A Robust Joint Model of Longitudinal Trajectories and Time-to-Event Data at Biobank Scale

• Activity Number: 137 - Joint Modeling for Longitudinal and Survival Outcomes in Health Studies
• Type: Topic Contributed
• Date/Time: Monday, August 8, 2022 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Computing
• Author(s): Hua Zhou* and Jin Zhou and Gang Li
• Companies: UCLA and UCLA and University of California, Los Angeles
• Keywords: accelerated failure time (AFT); dynamic prediction; joint modeling

Abstract

Motivated by the analysis of massive electronic health record (EHR) and wearable device data in modern biobanks, we propose a robust and scalable M-estimator, termed the joint model robust estimator (JMRE), for estimating the accelerated failure time (AFT) model for a right-censored event time jointly with a linear mixed model (LMM) for the longitudinal biomarker trajectory. As a semiparametric estimator, JMRE is robust to distribution misspecification in both AFT and LMM models; scalable to biobank data with $10^5 \sim 10^8$ individuals, intensive longitudinal measurements, and a large number of random effects; able to model the time-varying effects on both mean and within-subject variance of the longitudinal biomarker simultaneously; and easily extensible to data with multiple longitudinal biomarkers.