1

Using the EQ-5D index score as a predictor of outcomes

in patients with type 2 diabetes.

WORK IN PROGRESS: PLEASE DO NOT QUOTE

Status: In press with Medical Care

Authors: Philip M. Clarke Ph.D1,2 Alison J. Hayes Ph.D1,2, Paul G. Glasziou Ph.D3,

Russell Scott FRACP4, John Simes FRACP1,2 Anthony C. Keech. FRACP1,2

Affiliations: 1. School of Public Health, Edward Ford Building, University of Sydney, Australia.

2. NHMRC Clinical Trials Centre, University of Sydney, NSW, Australia.

3. Department of Primary Care, Headington OX3 7LF University of Oxford, United Kingdom

4. Lipids and Diabetes Research, Christchurch Hospital, Private Bag 4710

Christchurch, New Zealand.

Address for corresponding author: A/Prof Philip Clarke, School of Public Health,

Edward Ford Building (A27), University of Sydney, NSW 2006 Australia; Phone.: +61 2

9036 7138; Fax.: +61 2 9351 7420; Email.: philipc@med.usyd.edu.au.

2

Abstract

Objective: To examine whether index scores based on the EQ-5D, a 5-item generic health

status measure, are an independent predictor of vascular events, other major complications

and mortality in people with Type 2 diabetes and to quantify the relationship between these

scores and future survival.

Subjects: 5-year cohort study involving 7348 patients with type 2 diabetes, aged between

50-75 years who had been recruited to the FIELD study from Australia and New Zealand.

Measures: Multivariate Cox proportional hazard regression models were used to estimate

the hazard ratio associated with index scores derived from the EQ-5D on: (i) cardiovascular

events (including coronary heart disease event, stroke, hospitalization for angina, or

cardiovascular death); (ii) other major diabetes-related complications (heart failure,

amputation, renal dialysis and lower extremity ulcer); and (iii) death from any cause. Life

table methods were used to derive expected survival for patients with different index

scores.

Results: After adjusting for standard risk factors, a 0.1 higher index score (derived from the

United Kingdom algorithm) was associated with an additional 7% (95% CI 4-11%) lower risk

of vascular events, a 13% (CI 09-17%) lower risk of complications and up to 14% (95% CI

8-19%) lower rate of all-cause mortality.

Conclusions: Index scores derived from the EQ-5D are an independent predictor of the risk

of mortality, future vascular events, and other complications in people with type 2 diabetes.

This should be taken into account when extrapolating health outcomes such as Quality

adjusted life years (QALYs).

Keywords: EQ-5D; utility; mortality; survival; diabetes; Quality Adjusted Life Years

Abbreviations: BMI, Body mass index; FIELD, Fenofibrate Intervention and Event Lowering

in Diabetes, QALYS, Quality Adjusted Life Years, HRQL, Health Related Quality of Life.

3

Introduction Quality-adjusted life years (QALYs) are becoming a common metric for quantifying health

outcomes, as they are able to combine measures of both life expectancy and health-related

quality of life (HRQL) into a single summary measure. QALYs are usually calculated by

weighting each year of a patients’ remaining life by a HRQL index, which has been referred

to as a health state utility1, where 1.0 is equivalent to “full health” and 0.0 is equivalent to

being “dead”. Health economists have traditionally used techniques such as the time trade-

off method on patients or the general public (based on questionnaires involving health state

descriptions) in order to estimate preferences for health states. However, it is often difficult

to directly measure health state utilities as the elicitation techniques often require extensive

personal interviews. In practice these values are often obtained from algorithms or value

sets developed for multi-attribute instruments such as the EQ-5D for quantifying HRQL2.

The calculation of QALYs requires either knowledge or prediction of survival and measures

of a health state utility over time. When follow-up data on patients is available, a QALY can

be calculated by multiplying the utility or preference weight associated with each health

state by the duration of time in that state and then summing these quantities over time.

