group with a high burden of di-abetes. This
is the only population-basedstudy of clinically assessed dementia
inMexican Americans. Because our analysisaccounted for the
competing risk of death, our results can be interpreted asthe
absolute risk of dementia/CINDamong people with diabetes
comparedwith those without, making our resultsrelevant for clinical
decision making andpublic health predictions.
This study also has some limitations. As for any study of older
adults, individ-uals had to survive at least to 60 years of age to
participate in this study. Selectiondueto premature mortality among
peoplewith diabetes before 60 years of age may have occurred and
affected the risk esti-mate in our sample. Mortality was theprimary
source of attrition, but some
Table 2 d HRs from competing risk regression models relating
diabetes and incidence of dementia/CIND
Competing risk models Cox models
Model HR (95% CI) HR (95% CI)Model 1
No diabetes (ref) 1.0 1.0
Diabetes untreated 1.50 (0.94
2.40) 1.93 (1.22
3.07)Diabetes treated 2.06 (1.47 2.91) 2.38 (1.68 3.37)Female
vs. male sex 1.35 (0.98 1.86) 1.22 (0.88 1.68)Education (years)
0.96 (0.93 0.99) 0.97 (0.94 1.00)
Model 2No diabetes (ref) 1.0 1.0Diabetes untreated 1.73 (1.05
2.86) 2.16 (1.33 3.54)Diabetes treated 2.26 (1.57 3.27) 2.66 (1.85
3.83)Female vs. male sex 1.31 (0.93 1.83) 1.14 (0.81 1.61)Education
(years) 0.96 (0.93 0.99) 0.96 (0.93 0.99) Waist circumference
(inches) (TD) 0.97 (0.94 1.00) 0.96 (0.93 1.00)
Model 3No diabetes (ref) 1.0 1.0Diabetes untreated 1.55 (0.93
2.58) 1.88 (1.15 3.07)Diabetes treated 2.05 (1.41 2.97) 2.38 (1.65
3.44)Female vs. male sex 1.30 (0.92 1.82) 1.15 (0.82 1.61)Education
(years) 0.95 (0.92 0.99) 0.96 (0.93 0.99) Waist circumference
(inches) (TD) 0.97 (0.94 1.00) 0.97 (0.93 1.00)Stroke (TD) 2.95
(2.04 4.27) 3.28 (2.28 4.72)
Age adjusted for entry age is the timescale in all models. Ref,
reference; TD, time dependent.
Figure 2 d Cumulative incidence functions for dementia/CIND by
diabetes status, accounting for the competing risk of death, from
competing risk regression model adjusted for sex, years of
education, waist circumference, and stroke.Squares, no diabetes;
circles, diabetes un-treated; triangles, diabetes treated.
care.diabetesjournals.org D IABETES CARE 5
Mayeda and Associates
-
8/13/2019 dc12-2158.full
6/7
-
8/13/2019 dc12-2158.full
7/7
Valley Independent Elders Survey. J AmGeriatr Soc 2009;57:94
100
24. Haan MN, Mungas DM, Gonzalez HM,Ortiz TA, Acharya A, Jagust
WJ. Preva-lence of dementia in older latinos: thein uence of type 2
diabetes mellitus,stroke and genetic factors. J Am GeriatrSoc
2003;51:169 177
25. Teng EL, Chui HC. The Modi ed Mini-Mental State (3MS)
examination. J ClinPsychiatry 1987;48:314 318
26. Gonzalez HM, Mungas D, Reed BR,Marshall S, Haan MN. A new
verballearning andmemory test forEnglish- andSpanish-speaking older
people. J IntNeuropsychol Soc 2001;7:544 555
27. American Phsyciatric Association. Di-agnostic and
Statistical Manual of Mentaldisorders: DSM-IV-TR . 4th ed.
Wash-ington, D.C., American Phsyciatric Asso-ciation, 2000
28. McKhann G, Drachman D, Folstein M,Katzman R, Price D,
Stadlan EM. Clinicaldiagnosis of Alzheimers disease: reportof the
NINCDS-ADRDA Work Groupunder the auspices of Department of Health
and Human Services Task Force
on Alzheimers Disease. Neurology 1984;34:939 944
29. Chui HC, Victoroff JI, Margolin D, Jagust W, Shankle R,
Katzman R. Criteria forthe diagnosis of ischemic vascular de-mentia
proposed by the State of California Alzheimers Disease Diagnostic
and Treat-ment Centers. Neurology 1992;42:473 480
30. Radloff LS. The CES-D Scale: a self-reportdepression scale
for research in the gen-eral population. Appl Psychol Meas
1977;1:385 401
31. Fine JP, Gray RJ. A proportional hazardsmodel for the
subdistribution of a com-peting risk. J Am Stat Assoc
1999;94:496509
32. Pencina MJ, Larson MG, D Agostino RB.Choice of time scale
and its effect on sig-ni cance of predictors in
longitudinalstudies. Stat Med 2007;26:1343 1359
33. Beyersmann J, Dettenkofer M, Bertz H,Schumacher M. A
competing risks analy-sis of bloodstream infection after
stem-celltransplantation using subdistributionhazards and
cause-speci c hazards. StatMed 2007;26:5360 5369
34. Lau B, Cole SR, Gange SJ. Competing riskregression models
for epidemiologic data. Am J Epidemiol 2009;170:244 256
35. Pintilie M. Analysing and interpretingcompeting risk data.
Stat Med 2007;26:1360 1367
36. Wu JH, Haan MN, Liang J, Ghosh D,Gonzalez HM, Herman WH.
Diabetesas a predictor of change in functionalstatus among older
Mexican Americans:a population-based cohort study. DiabetesCare
2003;26:314 319
37. Wolbers M, Koller MT, Witteman JC,Steyerberg EW. Prognostic
models withcompeting risks: methods and applicationto coronary
riskprediction. Epidemiology 2009;20:555 561
38. TongIS, LuY. Identi cation of confoundersin theassessment of
therelationshipbetweenlead exposure and child development. Ann
Epidemiol 2001;11:38 45
39. Ford ES. Trends in the control of riskfactors for
cardiovascular disease amongadults with diagnosed diabetes:
ndingsfrom the National Health and NutritionExamination Survey
1999-2008*. J Di-abetes 2011;3:337 347
care.diabetesjournals.org D IABETES CARE 7
Mayeda and Associates
