Summary: We have made contributions to the methodological development for handling missing data problems in clinical trials and its applications. Among the many challenges that researchers face in using clinical data, missing data poses a substantial, complex and yet unappreciated threat to validity. Towards addressing issues of missing data, we developed unbiased, efficient statistical approaches by incorporating missing data methods into various statistical models for the analysis of health data with incomplete observations. We also developed statistical computing tools to make our approaches flexible and computationally easy to implement.

Summary: Treatment effects on longitudinal markers of HIV disease progression may vary with time since an unknown/censored time origin. Traditional longitudinal models that misplace the time origin by ignoring censoring may lead substantial bias. We proposed more accurate estimation methods for treatment effects using expectation-maximization approach to deal with the censored time origin. Another challenging problem is that treatment effects may also vary with time or other covariates since treatment randomization/switching in HIV vaccine efficine trials. We proposed generalized semiparametric mixed varying-coefficient effects models to accommodate a variety of link functions and flexibly model different types of covariate effects, including time-constant, time-varying and covariate-varying effects.