Clinical Input from Project Dulce
and provide baseline demographic and clinical characteristics and baseline complications, respectively, for each of the four health insurance cohorts. Overall, the mean age was 51.2 (SD=12.7), 61 percent were female, and 48 percent were Latino. The CMS and Medi-Cal cohorts had similar demographic and clinical characteristics. The uninsured cohort has the highest percentage of Latinos (81 percent) and the highest baseline A1c (9.4). The commercial cohort had the lowest percentage of Latinos (16 percent) and the lowest baseline A1c (7.8). Interestingly, the uninsured cohort had the lowest levels of cardiovascular complications at baseline. shows the observed clinical changes in the four cohorts. The largest decreases in A1c were seen among the uninsured cohort, (−1.3 percent), followed by the CMS (−0.8 percent), Medi-Cal (−0.5 percent), and commercial cohorts (−0.4 percent).
Baseline Characteristics of the Health Insurance Cohorts
Baseline Complications in the Health Insurance Cohorts
Estimated Treatment Effects Associated with Project Dulce in Health Insurance and A1c Simulation Cohorts
Long-Term Simulation Results
The Dulce diabetes case management program was projected to improve life expectancy in all four cohorts defined by medical insurance coverage in the base case analysis (). The standard deviations represent the expected distribution of results for simulations of large groups of patients at the mean value of input parameters without uncertainty in these parameter values. Thus, although they are small compared with the means within each group, they represent a higher proportion of the mean differences between groups. Projected gains in life expectancy were highest in the uninsured cohort (mean 1.1 years), followed by the CMS cohort (0.6 years), the Medi-Cal cohort (0.3 years) and the commercial cohort (0.2 years). These differences in improvements in life expectancy are the result of differences between cohorts at baseline. For example, mean age and duration of diabetes was lower, and baseline A1c was higher, in the uninsured cohort than in the other three groups, leading to greater clinical benefits with treatment. Mean baseline A1c, SBP, and other risk factors were lowest in the commercial insurance cohort, and therefore smaller incremental gains were observed in this patient group. Similarly, the Dulce intervention was associated with improvements in quality-adjusted life expectancy in all four insurance cohorts. The largest improvement was observed in the uninsured cohort (0.9 QALYs), then in the CMS cohort (0.4 QALYs), the Medi-Cal cohort (0.3 QALYs), and the commercial cohort (0.2 QALYs).
Summary of Results for the Health Insurance Cohorts
Mean direct medical costs over patient lifetimes ranged from $57,530 to $82,225 per patient. The Dulce diabetes treatment program was associated with higher lifetime costs than control in all four health insurance cohorts. Lifetime incremental (net) costs were comparable in all three insured cohorts, with the largest incremental cost projected for the commercial cohort ($12,368 per patient), followed by the Medi-Cal cohort ($11,792), the CMS cohort ($10,921), and the uninsured cohort ($8,991). A breakdown of costs showed that approximately one-third of the additional costs of implementing the Dulce intervention were offset by reduced costs of diabetes-related complications over patient lifetimes.
Calculation of incremental cost-effectiveness ratios (ICERs) for each of the health insurance cohorts showed that the Dulce intervention was in the range considered to represent a good value by currently accepted standards. The lowest ICER for Dulce versus control was projected for the uninsured cohort ($10,141 per QALY gained), followed by the CMS cohort ($24,584 per QALY gained), the Medi-Cal cohort ($44,941 per QALY gained) and the commercial cohort ($69,587 per QALY gained).
Plotting the incremental costs and incremental effectiveness (in terms of quality-adjusted life expectancy) for each of the 1,000 mean values generated by the model allowed us to evaluate the percentage of these values that fall below a specific willingness to pay values and to generated acceptability curves for each of the four cohorts (). These cumulative probabilities in the acceptability curves represent the expected distribution of results for simulations of large groups of patients, with no uncertainty in the value of input parameters. With a willingness to pay of $50,000 per QALY gained, the acceptability curves indicate that the Dulce intervention had a 100 percent likelihood of being cost-effective versus control in the uninsured cohort, a 92 percent probability in the CMS cohort, a 57 percent probability in the Medi-Cal cohort, and a 31 percent probability in the commercial cohort. With a willingness to pay of $100,000 per QALY, these probabilities were 100, 98, 81, and 62 percent, respectively.
Decreasing the time horizon to 20 years reduced the improvements in life expectancy and quality-adjusted life expectancy as well as incremental costs with Dulce versus control compared with base case (). These changes led to a lower ICER in the uninsured cohort, but an increased ICER values in the commercial cohort. Reducing the A1c treatment effect by 50 percent raised the ICERs for all groups to the point where treatment for commercial populations was not cost-effective. Introducing cohort specific case management costs slightly increased (reduced) the ICER for the uninsured (commercial) cohort. Varying the discount rate between 0 and 6 percent per annum on costs and clinical outcomes did not substantially change the outcomes from base case.