A0959
Title: Heterogeneous treatment effect estimation for longitudinal outcomes
Authors: Tianyu Pan - University of California, Irvine (United States)
Lu Tian - Stanford University (United States)
Vivek Charu - Stanford University (United States) [presenting]
Abstract: Developing trials around clinical endpoints for chronic kidney diseases, such as the event-time outcomes of end-stage kidney diseases, is often challenging due to the low likelihood of occurrence likeliness within ten years post-treatment. A slope-based endpoint, the rate of decline in eGFR in the first three years post-treatment, has been suggested as an effective surrogate for clinical endpoints. Existing research has focused on relaxing linear trend assumptions on eGFR and non-informative censoring assumptions or analyzing treatment effects in predetermined subgroups. Yet, none have explored data-driven sub-group identification and heterogeneous treatment effect estimation. The aim is to propose a method that introduces a Bayesian decision tree structure into a shared-parameter model, combining a survival model with a two-slope spline model that characterizes the rate of decline in eGFR. The proposed model simultaneously estimates the eGFR slope in the presence of informative censoring and provides digestible clinical decisions for sub-grouping governed by slope-based treatment effect heterogeneity. Simulation studies showcase that our proposed model effectively captures subtle heterogeneity in slope-based treatment effects. The model is also applied to the modification of diet in renal disease (MDRD) trial, providing Bayesian evidence that the patients with higher kidney failure risk score benefit more from the treatment.
