A Two-Stage Linear Mixed/Cox Model for Longitudinal Data and Disease Progression in Pancreatic Cancer Patients

• Activity Number: 359
• Type: Contributed
• Date/Time: Tuesday, August 2, 2011 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Wei Qiao*+ and Ning Jing
• Companies: The University of Texas MD Anderson Cancer Center and The University of Texas Health Science Center at Houston
• Keywords: two-stage model ; linear mixed model ; longitudinal ; Cox model

Abstract

Cancer studies often collect time-to-event data and repeated measurements, e.g. biomarkers, for each subject. In many cases, the research interest is to evaluate if the individual level and progression rates of repeatedly measured biomarkers can quantify the severity of the disease and predict the subject's susceptibility to disease progression. In this study, we present a two-stage model that takes into account the dependency and association between longitudinal measurements of biomarkers and time-to-event data. In the first stage, the subject-specific biomarker trajectories need to be modeled and estimated using linear mixed-effect model ; in the second stage, the subject-specific biomarker trajectories estimated from the first stage are used as coviarates in the Cox model for the disease progression. Consequently, the effects of the biomarker trajectories on the time-to-progression can be assessed. The information matrix of the partial likelihood in the second stage cannot be used to make inference for the estimated risk coefficients, since it does not take into account the uncertainty of the estimated biomarker trajectories. Alternatively, the bootstrap procedure can be u