|Institution:||University of Washington|
|Keywords:||Immune correlates of protection; Longitudinal data; Measurement error; Proportional hazards model; Two-phase sampling; Biostatistics|
|Full text PDF:||http://hdl.handle.net/1773/27428|
Assessing immune correlates of protection, the immune responses that reliably predict the vaccine efficacy on the clinical endpoint, has always been an important objective in vaccine efficacy trials. In this dissertation, we study the continuous and dichotomized trajectory of time-varying immune response as the immune correlate of protection in two-phase sampling design cohort studies. We adopt the joint modeling framework that models the immune response data measured longitudinally and with error and the time-to-event clinical endpoint simultaneously. The inherent evolution of the time-varying immune response is characterized by a random effects model, and its relationship with the instantaneous risk of the clinical event is modeled by the Cox proportional hazards regression. This regression model allows for direct assessment of immune correlates of protection in Prentice's framework. This evaluation is different from the traditional work that is based on measured values of biomarkers. Instead, by studying the underlying trajectory, the application is to generate hypotheses about the biological mechanisms of protection. The main objective of the dissertation is to develop statistical methods to make inference on the regression model accounting for the missing immune response data due to two-phase sampling. For the inference on the continuous immune response trajectory, we extend the existing conditional score method to the two-phase sampling design cohort studies by using the technique of weighting the complete cases by the inverse probabilities of observing the immune response data, and the augmented inverse probability weighting. For the dichotomized immune response trajectory, we propose estimating equations based on regression calibration method. We also generalize it to two-phase samples by the inverse probability weighting method. We finally apply the proposed methods to the AIDS Clinical Trials Group (ACTG) 175 dataset, a randomized clinical trial comparing monotherapy with combination therapy among HIV-1-infected subjects.