A0593
Title: Bayesian nonparametric methods for longitudinal mediation with informative continuous-time treatment decisions
Authors: Jason Roy - Rutgers University (United States) [presenting]
Arman Oganisian - Brown University (United States)
Abstract: The time between treatment courses might be informative in many clinical settings, such as cancer chemotherapy. In such settings, when there are questions about treatment's direct and indirect effects, available statistical methods are limited. Flexible Bayesian models are proposed for the continuous time decision process, time-varying mediator, and time-varying covariates. Then a g-computation approach is used to obtain the posterior distribution for the direct and indirect effects. The motivating example involves quantifying the contribution of chemotherapy-associated sepsis (time-varying mediator) to transient cardiac toxicity, which may help mitigate premature discontinuation of anthracycline chemotherapy agents. The performance of the models is assessed via simulations, and it is applied to data from a study of acute myeloid leukaemia.
