09/20/2013

Post date: Sep 28, 2013 9:41:44 PM

Zhen Chen

Biostatistics and Bioinformatics Branch, NICHD/NIH

Topic: Bayesian causal inference for pregnancy related outcomes

Abstract

In many reproductive and fecundibility studies, the research interest usually centers at the effect of a pre-conception intervention on a pregnancy-related outcome. For example, the EAGeR study of NICHD aims at investigating the effect of preconception low dose aspirin on gestational age (GA). In these situations, a standard treatment/control comparison of the outcome is not appropriate, as those without a conception will not have the desired outcome observed. Motivated by the EAGeR study, we propose to use the principal stratification approach, treating time to pregnancy (TTP) as the post-randomization variable. We use a Gaussian copula to model the TTPs under treatment and control, with each marginal formulated by a discrete time proportional hazards model. The potential outcomes of GA are modeled through linear regressions where the potential TTP are used as covariates. For estimation and inference, we use a Bayesian approach which provides a natural framework for handling the potential outcomes and censored TTPs. We illustrate the proposed approach through simulation studies and a preliminary analysis of the EAGeR data.