When survival time is censored a method such as quality adjusted survival analysis must

be employed.3 Without follow-up data, extrapolation of both life expectancy and the HRQL

profile is required. In regard to the latter, it may be important to take account of the potential

relationship between current levels of HRQL and future morbidity. For example, if patients

with lower health state utilities were at higher risk of events this would impact on their

expected future QALYs. In a similar fashion index scores from generic HRQL instruments

such as the EQ-5D, which have also been used to construct QALYs, may also be predictive

of future morbidity and mortality. Until recently there has been little empirical evidence of

the association between index scores and subsequent outcomes. Index scores derived

from the EQ-5D have also been shown to be predictive of short-term survival of terminal

cancer patients4 and of one-year survival after coronary revascularisation.5 The degree to

which index scores can be predictive of outcomes over longer periods and for most chronic

diseases has not yet been explored.

4

The primary purpose of this study is to test whether measures of HRQL based on index

scores derived from the EQ-5D are independent predictors of morbidity and mortality of

people with type 2 diabetes. The study is based on Australian and New Zealand patients

with type 2 diabetes who participated in the Fenofibrate Intervention and Event Lowering in

Diabetes (FIELD) study.6,7 A secondary purpose is to calculate estimates of the difference

in life expectancy for different levels of index score.

Methods Background Complications of diabetes which include events such as myocardial infarction, stroke,

amputation and renal failure elevate the risk of mortality and reduce HRQL. These events

are known to be influenced by risk factors such as blood pressure.8 Risk equations are now

available for people with diabetes9,10 and these form the basis of simulation models for

estimating outcomes such as QALYs in health economic evaluations. Typically such

models focus on risk-factors derived from clinical assessment or biochemical measurement,

rather than subjective measures of HRQL for predicting future morbidity and mortality11.

Patients The FIELD study is a double blind, placebo controlled trial performed in 63 centers in

Australia, New Zealand and Finland, details of the design have been published

elsewhere6,7. In brief, 9,795 patients with type 2 diabetes, aged between 50-75 years, were

randomized to fenofibrate 200mg or matching placebo capsules. Patients were recruited

from hospital clinics and community based sources. At entry to the main study all

individuals had an initial plasma total-cholesterol concentration between 116.1 mg/dl and

251.4 mg/dl, plus either a total-cholesterol/ HDL-cholesterol of 4.0 or more, or a plasma

triglyceride concentration of between 88.6 mg/dl and 442.9 mg/dl. Exclusion criteria

included taking lipid-modifying therapies, known renal impairment, chronic liver disease,

symptomatic gallbladder disease, or a cardiovascular event within 3 months of recruitment.

The study design conforms to the guidelines of the Declaration of Helsinki and it was

approved by the local and national ethics committees. All patients gave informed written

consent. The study is registered as an International Standard Randomized Controlled Trial,

ISCTN 64783481.

5

Assessment of health-related quality of life To measure HRQL, the EQ-5D questionnaire was administered to FIELD patients residing

in Australia and New Zealand at a mean (SD) time of 2.87(0.81) years after randomisation

to therapy. The EQ-5D is a multi-attribute instrument with a descriptive system covering five

dimensions of self-reported health (mobility, self-care, usual activity, pain/discomfort,

anxiety and depression) with three levels (no problem, some problem, or extreme

problems).13-15 For example, 12111 stands for a health state where there are “no problems”

in 4 of the 5 dimensions, but “some problems” with self-care. The developers of the EQ-5D

have generated value sets in several countries to calculate a preference-based index for

the 243 health states defined by responses to the five questions of the EQ-5D, using a

scale on which 0.0 represents being dead and 1.0 full health. Values of the index can be

negative for states that are deemed to be “worse than death”: so for example, the minimum

value in the UK based value set2 is -0.59 to represent the worst possible health state (i.e.,

33333). As there is no value set based on Time Trade Off developed for the Australian or

New Zealand population we have used the two most commonly used values currently

available for the EQ-5D. In this study the primary analysis used the EQ-5D value set

developed from a valuation of key health states using the time trade off method conducted

in the United Kingdom2 and in a secondary analysis used valuations of the same health

states based on a value set for the US population.16

A total of 7704 (91.7%) of the 8402 Australian and New Zealand patients participating in the

FIELD study completed at least one EQ-5D. Of these, 356 patients had one or more

missing responses on the EQ-5D questionnaire and hence were excluded from the

analysis. The age, baseline systolic blood pressure, total cholesterol:hdl ratio, HbA1c levels

and BMI of the excluded patients were not significantly different from those remaining in the

analysis population (n=7348). For a small number of patients (<1%), mean imputation was

used to estimate values for missing risk factors.

