Abstract
BACKGROUND In the presence of non-adherence, intention-to-treat analysis preserves randomization, but does not lead to a valid comparison of outcome between the assigned groups. Using a reanalysis of Diabetes Prevention Program, this study aimed to estimate the causal effect of treatment with intensive lifestyle intervention or metformin vs. placebo on blood pressure and lipid profile using G-estimation after accounting for non-adherence. METHODS The Diabetes Prevention Program randomized 3,052 pre-diabetic individuals to metformin (N = 1015), placebo (N = 1014), or an intensive lifestyle intervention (N = 1023). G-estimation was used to estimate the causal effect of intensive lifestyle intervention or metformin vs. placebo on blood pressure and lipid profile in 2,973 patients who had adherence data. For comparison, we also performed the standard intention-to-treat analysis. RESULTS The G-estimation results showed that intensive lifestyle substantially improves systolic and diastolic blood pressure and lipid profile. The G-estimates of the effects of metformin vs. placebo as well as intensive lifestyle intervention vs. metformin on blood pressure and lipid profile were also stronger than the intention-to-treat effect estimates. CONCLUSION G-estimation suggests that intensive lifestyle modification improves known risk factors for cardiovascular disease, including systolic blood pressure, diastolic blood pressure, triglyceride, and HDL levels more than what standard ITT analysis suggests. Adherence to the assigned treatment should be measured in all randomized trials, and G-estimation should be the standard analysis of randomized trials with substantial non-adherence.
