Disease Progression Monitoring: Bayesian Joint Modeling of Latent Time Series Measures of Longitudinal Data and Time-to-Event Outcomes

• Activity Number: 435
• Type: Invited
• Date/Time: Wednesday, August 6, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Korean International Statistical Society
• Author(s): Mi-Ok Kim and Sungduk Kim*+
• Companies: Cincinnati Children's Hospital Medical Center and NIH/NICHD
• Keywords: Joint Modeling ; Latent time series structure ; Bayesian analysis ; Skewed distributions

Abstract

Biostatistics research that deals with the joint modeling of longitudinal and time-to-event data has flourished for the past decade in order to investigate how a marker measured repeatedly in time is associated with a time to an event of interest. Existing approaches heavily rely on the Gaussian framework for the longitudinal marker process, whereas the marker process can be skewed in distribution with both skewedness and variance varying over time. Data from disease progression monitoring of lung functionality and survival/lung transplant prediction study in cystic fibrosis patients reveals these characteristics. We propose a Bayesian method that uses latent time series structure to analyze heavily skewed longitudinal marker data with growing variance over time. Time series modeling facilitates treatment of missing observations, and the joint modeling with survival time addresses informative drop-outs. We predict 5-year longitudinal trajectory of lung functionality measure and death or lung transplant probability. Markov chain Monte Carlo sampling is used to carry out Bayesian posterior computation. Several variations of the proposed model are considered and compared via the devi