Estimating and Using the Attained Power Distribution to Ensure We Get the Trial Power We Expect

• Activity Number: 426 - SPEED: Biopharmaceutical and General Health Studies: Statistical Methods and Applications, Part 2
• Type: Contributed
• Date/Time: Tuesday, July 30, 2019 : 3:05 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Yongdong Ouyang* and Hubert Wong and Ehsan Karim and Paul Gustafson
• Companies: University of British Columbia and University of British Columbia and University of British Columbia and University of British Columbia
• Keywords: Clinical Trial; Attained power; Stepped-wedge Design; Power

Abstract

The attained power (AP) in a clinical trial can depend on the allocation-a trial that randomized 15 people to one arm and 25 people to the other arm has a lower AP than if it randomized 20 people to each arm. Because trial power is quoted as a single “design” value (which in principle should reflect the average power generated by the randomization algorithm (RA)), the AP may be substantially lower than the design power. This risk can be judged using the attained power distribution (APD) of the RA which describes the probability that the AP will fall below any given value. In a stepped-wedge (SW) design with unequal cluster sizes, this risk may be unacceptably high. Identifying allocations which yield low AP would enable use of randomization restrictions to construct RAs which achieve an acceptable APD. We use simulation to evaluate the AP of different allocations and to identify predictors of low AP. Typically, the number of possible allocations is too large to allow evaluating them all. Based on predictive models built on random samples of allocations, we found that the AP can be predicted accurately using the correlation between (individual) treatment allocation and time-step.