Estimating Controlled Direct Effects for Time-Varying Treatments Using Structural Nested Mean Models

• Activity Number: 423
• Type: Contributed
• Date/Time: Tuesday, August 2, 2011 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Tomohiro Shinozaki*+ and Yutaka Matsuyama and Yasuo Ohashi
• Companies: The University of Tokyo and The University of Tokyo and The University of Tokyo
• Keywords: direct effect ; g-estimation ; longitudinal data ; structural nested mean model ; time-dependent confounding

Abstract

Estimating direct effects without bias requires that two assumptions hold, that is, the absence of unmeasured confounding for treatment and outcome, and the intermediate and outcome. Even if the above two assumptions hold, one cannot estimate direct effects via standard methods such as stratification or regression modeling if treatment affects confounding factors, namely, time-dependent confounders. Sequential g-estimation method for the structural nested mean models has been developed for estimating direct effects in the presence of time-dependent confounders. In this talk, we extend this method for data with time-varying treatments and repeated measured intermediates. Simulation studies showed that usual regression approaches were heavily biased in the presence of time-dependent confounders, but our sequential g-estimator remained unbiased. The proposed method was applied to data from a large primary prevention trial for coronary events in which plavastatin was used for lowering cholesterol. Our analyses showed that, for patients with moderate hypercholesterolemia, the benefit experienced by pravastatin could not be attributed to the effect of treatment on the cholesterol levels.