231 – Data Collection, Processing, and Analysis in the Energy Industry
Uncertainty in Pilot Parameter Estimates: A Comparison of Methods to Size Full Trials
Elizabeth Handorf
Fox Chase Cancer Center
Eric A. Ross
Fox Chase Cancer Center
Pilot data is a valuable resource for sizing full studies, but parameter estimates based on pilot data can be highly variable. This can lead to substantial under- or overestimation of the necessary sample size, resulting in low power or wasted resources. Several approaches may help decrease the likelihood of underpowering a study, including the use of upper confidence limits of pilot parameter estimates, adaptive approaches with planned interim analyses, and sample size re-estimation without unblinding. In this study, we explore how various methods affect estimated sample size and resulting power. We show that certain re-estimation strategies may substantially improve power, particularly when the pilot study is small. Furthermore, although using upper confidence limits provides increased power, this method leads to large overestimates of sample size. Finally, we show that in most cases considerable benefits come from increasing the number of observations in the pilot study.