The Use of 'Fuzzy Promising Zones' to Mitigate Against Operational Bias in Adaptive Sample Size Re-Estimation Designs

• Activity Number: 225
• Type: Topic Contributed
• Date/Time: Monday, August 4, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Section on Medical Devices and Diagnostics
• Author(s): Brendan Keenan*+ and Greg Maislin
• Companies: Biomedical Statistical Consulting and Biomedical Statistical Consulting
• Keywords: Adaptive Design ; Sample Size Re-estimation ; Promising Zone ; Operational Bias ; Conditional Power

Abstract

Research has demonstrated that adaptive designs with sample size re-estimation based on assessments of a 'promising zone' can improve overall power without increasing type I error. However, potential exists for operational bias, since the treatment effect size can be inferred. To mitigate this bias, we propose a novel 'fuzzy promising zone' approach. A small sample size expansion may occur even when results fall outside the promising zone, based on a Poisson(?) random draw. We utilize a real-world example to explore the operating characteristics of this type of design. Under various ?, 10 sets of 5,000 trials were simulated to estimate type I error (a) and unconditional power. When ?=3, a was maintained at <0.025; power was at least as good as the original design. Moreover, the impact on expected sample size is small. We conclude that the introduction of "fuzzy promising zones" to sample size re-estimation designs has negligible impact on operating characteristics while reducing the potential for operational bias. Specifically, for small values of the distributional parameter ?, the design maintains the power increases with minimal expected sample size increases and without inflating type I error.