Assessment of other risk factors

Patients were required to attend scheduled study visits every 4-6 months, and continued to

receive their normal care from their health professionals. Information was recorded on

demographic and clinical characteristics including height, weight, blood pressure, smoking

6

status and pre-existing medical conditions. All blood and urine samples were analyzed in a

single central laboratory in Adelaide, Australia. Values of all risk factors were calculated as

an average of available measures between randomisation and the time of the first

administration of EQ-5D. Each patient’s mean risk factor values were used as covariates in

the statistical analysis, which is comparable with methods used to develop previous

diabetes risk prediction equations.10

Outcomes We focused on three outcomes: (i) cardiovascular events (involving the first of any of a

coronary heart disease event, stroke, hospitalization for angina, or cardiovascular death);

(ii) other major diabetes-related complications (i.e. heart failure, amputation, renal dialysis

and lower extremity ulcer); (iii) death from any cause. A diagnosis of myocardial infarction

required at least two of three criteria: ECG changes, ischaemic symptoms, or raised cardiac

enzymes. A stroke required evidence of onset of focal neurological defects, lasting at least

24 hours, with cerebral imaging to confirm an ischaemic stroke. Congestive heart failure

was ascertained from a hospital admission with a principal diagnosis code of I50 ICD-10.

Angina events were determined from hospital admissions with a pincipal diagnosis of ICD-

10 codes I20.00, I20.80 and I20.90. Amputation included all non-traumatic amputations of

either digit or limb. Renal failure was defined as time of first dialysis, and lower extremity

ulcer was defined by ICD-10 codes E11.73, E11.69 and L97.00.

All major cardiovascular disease events, including deaths and all deaths from other causes

were adjudicated by an outcomes assessment committee blinded to treatment allocation.

The follow-up period for this study was from the time of administration of the first EQ-5D

questionnaire until death, or close out of the FIELD study.

Statistical methods

The statistical analyses were undertaken in STATA 9.217 and were divided into three

phases. Unless otherwise noted all statistical inferences for significance were drawn at the

5% level.

7

Firstly, to visually examine the degree to which risk varies across ranges of index scores,

estimates of Nelson Aalen cumulative hazard were plotted for the following groups (mean of

each group reported in brackets): index score < 0.75 (0.59); index score >=0.75 and index

score < 1.00 (0.81), index score = 1.00 (1.00). Each group represents an approximate

tertile of the data. Standard log-rank methods without adjustment for covariates were used

to test for significant differences between these groups for each specified endpoint.

Multivariate Cox proportional hazard models were used to assess whether index scores

and responses to dimensions were independently predictive of outcomes. This provided a

hazard ratio associated with responses to EQ-5D index scores after controlling for baseline

covariates including age, sex, duration of diabetes, HbA1c, systolic blood pressure, BMI,

lipid levels and previous complications. These factors were chosen as they have been

shown to be significant predictors of diabetes related complications in previous risk models

based on the UKPDS9,10. Following a similar approach to the development of other risk

models18 factors were dropped through backwards stepwise elimination if their hazard

ratios were not significant at the 10% level. For death from any cause, separate analyses

were undertaken for those patients with prior events as this may impact on survival

(including diagnosis of cancer or cardiovascular disease, or other major complications). To

investigate whether there was heterogeneity in the hazard ratio of the index score by

treatment allocation (i.e. fenofibrate or placebo) we interacted the index score weight by the

treatment allocation indicator and tested for differences. The proportional hazards

assumption underlying all models was examined using a test based on Schoenfeld

residuals.19

Secondly, to gain a measure of how an EQ-5D index score is associated with absolute risk

an estimate of cumulative hazard for selected index scores (0.60 and 1.00) were calculated

for an individual aged 60 with covariates set to their mean values (male, non-smoker, BMI

30.6, HbA1c 7%, Systolic Blood Pressure 137mm Hg, Total: LDL cholesterol ratio 4.6) This

involved multiplying the baseline hazard by the estimated hazard ratio. Incidence rates

expressed as events of 100 person-years at risk for each type of event were reported to

provide an overall summary measure of the differences in risk across levels of the index

score.

8

Finally, estimates of the impact of index score on survival were obtained from a partial life

table constructed using methods outlines in the statistical appendix. These tables were

used to calculate the average life years lived for men and women aged between 60-80

years for different levels of index score by gender and smoking status.

