Analysis of Repeated Measures in the Presence of Missing Observations due to Dropout

Abstract

Incomplete data is common in both observational studies and clinical trials. Ignoring missing data may produce seriously biased estimators and could lead to misleading results. During the last three decades, a vast amount of work has been done in this area. The approaches can be classified into the following main categories: imputation methods, likelihood-based methods and inverse probability weighting methods. Longitudinal and crossover studies with repeated measures are particularly subject to missing observations. Various methods, including generalized estimating equations (GEE) (Liang and Zeger, 1986), weighted GEE (WGEE) (Robins, Rotnitzky and Zhao, 1995) and multiple imputations, have been proposed to cope with missing data in longitudinal studies. However, very few researchers have explored the missing data issue in crossover studies. In addition to reviewing and critiquing the methods dealing with missing observations in general and in repeated measures, in this dissertation, we propose a new weighting approach for GEE to estimate the regression parameters in crossover studies. The proposed method provides consistent and asymptotically normally distributed estimators. Simulation and asymptotic efficiency results indicate that the proposed estimators are more efficient than both regular GEE and WGEE. Applications of the proposed method are illustrated with real data.