Multiply-robust adjustment for dependent drop-out in longitudinal studies
*Eric Tchetgen, Harvard University  

Keywords: longitudinal , drop-out, GEE, doubly-robust, multiply-robust

Longitudinal studies are now commonly used in health sciences research, however many such studies suffer from severe patient drop-out. Often, dropout can be explained by incorporating into the analysis, a large vector of observed auxiliary longitudinal data which is otherwise not of primary scientific interest. Inverse-probability-weighting has been proposed as an approach to adjust for this form of drop-out by incorporating these auxiliary data through weighting. Unfortunately bias due to model mis-specification of the probability weights present a serious threat to the validity of the resulting inferences. Double-robust inferences have been proposed that partially relax the requirement of correctly specified weights. In this talk, we review existing double-robust methodology and describe recent multiple-robust estimators that improve on previous estimators in the sense that the new estimators remain consistent and asymptotically normal under many more possible data generating mechanisms. Simulation studies and a data example are given to illustrate the finite sample performance of the various estimators.