OBJECTIVES

Knowler et al published the initial results of the DPP trial in 2002. The study reported a crude incidence density (ID) of 4.8, 7.8 and 11.0 cases per 100 person-years for lifestyle, metformin, and placebo respectively. The three-year cumulative incidence (CI) of diabetes was 14.4%, 21.7% and 28.9% respectively. These results adjusted neither for baseline covariates nor for treatment interactions. While randomization is expected to ensure baseline balance, that is not guaranteed. Using subgroup analysis to handle the heterogeneity of treatment effect (HTE) is problematic due to multiple testing and inadequate statistical power. The main purpose of the study is to estimate adjusted CI, assessing the HTE using multiple logistic regression (MLR).

METHODS

A 95% subset of the original DPP data was available. Fitted multiple logistic regression was used to compute an adjusted CI for each treatment.

RESULTS

Adjusted CIs controlling for baseline covariates and including treatment interaction terms were 14.5%, 20.9% and 28.3%, similar, though not identical to the unadjusted full sample. Based on these CI, the ID was 5.2, 7.8 and 11.1 cases per 100 person-years. There was a statistically significant HTE effect by age and BMI. The odds ratios of lifestyle and metformin compared to placebo in 40-64 years old participants compared to 25-39 years old were 0.49; [ 95% CI: 0.25,0.97] and 0.57;[95% CI: 0.29,1.12], respectively. The analogous odds ratios in those with BMI of ≥35kg/m2 compared to those with BMI of <30kg/m2 were 1.18; [95% CI: 0.63,2.21] and 0.36; [95% CI: 0.20,0.63].

CONCLUSIONS

Our findings from MLR generated similar results for CI and ID as in the DPP despite differences in sample sizes and the methods used (life table versus MLR). Both lifestyle and metformin were similarly effective in reducing the incidence of diabetes compared to placebo after adjustment. Age and BMI levels were identified as effect modifiers.