An application of a joint model for interval-censored disease progression data with informative observation times.

• Activity Number: 377
• Type: Contributed
• Date/Time: Tuesday, August 6, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Veronica Sabelnykova*+
• Companies: University of Victoria

Abstract

Multi-state forward models are readily used for modeling progression of a disease, where states represent health status of a patient. In practice, patients are often observed periodically. Other times patients are observed when necessity strikes. This implies the life history and observation process are not independent, making the examination process informative. Since the life history process and the follow up process are not independent the aim of this study was to model the two processes jointly to ensure valid inference. To illustrate the benefits of the joint model we use a data set with examination times that were not fixed in advance and instead were subject to random fluctuations usually depending on how the disease is progressing for a certain individual. In addition, the number of examination times was also random and the transition times between disease states are interval censored. We use a proportional hazards model to model the transition intensities and link the two processes via random effects. Using a Bayesian framework we show advantages of the joint model.