B1039
Title: Bayesian semiparametric models for dynamic treatment rules with incomplete time-varying covariates
Authors: Arman Oganisian - Brown University (United States) [presenting]
Abstract: A Bayesian semiparametric model is developed for assessing the impact of anthracycline chemotherapy (ACT) on survival among patients diagnosed with pediatric acute myeloid leukaemia (AML). The data are from a phase III clinical trial in which patients move through a sequence of four treatment courses. At each course, a decision is made to administer ACT. Since ACT is cardiotoxic, left ventricular ejection fraction (EF) is sometimes, but not always, measured and used to help inform the ACT decision ahead of each course. The inconsistent EF assessment induces informative missingness in the time-varying covariate. Moreover, patients may die or be withdrawn from the study before ever completing the sequence. The problem is framed in terms of a joint dynamic treatment rule (DTR) that outputs both an EF monitoring decision and a subsequent ACT treatment decision. Bayesian semiparametric models are used to model continuous-time transitions between treatment courses (recurrent states) and death (the absorbing state). A g-computation procedure is used to compute posterior marginal survival probabilities under hypothetical monitoring-treatment schemes.