Biostatistics Seminar speaker, Dr. Abdus S. Wahed, Biostatistics, University of Pittsburgh, will present, “Inference for Dynamic Treatment Regimes in the Presence of Drop-Out.”
Statistical Inference for dynamic treatment regimes from observational studies or randomized trials are generally done via inverse-probability-of-treatment-weighting or Robin’s g-Computation. G-computation is a straightforward approach that fits a regression model of the outcome as a function of the treatment along with observed patient characteristics and then combines the appropriate treatment subgroup means using total probability formula. Inverse probability weighting instead estimates a patient's probability of receiving treatment and weights the outcomes by the inverse of this probability. In this talk, we discuss these techniques and then extend them to estimate the effects of dynamic treatment regimes in the presence of drop-out that are missing at random. We propose a new inverse probability of treatment weighted estimator that accounts for such drop-outs, and derive it’s variance. We compare variations of this method with the existing methods via simulation. We utilize the new class of estimators in the analysis of the STAR*D trial to estimate the mean outcome of patients on different regimes for the treatment of non-psychotic major depressive disorder.