Results

The characteristics of the patients completing the initial EQ-5D are shown in Table 1. The

average (SD) age of the 7,348 respondents was 66.0 (6.9) years and the average (SD)

index score based on the UK valuation algorithm was 0.81 (0.22).

The average EQ-5D index score differed between those with and without prior events. For

those patients with no prior complications the mean (SD) index score score was 0.82

(0.20), where as it was lower for those who had experienced events: for those, with prior

vascular events it was 0.76 (0.25) and for other major complications it was 0.65 (0.29).

The mean (SD) duration of follow-up after EQ-5D completion was 2.4 (0.49) years with the

longest duration being 4.5 years. During this period, 453 patients had one or more vascular

events, comprising: vascular deaths (89), MI (129), stroke (83), angina (152); 193 patients

experienced other complications, as defined above, with the breakdown of first complication

being heart failure (95), ulcers (44), amputations (30), ulcer plus amputation (7) and renal

dialysis (17); there were a total of 284 deaths including 180 non-vascular deaths. Figure 1

shows the cumulative hazard stratified by ranges of the index score. Across all three

outcomes the lowest risk patients are those with an index score equal to 1 (i.e. at full-

health) followed by those with a EQ-5D index score between 0.75 and 1.00.

For vascular events, other major diabetes-related complications and death from any cause

in each case the EQ-5D index score is a significant independent predictor of risk after

accounting for established risk factors (Table 2). In regard to interpretation, the HR

estimates indicate that a 0.10 higher UK index score (which ranges from -0.59 to 1.0) is

associated with a 7% (95% CI 4-11%) lower risk of vascular events and a 13% (CI 09-17%)

lower risk of complications. For all-cause mortality a 0.1 increase in the index score was

9

associated with a 14% (95% CI 8-19%) reduction in risk for patients with no known prior

diabetes-related events or cancer and a 12% (95% CI 7-16%) reduction of risk for patients

with a co-morbidity. A global test based on Schoenfeld residuals did not indicate a violation

of the proportional hazards assumption of any model. Testing for heterogeneity in the

hazard ratio of the index score between therapies did not indicate a significant difference for

patients allocated to fenofibrate or placebo.

The adjusted incidence of vascular events for those without prior events ranged from 0.86

to 0.60 per 100 person-years at risk for those with a index score of 0.60 and 1.00

respectively. The comparative rates for other outcomes were: 0.63 and 0.35 for other

complications; 0.56 and 0.31 for mortality in individuals without prior events; and 2.57 and

1.44 for those with prior events.

A secondary analysis, using US EQ-5D index scores instead of UK values in the Cox

regression model also produced significant results with larger effects. A 0.1 increment in

index score was associated with: an 11% (95% CI 07-16%) lower risk of vascular events; a

19% (CI 13-24%) lower risk of complications; and in regard to mortality a 20% (95% CI 12-

27%) lower risk for patients without prior events; 17% (95% CI 11-23%) lower risk with prior

events.

Figure 2 shows the hazard ratios associated with items of the EQ-5D after controlling for

other risk factors. The top half reports the hazard ratios with the degree and number of

problems relative to category “no problems” on any dimension. Having some problems (i.e.

on the second level) on two or more dimensions or having “extreme problems” (i.e. on the

third level) on one dimension is associated with higher risk for all types of events. In regard

to the individual dimensions having problems with mobility appears to be the strongest

predictor of future risk (i.e. significant for all dimensions except mortality for those with prior

events).

Table 3 reports estimates of the average life years lived by gender and smoking status for

people with different index scores derived using a partial life table, that contains estimates

of risk over a twenty year period from 60 years of age. The left hand side of the table

10

reports expected life years lived for smokers and the right hand side for non-smokers.

Higher index scores are associated with substantially longer life expectancy. For example,

the average estimated life years over a twenty year period for 60 year old male non-

smokers who rate themselves at full health (i.e. index score of 1) is 17.87 (17.42, 18.34) or

1.85 (1.05, 2.67) years more than those with a index score of 0.60.

