The effect of comorbid illness and functional status on the expected benefits of intensive glucose control in older patients with type 2 diabetes: a decision analysis

Overview

This decision analysis is an integration of multiple prediction models from the fields of diabetes and geriatrics. All prediction models were housed within the structure of an existing model of diabetes complications (National Institutes of Health (NIH) Model) (17, 22). This Monte Carlo simulation model is framed by simultaneous progression of disease through individual diabetes-related complications and mortality (Figure 1). Within a 1-year cycle length, patients move from one disease state to another or stay in the current disease state until death or age 95 (Microsoft Excel 2000, Microsoft, Seattle, WA and @Risk 4.0, Palisades, Inc., Newfield, NY). For each specific model setting (e.g., population characteristics, glucose level), the model is run 10,000 iterations (each iteration representing a patient life).

In the following sections, we describe the individual prediction models, the population of interest, the comparison treatments, the outcomes of interest, and sensitivity analyses. Please see the Technical Appendix for details.

Diabetes Complications

The diabetes complication models in this analysis are all derived from United Kingdom Prospective Diabetes Study (UKPDS) results (3, 20, 23–24). Prediction models for all major diabetes-related complications have been developed by the UKPDS study group. These models have been internally validated and externally validated with cardiovascular trial data (25). We could not use the UKPDS prediction model for end-stage renal disease because this model does not include glucose control as a predictor. Instead, we modeled the development of microalbuminuria and proteinuria which are linked to the intensity of glucose control (19, 26). For probabilities under moderate control, we used prediction models developed using optimization procedures to fit observations from the UKPDS control arm to a functional form used in the original NIH model (27) (Technical Appendix). To determine the transition probabilities for intensive glucose control, we used a multiplier derived from the comparison of the overall results of UKPDS for individual complications. For the transition between proteinuria to end-stage renal disease we used probabilities from an observational study (28).

Incorporating functional status and comorbidity into background mortality

We used mortality rates from a 4-year mortality index developed from the Health and Retirement Study (21), rather than mortality rates from life tables (17, 22, 29). This index was developed and validated with a split-sample approach. The index has a total score of 26 and each comorbid illness or functional impairment contributes one to two points to the index score. To calculate background mortality rates for the general population, we subtracted cardiovascular mortality rates for the general population from the mortality rates associated with each index score (30). These mortality rates were multiplied by 2.75 as previously done to reflect higher background mortality rates for patients with diabetes (22).

The baseline index score for each hypothetical subgroup included points for age group and gender. We then systematically increased the mortality index score by as much as 14 additional points. Apart from these changes, we retained the NIH model assumptions about mortality due to other specific causes of death (31–33).

Population of interest

We performed simulations for hypothetical patients with type 2 diabetes at ages 60 to 80 years of age with no prior history of diabetes-related complications. The patients were assumed to have the demographic and clinical characteristics of diabetes patients over 60 years of age, found in the National Health and Nutrition Examination Surveys (NHANES)(1999–2002) (34).

Hypothetical cohorts were divided into 5-year age groups with a uniform age distribution. The age groups correspond with major groupings of older diabetes patients. We also varied the duration of diabetes for such patients (new onset, 0–5 years, 5–10 years, and 10–15 years). We assumed that the population had the gender and ethnic/racial distributions observed in NHANES. We did not assume any additional impact of race on complication rates, because the major ethnic minorities included in the UKPDS data do not completely correspond with the major ethnic minorities in the United States.

Comparison treatments

We compared the projected health effects of moderate glucose control (HbA_1C 7.9%) and intensive glucose control (HbA_1C 7.0%)(11, 35). We assumed that patients maintained these glucose control levels throughout their lives. In sensitivity analyses, we also compared HbA_1C of 9.0% versus 7.9% and HbA_1C of 7.0% versus 6.5% (36).

Other treatment assumptions

All other elements of care were held constant in the two scenarios (37–39). Hypothetical patients had cardiovascular risk factor levels selected from the age, race, and gender specific subgroup distributions for systolic blood pressure, diastolic blood pressure, total cholesterol, high density lipoprotein cholesterol, and smoking for diabetes patients found in NHANES (1999–2002).

Outcomes of interest

Outcomes of interest included the lifetime incidence of individual complications and life expectancy. The benefits of intensive versus moderate glucose control are reported as an average absolute risk reduction in complications and added days of life. The primary outcome of interest was the average difference in quality-adjusted days. To calculate quality-adjusted benefits, we used utility weights for major complications used in prior analyses (19, 27, 40–44). We assumed no disutility of life with different treatments (45). When multiple health states occurred, we used the minimum health state method (46).

Sensitivity Analysis

The UKPDS prediction models assume that the glucose level is a modifiable risk factor for coronary heart disease in type 2 diabetes, when in fact this remains a highly debated and studied topic (47). To assess the impact of this assumption, we replaced the UKPDS models for coronary heart disease and stroke with Framingham models (48, 49). We also conducted sensitivity analyses on other important model assumptions (e.g., background mortality rate inflator). Subgroup analyses were performed for men, women, Whites, African-Americans, and Latinos.

Role of the Funding Source

The funding sources had no role in the design, conduct, or analysis of the study or in the decision to submit the manuscript for publication.

Base Case Results

The model predicted that cardiovascular complications would be the most frequently experienced complications in all patient subgroups (44–54% lifetime incidence coronary heart disease) (Table 1). Microvascular complications had lower lifetime incidences that varied with starting age of the simulation and duration of diabetes. The incidence of end-stage renal disease and amputation declined with increasing starting age but rose with increasing duration of diabetes. Conversely, incidence of blindness increased with increasing starting age but declined with increasing disease duration; these patterns reflect the inclusion of age at onset of diabetes as a predictor in the UKPDS blindness model.

