Two-phase Targeted Maximum Likelihood Estimation for Mixed Data Meta-Analysis

Abstract

This thesis focuses on extension of Inverse Probability of Censoring Weighted Targeted Maximum Likelihood Estimation (IPCW-TMLE) which was initially proposed for two-stage sampled data. We adapt this framework to the setting of mixed Aggregate Data (AD) and Individual Patient Data (IPD) meta-analysis. Our methods are motivated by a systematic review investigating treatment effectiveness for Multi-Drug Resistant Tuberculosis (MDR-TB) where studies consist of mixed IPD and AD, and where treatments are not necessarily observed across all studies. We focus on the estimation of the expected potential outcome under a given treatment and then compare the results using different methods in two simulation studies. We also discuss the challenges and demonstrate our estimation approach when there exist studies that do not have access to the treatment of interest, using the concept of transportability. We use the jackknife estimator to estimate the variance and evaluate the coverage probability of different methods in the simulation study. The results showed near unbiased results along with a close to nominal coverage probability when using IPCW-TMLE method.