Discussion

This study has demonstrated that index scores derived from the EQ-5D can predict the risk

of vascular events, diabetes-related complications and all-cause mortality in people with

Type 2 diabetes. The multivariate analyses indicate that the index score provides additional

information on risk of these outcomes that is over and above that determined from the

major established risk factors such as HbA1c, systolic blood pressure, lipid levels and

smoking status. Patients who had a score 0.10 points higher on the UK based index scores

had 7% lower rates of cardiovascular events, regardless of their clinical history, and 13%

lower rates of other major diabetes-related complications if they had no history of diabetic

complications. This difference in index score was also associated with a reduction in

mortality of between 12-14%. The hazard ratios associated with US index scores were

generally higher, but showed a similar pattern across outcomes.

The main implication of this study for economic evaluations, is that patients with lower index

scores are likely to have a lower HRQL over their remaining lifetime, not only because a

lower score will mean they accumulate QALYs at a slower rate, but also because they have

a higher incidence of events contributing to both morbidity and mortality in the future. For

example, the incidence rate of some major complications and the risk of mortality for a

patient with an index score of 0.60 is almost double the rate of those with an index score of

1.00. When this difference in mortality is extrapolated using life table methods, this level of

difference in risk produces up to a three years difference in life expectancy.

Our results are likely to be particularly relevant to economic evaluations of therapies for

diabetes, when the analysis is conducted alongside randomized controlled trials. In addition

to the FIELD study, the EQ-5D has been used to measure HRQL in several recent large

11

trials of patients with type 2 diabetes such as the ADVANCE study.20 For example, a

potential application is adjustment of the outcomes to take account of differences in HRQL

between patients in the trial population and those in a community setting. Such differences

may arise when selection criteria for recruitment to a randomized control trial, or the quality

of care has an impact on HRQL. It also suggests that differences in EQ-5D index scores

could potentially be a useful intermediate outcome measure of the efficacy of a therapy. For

example, if there were statistically significant improvements in EQ-5D index scores by a

therapy at the end of a study, this would not only indicate potential difference in QALY

within the trial, but also potentially continuing benefits as a consequence of the lower risk of

future events. EQ-5D index scores could have value in a clinical setting where QALYs have

recently been proposed as a performance measure.21

It is important to stress that EQ-5D index scores are derived from responses to the five

questions making up the EQ-5D questionnaire. It is interesting to note that having

problems with mobility appears to be the only dimension that is consistently an independent

predictor of future events. What appears to be more useful at identifying high risk subjects,

are those who report problems on more than one dimension, or having extreme problems in

any dimension. In this regard, a holistic summary measure such as the EQ-5D index score

would appear to be necessary in identifying high risk patients as it is based on the pattern

of responses across all dimensions. It should be noted that the EQ-5D is just one of a

number of instruments that can be used to generate preference based values for assessing

HRQL. It would be interesting to explore whether the preference scores generated by

instruments such as the SF-6D22 and HUI323 are also independent predictors of future

morbidity and mortality in people with type 2 diabetes or other chronic diseases. A common

feature of all these instruments is that they do not include a general self-reported health

question and so the scores for health states are produced by responses to combinations of

questions rather than a global rating of health.

It is unclear why index scores an independent predictor of risk for subsequent fatal and

nonfatal events. It is possible that this relates simply to additional prognostic factors which

are linked to both worse quality of life and poorer outcomes. It is also unclear whether the

patients health status and hence health related quality of life HRQL is the major driver or

whether a patient’s score for a given health state is a factor. For example, index scores may

12

be an indicator of patient behavior, or mood that might more directly influence subsequent

events. Whatever the underlying mechanism, this does not negate the practical value of

these results in making a more valid assessment of the net clinical benefit of therapy

measured in terms of long term QALYs.

In regard to limitations, the models in this study were confined to risk factors used in

previous diabetes specific risk equations based on the UKPDS10 and did not contain risk

factors, such as microalbuminuria, which have been proposed as a potential additional risk

factors for cardiovascular disease.24 As it is possible that the impacts of different index

scores are mediated by some of these omitted factors, the influence that additional clinical

information has on the association between the index scores and these outcomes could be

explored in future work. Patients in this study were followed for an average of two and a half

years, which is a relatively short time compared with the patients’ remaining lifetimes.

Hence, it would be useful to examine longer follow-up periods so as to understand the

duration over which index scores can predict elevations in risk. It would also be useful to

examine whether index scores are predictive of other types of complications, such as

diabetic retinopathy, in future work. Finally, while we have quantified how EQ-5D index

scores influence event rates and life expectancy, it would be useful to quantify the impact in

terms of a holistic measure such as expected QALYs. This would require the development

of a simulation model in which the transition probabilities between states depend not only

upon the current state, but preference based values for these states.

