Sample-Size Calculation for Time-to-Event Outcome with Treatment Lag Effect

• Activity Number: 345
• Type: Contributed
• Date/Time: Tuesday, August 3, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Yulei Zhang*+ and Tai-Tsang Chen
• Companies: Columbia University and Bristol-Myers Squibb/Columbia University
• Keywords: sample size ; power function ; time-to-event ; treatment lag ; design comparison

Abstract

We reviewed classical sample size determination methods as Freedman (1982) and Rubinstein, Gail and Santner (1981) for constant hazard ratio design. For non-constant hazard ratio design, we reviewd Lakatos' Markov method and treated it as our starting point.

To obtain sample size addressing more flexible follow up duration, treatment lag length, non-compliance pattern and abrupt treatment effect change, Lakatos' algorithm was modified. We also rewrote its SAS code with Proc IML to replace the older version of Proc Matrix. The power function along increasing number of events was obtained.

To illustrate how the treatment lag length influenced the power accumulation pattern, we compared the power functions with different treatment lag length designs. It helps us decide whether sample size is feasible in different scenario of treatment lag effect.

Numerical examples were provided.