A Pragmatic Approach to Blinded Sample Size Reestimation

• Activity Number: 176
• Type: Contributed
• Date/Time: Monday, August 4, 2003 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Anthony J. Rodgers*+ and Norman R. Bohidar and Erluo Chen
• Companies: Merck & Co., Inc. and Merck & Co., Inc. and Merck & Co., Inc.
• Keywords: sample size ; sensitivity ; risks

Abstract

Although both blinded and unblinded approaches to sample size reestimation exist, the blinded approach is less prone to controversy. However, there are no straightforward decision rules that can be applied to decide whether a change (increase or decrease) of sample size is necessary based on a blinded sample size reestimation. Without unblinding, there are many unknowns. For example, it is impossible to differentiate the source of variability (between group or within group). However, if there is supportive prior knowledge of the treatment effect, blinded reestimation can be a useful way of protecting against misspecification of variance. The "Simple Approach" for blinded sample size reestimation when there are only two treatment groups, proposed by Gould and Shih (1992) will be described. A generalization of this approach to cases in which there are more than two treatment groups was developed and will be described. An example will be presented summarizing results from a series of simulations to assess the sensitivity of this procedure. Risks associated with choice of criteria for increasing the sample size used in the procedure will be discussed.