Title: A flexible joint longitudinal-survival modeling framework for incorporating multiple longitudinal biomarkers
Authors: Daniel Gillen - University of California, Irvine (United States) [presenting]
Sepehr Akhavan - Cylance (United States)
Alexander Vandenberg-Rodes - Obsidian Security (United States)
Babak Shahbaba - UC Irvine (United States)
Abstract: When monitoring the health of subjects it is common for multiple biomarkers to be measured longitudinally over time. While the associations between the collected biomarkers and a time-to-event endpoint are often of primary scientific interest, modeling the longitudinal risk factors simultaneously can be beneficial, particularly when the density of measurements is differential across biomarkers or when data on some biomarkers are intermittently missing. We propose a joint longitudinal-survival framework with the longitudinal component modeled via a Gaussian process that allows for the correlation between biomarker trajectories to be estimated and utilized. Biomarker trajectories are then linked to survival times via a multiplicative hazards model. Joint estimation of the longitudinal and survival models is performed to account for uncertainty in the estimated time-dependent biomarkers. The proposed methodology is robust to strong parametric assumptions on the mean and covariance structure of the longitudinal component, while also allowing for subject-specific baseline hazard functions in the survival component. Simulation studies are presented to illustrate the performance of the proposed method. We further use the approach to estimate the association between multiple serum-based nutritional biomarkers and survival among end-stage renal disease patients.