In summary, our results show that index scores derived from the EQ-5D can be used to

identify diabetic patients who are at higher risk of complications and death. It suggests

these scores represent a potential intermediate outcome measure and could be used to

improve predictive models where morbidity and mortality are important outcomes of

economic evaluations of therapies for diabetes.

Appendix: Construction of a partial life table to estimate life

expectancy A partial life table reflecting risk between 60 and 80 years was used to construct estimates

of average number of life years by level of EQ-5D index scores for smokers and non-

smokers. This was based on a proportional hazards model in which current age was time at

risk (used to calculate the baseline hazard) and gender, smoking status, and duration of

diabetes were covariates. The last covariate was included to capture the higher risk

associated with long durations of diabetes.

The hazard ratios (SE) for the estimated model were: index score 0.83 (0.02) ; female

gender 0.64 (0.09) ; smoker 2.6 (0.42); duration of diabetes 1.14 (0.10). Tests of

heterogeneity in the hazard ratio associated with index score for smokers and non-smokers

indicated a non-significant difference (P=0.424).

The annual probability of mortality ( ) between the ages of 60 and 80 years was

calculated using the baseline survivor function ( ) using the

following , where t (t=1…20) reflects number of years from the

persons 60th birthday. This probability was then adjusted for duration of diabetes. Separate

probabilities were estimated for females and smokers by multiplying the annual probability

by the relevant hazard ratio.

)(tq

)(0 tS

)(/)1(1)( 00 tStStq +−=

Life tables were then constructed in the traditional manner, in which was used to

estimate the number of persons alive at each age between 60 and 80. Based on the

assumption of constant annual mortality, the number of person-years lived was then

calculated as and the average life years lived at the age of 60 years

was calculated: .

)(tq

)(tl

2/))1()(()( ++= tltltL

∑= )1(/)()1( ltLe

Non-parametric bootstrapping is used to estimate confidence intervals for life-expectancies.

This involves drawing with replacement 1000 samples from the original data and calculating

life-expectancy with each replication. The confidence intervals can be calculated by

selecting the 26th and 975th of the bootstrap replicates.

13

14

References 1. Torrance GW.Measurement of health state utilities for economic appraisal. J Health Econ. 1986; 5:1-30. 2. Dolan P. Modeling valuations for Euroqol Health States. Med Care. 1997; 35:1095 1108. 3. Glasziou PP, Simes RJ, Gelber RD. Quality adjusted survival analysis. Stat Med. 1990;9:1259 1276. 4. Park SM, Park MH, Won JH, et. al. EuroQol and survival prediction in terminal cancer patients: a multi-center prospective study in hospice-palliative care units, Support Care Cancer. 2006; 14: 329 333. 5. Lenzen MJ, Scholte WMJ, op Reimer WJ, et. al. The additional value of patient-reported health status in predicting 1-year mortality after invasive coronary procedures: a report from the Euro Heart Survey on Coronary Revascularisation. Heart. 2007;93:339 344. 6. FIELD Study Investigators: The need for a large-scale trial of fibrate therapy in diabetes: the rationale and design of the Fenofibrate Intervention and Event Lowering in Diabetes (FIELD) study. Cardiovasc Diabetol 2004; 3:9. 7. FIELD study investigators Effects of long-term fenofibrate therapy on cardiovascular events in 9795 people with type 2 diabetes mellitus (the FIELD study): randomised controlled trial. Lancet 2005; 366: 1849 1861. 8. Stratton IM, Adler AI, Neil HA, et al. on behalf of the UK Prospective Diabetes Study Group. Association of glycaemia with macrovascular and microvascular complications of type 2 diabetes (UKPDS 35): prospective observational study. BMJ 2000; 321: 405–412. 9. Stevens RJ, Kothari V. Adler, AI, et. al. The UKPDS risk engine: A model for the risk of coronary heart disease in Type II diabetes (UKPDS 56). Clinical Science. 2001; 101: 671–679 10. Clarke P, Gray A, Briggs A, et. al. A model to estimate the lifetime health outcomes of patients with Type 2 diabetes: the United Kingdom prospective diabetes study (UKPDS) outcomes model (UKPDS 68). Diabetologia. 2004;47: 1747 1759. 11. The Mount Hood 4 Modeling Group. Computer Modeling of Diabetes and its Complications: A report on the Fourth Mount Hood Challenge Meeting, Diabetes Care. 2007; 30: 1638 1646. 13. Rabin R, de Charro F. EQ-5D: a measure of health status from the EuroQol Group. Ann Med. 2001; 33:337–343 14. Brooks R, Rabin R, de Charro F, eds. The Measurement and Valuation of Health Status Using EQ-5D: A European Perspective. Dordrecht: Kluwer Academic Publishers; 2003.

