Joint Modeling of Longitudinal and Survival Data to Estimate Treatment Effects

• Activity Number: 150
• Type: Invited
• Date/Time: Monday, August 1, 2011 : 10:30 AM to 12:20 PM
• Sponsor: SSC
• Author(s): Jeremy Michael George Taylor*+
• Companies: University of Michigan
• Keywords: prostate cancer ; causal effects ; treatment by indication ; joint models

Abstract

Estimating a treatment effect of an intervention from longitudinal observational data, when the treatment is given by indication, is challenging. We utilize a longitudinal and hazard modeling approach to estimate the effect of salvage hormone therapy for patients who are being followed after treatment for prostate cancer. The longitudinal data are prostate specific antigen (PSA) counts and the survival data are time of recurrence. Patients with higher PSA values are at higher risk for recurrence and they also tend to be given hormone therapy. We are interested in estimating the reduction in the hazard of the recurrence of the prostate cancer for receiving hormone therapy for that patient conditional on their longitudinal PSA data compared to not receiving hormone therapy. The joint longitudinal-survival modeling approach requires a model for what PSA would have been had the person not taken hormone therapy. The method will be compared with a sequential stratification propensity score method and marginal structural models.