Time-to-Event Prediction Based on Longitudinal Biomarkers Using Bayesian Hierarchical Changepoint Mixture Models

• Activity Number: 173 - Bayesian Methods Applied to Biometric Problems
• Type: Contributed
• Date/Time: Monday, July 29, 2019 : 10:30 AM to 12:20 PM
• Sponsor: ENAR
• Author(s): Lynette Smith* and Yeongjin Gwon and Morshed Alam and Sukhwinder Kaur
• Companies: University of Nebraska Medical Center and University of Nebraska Medical Center and University of Nebraska Medical Center and University of Nebraska Medical Center
• Keywords: biomarker; changepoint; mixture model; Bayesian

Abstract

Pancreatic cancer (PC) is an extremely aggressive malignancy and while identification of patients at a resectable stage results in increased patient survival, PC patients are often diagnosed at late stages. Biomarkers are needed that can be used to screen individuals to detect disease and predict survival. Longitudinally measured biomarkers can show a variety of trends over time. PC patients with progressive disease could reflect either no change or rapid changes in biomarkers prior to death from the disease. The rate of change in biomarker levels has potential to predict presence/progression of disease and patient survival. Here we utilize a Bayesian changepoint mixture model to predict time to event outcomes. The changepoint model looks at trends over time and determines where changes in the trends occur. However, due to genetic, or other differences, not all subjects will experience changes in biomarkers over time, and the mixture piece of the model captures the fact that groups of individuals behave differently. Goodness of fit of the model is assessed by Q-Q plots and sensitivity to prior distribution selection assessed by comparing multiple priors posterior summaries.