Affiliation: Gillings School of Global Public Health, Department of Biostatistics
A common objective of biomedical cohort studies is assessing the effect of a time-varying treatment or exposure on a survival time. In the presence of time-varying confounders marginal structural models fit using inverse probability weighting can be employed to obtain a consistent and asymptotically normal estimator of the causal effect of a time-varying treatment. This document considers estimation of parameters in the semiparametric marginal structural Cox model (MSCM) from a case-cohort study. Case-cohort sampling entails assembling covariate histories only for cases and a random subcohort, which can be cost effective, particularly in large cohort studies with low outcome rates. Following Cole et al. , we consider estimating the causal hazard ratio from a MSCM by maximizing a weighted-pseudo-partial-likelihood. The estimator is shown to be consistent and asymptotically normal under certain regularity conditions. Computation of the estimator using standard survival analysis software is discussed and results from a simulation study are presented. In the standard (associational) case-cohort Cox analysis, various methods have been proposed to improve efficiency from maximum pseudolikelihood estimators of Prentice  or Self and Prentice . As the presented theory of MSCM parameter estimator is developed based on Self and Prentice  we briefly review those methods and discuss extension of the methods to the MSCM analysis. In addition, we proposed a new method to improve efficiency of the case-cohort MSCM analysis from a biomedical study that aims to evaluate the causal effect of treatment on a time to event. We seek to improve the efficiency by multiple imputation method which can make fuller use of covariate information that are available from full cohort. The proposed method is applied to the Multicenter AIDS Cohort Study (MACS) and the Women's Interagency HIV Study (WIHS).