Efficacy studies of malaria treatments can be plagued by indeterminate outcomes for some patients. The study motivating this paper defines the outcome of interest (treatment failure) as recrudescence and for some subjects, it is unclear whether a recurrence of malaria is due to that or new infection. This results in a specific kind of missing data. The effect of missing data in causal inference problems is widely recognized. Methods that adjust for possible bias from missing data include a variety of imputation procedures (extreme case analysis, hot-deck, single and multiple imputation), inverse weighting methods, and likelihood based methods (data augmentation, EM procedures and their extensions). In this article, we focus on multiple imputation, two inverse weighting procedures (the inverse probability of censoring weighted (IPCW) and the doubly robust (DR) estimators), and a likelihood based methodology (G-computation), comparing the methods' applicability to the efficient estimation of malaria treatments effects. We present results from a simulation study as well as results from a data analysis of malaria efficacy studies from Uganda.



Included in

Biostatistics Commons