Biostatistics Seminar Series - Kristin A. Linn, PhD

Title: Preliminary Ideas for Estimating Individualized Treatment Rules with Multiply Imputed Data   Abstract: The goal and mantra of personalized medicine is to determine the right treatment for the right patient at the right time. A large number of methods exist for estimating optimal dynamic treatment regimes (DTRs), which are alternatively called adaptive treatment strategies or individualized treatment rules, among other names. Despite significant attention given to estimation of DTRs, little attention has focused on how to estimate optimal DTRs using a collection of data sets that are output from a multiple imputation model. Typically, the evaluation of DTRs focuses on the Value of the estimated strategy, i.e., the expected value of the outcome of interest that would be observed if the DTR were applied to assign treatments in the population. Thus, standard rules for combining multiply imputed data that target inference for individual model parameters do not directly apply. In this talk, we motivate this problem using data from a 4-arm randomized trial studying the effects of several social interaction-based gamification interventions on activity levels (measured by step counts) of participants. We present preliminary results and propose several ideas for future methodological work, with the goals of obtaining early feedback and prompting interest in methodological collaboration on this problem.