Comparing Different Statistical Approaches in Analysis of Oncology Trial Data with Patient Crossover

• Activity Number: 291
• Type: Topic Contributed
• Date/Time: Wednesday, August 11, 2004 : 8:30 AM to 10:20 AM
• Sponsor: Biopharmaceutical Section
• Author(s): Tailiang Xie*+
• Companies: Imclone System, Inc.
• Keywords: treatment crossover ; survival data ; causal inference ; cancer clinical trial

Abstract

Treatment crossover from an active control therapy to an experimental therapy occurs commonly in randomized oncology clinical trials, where continuation of therapy in active control arm when disease progression was observed is considered unethical. When treatment crossover occurred, treatment effect of the experimental therapy was confounded and assessment of treatment benefit became very difficult. If the experimental therapy had a beneficial treatment effect, treatment crossover from active control would result in an underestimation of treatment difference, whereas if the experimental therapy had some harmful effects, treatment crossover from active control would result in an over-estimation of treatment difference. Except for many naïve approaches, there is no established method to handle clinical data with treatment crossover. We will discuss commonly used approaches for analyzing survival data in which treatment crossover had happened. We will also propose several alternative approaches. These approaches will be compared via extensive simulations and real examples.