A0695
Title: Bayesian joint model for longitudinal biomarker and competing risks with cure fraction, accounting for masked causes
Authors: Md Tuhin Sheikh - Arizona State University (United States) [presenting]
Abstract: In medical studies, subjects who never develop a disease or are cured of it pose challenges, especially in competing risks survival data with partially masked causes. This research is motivated by SELECT data, where prostate cancer can arise from observed or masked causes, and prostate-specific antigens are collected longitudinally as a potential biomarker. To investigate the association between longitudinal biomarker and cause-specific risks for prostate cancer accounting for cure fraction and masked causes, a Bayesian joint model is proposed including a mixed effects regression sub-model for longitudinal data and a double-regression promotion time cure rate sub-model for competing risks survival data. A deviance information criterion is developed to assess model fit in cause-specific survival data. A simulation study is conducted to assess the empirical performance of the methodology and analyze SELECT data to demonstrate the utility of the proposed model and assessment criteria.