Life expectancy declined with increasing starting age and increasing duration of diabetes. The model’s life expectancy predictions for healthy older patients with new onset diabetes matched expectations from epidemiological studies of mortality and diabetes (50, 51). Quality-adjusted life expectancy was 1–1.5 years less than unadjusted life expectancy.

The overall magnitude of expected benefits of achieving intensive glucose control compared to moderate glucose control declined as the age of hypothetical patients rose (Table 1). With rising starting simulation age, the level of absolute risk reduction declined for end-stage renal disease and amputation but remained stable for blindness and coronary heart disease. Life expectancy and quality of life benefits declined with rising age. The expected benefit of intensive glucose control was 106 (95% Confidence Intervals (CI), 95–117) quality-adjusted days at age 60–64 years of age and declined to 52 (CI, 46–58) days at 75–79 years of age among individuals with no comorbid illness or functional impairment. Increasing duration of diabetes had the opposite effect of increasing the overall expected quality of life benefits of intensive glucose control. For 60–64 year old patients, the overall quality of life benefit increased from 106 days for new-onset diabetes to 114 days for duration of diabetes beyond five years.

Larger differences in the expected benefits of intensive glucose control were observed within each age group with changes in the mortality index score (Figure 2, Appendix Figure 1). As the index score increased, life expectancy declined. In the case of patients 60–64 years of age with new diabetes, life expectancy declined from 14.6 (CI, 14.4–14.7) years at baseline, to 9.7 (CI, 9.6–9.9) years with 3 additional index points, and to 4.8 (CI, 4.7–4.9) years with 7 additional index points. As life expectancy declined, so did expected benefits. Over the same interval of index points, life expectancy benefits declined from 114 (CI, 101–128) days to 41 (CI, 34–48) days to 5 (CI, 3–8) days, and quality-adjusted benefits declined from 106 (CI, 95–117) days to 44 (CI, 38–50) days to 8 (CI, 5–10) days.

The negative associations between life expectancy, benefits, and the mortality index score were also observed for patients with greater duration of diabetes. Again, life expectancy declined with an increasing index score. For patients 60–64 years of age and 10–15 years of diabetes, life expectancy declined from 13.5 (CI, 13.3–13.6) years at baseline to 8.0 (CI, 7.9–8.1) years with 4 additional index points and to 3.9 (CI, 3.8–4.0) with 8 additional index points. Over the same intervals, the expected benefits declined from 116 (CI, 103–129) quality-adjusted days, to 36 (CI, 29–43) days, to 8 (CI, 6–11) days (Figure 2d).

When examining these interrelationships by specific complications, we found distinct patterns for different complications (Figures 3, Appendix Figure 2). For end-stage renal disease, there is a decline in absolute risk reduction as the mortality index score rises, even among patients with extended duration of diabetes (Figures 3b and 3d). On the other hand, the benefits of preventing amputation declined but remained close to an absolute risk reduction over 0.5% percent at high index scores, when duration of diabetes exceeded five years (Figure 3b and 3d). These distinct patterns reflect differences in the size of baseline incidence rates and differences in assumptions regarding how glucose control affected individual complication rates.

Sensitivity analyses

When we used Framingham models, the predicted rates of cardiovascular disease were lower and life expectancies were higher than in the base case analysis, but expected benefits of intensive glucose control were lower. For patients 60–64 years of age with new diabetes and moderate glucose control, the incidence of coronary heart disease declined from 52% to 37 (CI, 36–38)% and life expectancy increased from 14.6 to 16.1 (CI, 15.9–16.3) years. Despite longer life expectancy, the expected benefits of intensive control were found to be less than half of those observed in the base case (e.g,44 (CI, 38–50) instead of 106 quality-adjusted days) due to the absence of glucose control as a predictor in the Framingham models. With all of these changes, the basic relationship between the mortality index score and expected benefits observed earlier was maintained. Expected benefits for the same 60–64 year old patients declined from 44 (CI, 38–50) days to 16 (CI, 13–19) days with 3 additional index points and to 3 (CI, 2–4) days with 7 additional index points.

In other sensitivity analysis, we generally found that results did not deviate significantly from the base case analysis. When we assumed no inflation of the non-diabetic background mortality rate, life expectancies increased by 2–3 years and the magnitude of expected benefits increased (e.g., 106 to 126 (CI, 115–137) quality-adjusted days for healthy 60–64 year old patents with new diabetes). Expected benefits continued to decline with a rising mortality index score but at slightly higher index scores compared to the base case. For example, in the same 60–64 year old patients mentioned earlier, expected benefits were reduced to less than 20 quality-adjusted days with 10 additional index points instead of 7 index points. In subgroup analyses, we found small differences in these patterns between men and women; for 60–64 year old patients, benefits dropped below 20 quality-adjusted days for men at 5 additional index points while women had this decline at 7 additional index points.

Different comparisons of glucose control levels altered the baseline magnitude of benefits of achieving lower glucose control _targets, but did not significantly alter the importance of the index score on the expected benefits. For healthy 60–64 year old patients with new diabetes the benefit was 131 quality-adjusted days for a comparison HbA_1C of 9.0% and 7.9% and 52 days for a comparison HBA_1C of 7.0% and 6.5%. In the case of the first comparison, the benefit declined to 10 days with 7 additional index points. In the case of the second comparison, the benefit declined to 3 days with 7 additional index points.

In our analyses of utility assumptions, we found that model results were highly sensitive to differences in treatment state utilities. If the every day quality of life experience of intensive glucose control was lower than moderate glucose control by more than 0.02 (i.e., a 2% reduction in daily quality of life from perfect health, 1.00 vs. 0.98), intensive control became a harmful therapy. Model results did not change significantly when using lower utilities for complication states.