Adaptive Sample Size Re-Estimation with Fisher's Exact Test and a Promising Zone

• Activity Number: 188
• Type: Contributed
• Date/Time: Monday, August 10, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Section on Medical Devices and Diagnostics
• Author(s): Tyson Rogers* and Scott McKane
• Companies: NAMSA and Respicardia, Inc.
• Keywords: Adaptive design ; Sample size re-estimation ; Fisher's exact test

Abstract

The adaptive sample size re-estimation method described by Mehta and Pocock (2011) that increases the sample size when conditional power at the interim falls into a "promising zone" has some attractive features. It permits equal weighting of each patient's data so that the conventional test statistic and critical value may be used at the end of the trial and maintains the planned alpha level by restricting the range of interim results that lead to a sample size increase. For moderate sized clinical trials with binary endpoints, Fisher's exact test is a common method to test the hypothesis whether the proportions in two groups differ. Since the Fisher's exact test statistic has a hypergeometric distribution, it's not obvious how to implement the promising zone method which was derived with normal distribution theory. We present a practical approximation and exact calculation for determining the sample size increase to yield the specified conditional power with Fisher's exact test. We calculate the exact Type I error for the adaptive design. The exact calculations apply generally for evaluating adaptive sample size re-estimation based on Fisher's exact test conditional power.