A Comparison of Statistical Approaches for Cluster Randomized Trials with Survival Outcomes

• Activity Number: 233
• Type: Contributed
• Date/Time: Monday, August 3, 2009 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Margaret R. Stedman*+ and David Gagnon and Robert Lew and Elena Losina and Daniel Solomon and M. Alan Brookhart
• Companies: Brigham and Women's Hospital and Boston University School of Public Health and Boston University and Boston University and Brigham and Women's Hospital and Brigham and Women's Hospital/Harvard Medical School
• Keywords: Cluster Randomized Trial ; Shared Frailty Models ; Survival Analysis ; Clustered Permutation Test ; Clustered Logrank Test ; Simulation Study

Abstract

Applying the characteristics of a physician randomized trial, we performed 3 simulation studies to compare conditional, marginal, and nonparametric methods for analyzing clustered survival data. We simulated data from a shared frailty model with (1) fixed cluster size, (2) variable cluster size, (3) non-normal random effects. Methods of analyses included: the accelerated failure time frailty model (AFT), Cox proportional hazards model with robust standard errors (COXPH), clustered logrank test (CLR), and the clustered permutation test (CPT). We compared Type I error, power, mean-squared error (MSE), and the coverage probability of the marginal treatment effect estimator. Nominal Type I error was preserved for all methods and scenarios tested. When the frailty was correctly specified AFT yielded less biased estimates with smaller MSE than COXPH.