Likelihood-Based Instrument Variable Analysis for Comparing Survival of Prostate Cancer Patients Receiving Hormone Therapy vs. Conservative Management

• Activity Number: 512
• Type: Contributed
• Date/Time: Wednesday, August 3, 2011 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Anjun Cao*+ and Dirk F. Moore
• Companies: University of Medicine and Dentistry of New Jersey and University of Medicine and Dentistry of New Jersey
• Keywords: IVA ; covariate adjustment ; prostate-specific survival ; PSA level

Abstract

Instrument variable analysis (IVA) is a method for comparing interventions using observational data. Unlike classical methods of covariate adjustment, it can adjust for both measured and unmeasured covariates, provided that essential assumptions are satisfied. The most important assumption is that an instrument can be found that is associated with the outcome of interest only through its association with the intervention. We present a two-stage likelihood-based IVA procedure for comparing survival data between two interventions. We show that the method can produce unbiased estimates of survival differences if the distribution of an unobserved confounder is known. We use the method to compare prostate-specific survival of patients receiving hormone therapy vs. conservative management, where the unobserved confounder is PSA level.