A1390
Title: A joint model for longitudinal change point analysis and recurrent events with application to cystic fibrosis outcomes
Authors: Jiayuan Shi - University of Colorado Anschutz (United States) [presenting]
Giovani Silva - Universidade de Lisboa (Portugal)
Brandie Wagner - University of Colorado Anschutz (United States)
Elizabeth Juarez-Colunga - University of Colorado Anschutz Medical Campus (United States)
Abstract: Many biomedical studies collect repeated biomarker measurements together with recurrent clinical event outcomes, where changes in biomarker trajectories may provide important information about underlying disease progression and recurrent event risk. A Bayesian joint model for longitudinal biomarkers with subject-specific trajectory changes and recurrent clinical events is proposed. The model links individualized biomarker trajectories to recurrent event risk through shared random effects, allowing event risk to be associated with biomarker levels, rates of change, and latent change-point timing. The longitudinal submodel captures sharp subject-specific shifts in biomarker trajectories, while the recurrent event submodel uses a counting-process formulation that accounts for left truncation due to variable entry ages. The model quantifies temporal associations between biomarker progression and recurrent outcomes. Application to data from the Early Pseudomonas Infection Control study of children with cystic fibrosis demonstrates the method's utility. The preferred two-random-change-point model estimated mean subject-specific change points at 6.7 and 15.6 years. The odds of Pseudomonas aeruginosa Infection increased modestly between the two change points, followed by a sharp increase after the second change point. Faster increases in Infection probability between change points were associated with elevated recurrent pulmonary exacerbation risk.
