Search

Incomplete data are unavoidable in studies that involve data measured or observed longitudinally on individuals, regardless of how well they are designed. Dropout can potentially cause serious bias problems in the analysis of longitudinal data. In the presence of dropout, an appropriate strategy for analyzing such data would require the definition of a joint model for dropout and measurement processes. This paper is primarily concerned with selection and pattern mixture models as modelling frameworks that could be used for sensitivity analysis to jointly model the distribution for the dropout process and the longitudinal measurement process. We demonstrate the application of these models for handling dropout in longitudinal data where the dependent variable is missing across time. We restrict attention to the situation in which outcomes are continuous. The primary objectives are to investigate the potential influence that dropout might have or exert on the dependent measurement process based on the considered data as well as to deal with incomplete sequences. We apply the methods to a data set arising from a serum cholesterol study. The results obtained from these methods are then compared to help gain additional insight into the serum cholesterol data and assess sensitivity of the assumptions made. Results showed that additional confidence in the findings was gained as both models led to similar results when assessing significant effects, such as marginal treatment effects.