Title: Dynamic treatment regimes - Nonparametric Bayes for causal inference
Abstract: We discuss inference for multi-stage clinical trials. The motivating example are multi-stage chemotherapy regimes for acute leukemia. Patients were randomized among initial chemotherapy treatments but not among later salvage therapies. We propose a Bayesian nonparametric (BNP) approach to account for the lack of randomization in the later stages. We argue that the BNP approach can provide an objective evaluation of a causal effect of competing treatment regimens, adjusting for the lack of randomization. In a simulation study we compare the BNP approach with standard doubly robust causal inference methods and show how the BNP approach compares favorably as an objective method that does not rely on particular model assumptions for a response or model for treatment assignment. The paper is Xu Y, Mueller P, Wahed A and Thall P., "Bayesian Nonparametric Estimation for Dynamic Treatment Regimes with Sequential Transition Times." JASA, in press https://arxiv.org/abs/1405.2656