15

15. Szende, A.; Oppe, M; Devlin, N. (Eds.) EQ-5D Value Sets: Inventory, Comparative Review and User Guide Series: EuroQol Group Monographs , Vol. 2 Springer, 2007 16. Shaw JF, Johnson J, Coons SJ, US Valuation of the EQ-5D health states. Development and Testing of the D1 valuation model. Med Care. 2005; 43:203 220. 17. StataCorp.. Statistical Software: Release 7.0. College Station, TX: Stata Corporation, 2001. 18. Wilson PWF, D'Agostino RB, Levy D, et. al. Prediction of coronary heart disease using risk factor categories. Circulation 1998; 97:1837-1847. 19. Schoenfeld D. Partial residuals for the proportional hazards regression model. Biometrika. 1982 69(1):239-241 20. Glasziou P, Alexander J, Beller E, et. al. Which health-related quality of life score? A comparison of alternative utility measures in patients with Type 2 diabetes in the ADVANCE trial, Health and Quality of Life Outcomes 2007, 5:21. 21. Schmittdiel, J. Vijan, S. Fireman, B. Lafata, JE, Oestreicher, N. Selby, JV Predicted Quality-Adjusted Life Years as a Composite Measure of the Clinical Value of Diabetes Risk Factor Control. Medical Care. 45(4):315-321, April 2007. 22. Brazier, J., Roberts J. and Deverill M. The estimation of a preference-based measure of health from the SF-36 Journal of Health Economics Volume 21, Issue 2, March 2002, Pages 271-292. 23. Feeny, D. Furlong, W. Saigal S and Suna J. Comparing directly measured standard gamble scores to HUI2 and HUI3 utility scores: group- and individual-level comparisons, Social Science & Medicine Volume 58, Issue 4, February 2004, Pages 799-809. 24. Cobb FR, Kraus WE, Root M, et. al. Assessing risk for coronary heart disease: Beyond Framingham. Am Heart J 2003; 146:572 580.

Figure 1: Estimated cumulative hazard curves showing proportion having events by EQ-5D index score level

0.0 0.5 1.0 1.5 2.0 2.50

2

4

6

8

10

Vascular events

p<0.0001

Cum

ulat

ive

haza

rd (%

)

0.0 0.5 1.0 1.5 2.0 2.5

Diabetic complications

p<0.0001

0.0 0.5 1.0 1.5 2.0 2.5

All cause mortality

p<0.0001

Time (years) since EQ5D survey

Key: Index Score < 75 dotted line; Index score >=75 and < 1 dashed line; Index score =1 Unbroken line

16

Figure 2: Hazard ratios for having problems by dimensions of the EQ-5D*

0 1 2 3 4 5 6

anxiety / depressionusual activities

painself caremobility

extreme in any dimension in 2 or more dimensions

in one dimension

Vascular events

HR (95% CI)0 1 2 3 4 5 6

Complications

HR (95% CI)

0 1 2 3 4 5 6

anxiety / depressionusual activities

painself caremobility

extreme in any dimension in 2 or more dimensions

in one dimension

No prior events

HR (95% CI)0 1 2 3 4 5 6

With prior events

HR (95% CI)

All cause mortality

Level of Problems

Level of Problems

Dimensions

Dimensions

Note: Hazard ratios (HR) are expressed relative to the response category “No problem”.

