Simulations on Comparisons Between 4-Factor Model and 3-Factor Model Using Stratified Logrank Test: A Case Study

• Activity Number: 598
• Type: Contributed
• Date/Time: Wednesday, August 7, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Shaoyi Li*+
• Companies: Celgene
• Keywords: prognostic factor ; randomization ; stratification ; logrank test ; controlled clinical trial ; simulation

Abstract

In controlled clinical trials, many prognostic factors seem all qualified for randomized stratification. However, too many strata may jeopardize the power of the stratified statistical test. Where to make the cut so that treatment allocation will be balanced for key stratification factors and clinicians don't worry about the down side? The author feels that the simulation approach could be a practical way to get some sense if the answer is not obvious. A 4-factor model vs. 3-factor model is considered to see which model would have more statistical power using stratified logrank test. Several typical scenarios based on number of patients in each stratum are explored: a. equal case; b. slightly different; c. dramatic different; d. a real case from literature with similar design and indication. Simulation results demonstrate that, for the given sample size, 4-factor model may be more powerful for cases a & b while 3-factor model may be more powerful for cases c & d. The allocation of patients in each stratum seems the major cause for the difference. For this particular case, the 3-factor model is recommended statistically. The 4-factor can be considered if medical reason is justified.