Findings from Simulation Studies on Treatment (Trt) Switching in Clinical Trial (CT)

• Activity Number: 334
• Type: Contributed
• Date/Time: Tuesday, July 31, 2007 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Dachuang Cao*+ and Hongmei Jiang
• Companies: Eli Lilly and Company and Northwestern University,
• Keywords: treatment switching ; marginal structural model ; generalized estimating equation (GEE) ; repeated measures analysis

Abstract

Some CTs had the core study period (CSP) and extension period (EP), with subjects randomized to be treated or untreated in CSP, and untreated subjects allowed receiving trt in EP. To examine trt effect at the end of EP, some planned analyses compared endpoint between treated group (TG) and untreated group (UG) by defining TG as subjects who ever received trt. Also, baseline was defined differently: for a subject without trt switching, baseline was the start of CSP; otherwise, the start of EP. Simulation was performed to compare such planned analysis and other methods: repeated measures analysis (RM), GEE, and marginal structural model (MSM) with stabilized weight (SW) and nonstabilized weight (NSW). The results showed that MSM with SW and GEE were most reliable, that MSM with NSW and planned analysis often had large variation, and that it was difficult to interpret the results of RM.