17

18

Table 1: Characteristics of patients completing the EQ-5D questionnaire Characteristics N 7348 Demographic and clinical Female 2794 (38.0%) Age at first EQ-5D survey (years, mean (SD)) 66.0 (6.9) Duration diabetes (years, mean (SD)) 10.1 (5.9) Risk factors Hba1c (%, mean (SD)) 7.0 (1.2) Total cholesterol:LDL ratio (mean (SD)) 4.6 (1.0) Body mass index (kg/m2 , mean (SD)) 29.8 (26.4-33.2) Systolic blood pressure (mmHg, mean (SD)) 136.9 (13.0) Current Smoker 624 (8.5%) Health-related quality of life Proportion of responses by EQ-5D dimension#: Mobility 69%/ 31%/ <1% Self-care 94%/ 5% /<1% Usual activities 74%/ 24%/ 2% Pain/ Discomfort 49%/ 47%/ 4% Anxiety and depression 77%/ 22%/ 1% EQ-5D index score UK mean (SD) 0.81 (0.22) EQ-5D index score US mean (SD) 0.85 (0.15) Clinical history prior to 1st EQ-5D questionnaire* Prior CVD MI 476 (6.5%) Stroke 342 (4.7%) Angina 927 (12.6%) Prior complications Amputation 40 (0.5%) Heart failure 83 (1.1%) Dialysis 5 (0.1%) Lower extremity ulcer 45(0.6%) Prior vascular events 1386 (18.9%) Prior diabetic complications 143 (2.0%) Cancer diagnosis 299 (4.1%)

# Proportion for each response category: Level 1 (i.e. no problems)/ Level 2 (some problems)/ Level 3 (extreme problems). *Includes clinical history prior to start of main trial and events that occurred between randomization and the first EQ-5D questionnaire.

Table 2: Hazard Ratios of risk factors and EQ-5D index score for vascular events, other complications of diabetes and all cause mortality based on multivariate proportional hazard models.

Vascular events

Diabetic complications

All cause mortality

All individuals All individuals With prior complications or cancer

Without prior complications or cancer

Number of Individuals 7348 7348 1693 5655 Number of events

2

453 193 151 133 PH test: statistic (P value) χ 11.40 (0.25) 11.72 (0.30) 9.70 (0.21) 3.31 (0.65) Variable HR P value HR P value HR P value HR P value EQ-5D index score per 0.10 point 0.93 <0.001 0.87 <0.001 0.88 <0.001 0.86 <0.001 Female 0.75 0.007 0.54 <0.001 0.58 0.006 - - Age per 10 years 1.47 <0.001 1.70 <0.001 1.72 <0.001 2.12 <0.001 Diabetes duration per 10 years - - 1.39 0.002 - - - - Hba1c per 1% increase 1.19 <0.001 1.42 <0.001 1.15 0.045 1.21 0.009 Total/HDL cholesterol ratio per 1% 1.13 0.006 - - - - - - Body Mass Index - - 1.04 <0.001 - - Systolic blood pressure 1.17 <0.001 1.13 0.026 - - 1.13 0.056

1.57 0.002 2.32 <0.001 1.78 0.017 3.21 <0.001 Current smoker Prior Vascular events 3.06 <0.001 1.86 <0.001 - - - - Prior Diabetic complications 2.36 <0.001 10.69 <0.001 2.64 <0.001 - - Cancer 3.75 <0.001 - -

Note: Hazard ratios (HR) for variables that were not significant at P<0.1 have been omitted from the table

19

20

Table 3: Estimated average (95% CI) life years between the age of 60 to 80 years by gender, smoking status and index score level

Smokers Non-smokers

Men Women Men Women Level of index score

0.60 11.84 (10.29,13.55) 14.37 (12.89,15.84) 16.02 (15.31,16.74) 17.50 (16.96, 17.98) 0.80 13.59 (12.15, 15.00) 15.77 (14.24,17.02) 17.07 (16.50,17.59) 18.19 (17.74, 18.59) 1.00 15.11 (13.89, 16.48) 16.88 15.69,17.91) 17.87 (17.42,18.34) 18.71 (18.36 ,19.03)

Difference 0.60 vs 0.80 1.75 (1.05, 2.67) 1.39 (-0.68, 3.28) 1.05 (0.14,1.95) 0.69 (0.01, 1.39) 0.60 vs 1.00 3.28 (1.23, 5.28) 2.50 (0.70, 4.31) 1.85 (1.05, 2.67) 1.21 (0.57, 1.84) 0.80 vs 1.00 1.51 (-0.29,3.45) 1.11 (-0.43, 2.88) 0.80 (0.07,1.53) 0.52 (0.00, 1.05)