Journal of Biometrics & Biostatistics

On Power Calculation Based on Effect Size in Clinical Research

Abstract

Yinuo Zhang* and Shein-Chung Chow

Power analysis for sample size calculation (power calculation) plays an important role in clinical research to guarantee that we have sufficient power for detecting a clinically meaningful difference (treatment effect) at a pre-specified level of significance. In practice, however, there may be little or no information regarding the test treatment under study available. In this case, it is suggested that power calculation for detecting an anticipated effect size adjusted for standard deviation be performed, reducing a two-parameter problem into a single parameter problem by taking both mean response and variability into consideration. This study systematically analyzes estimating sample sizes across diverse endpoints, including relative/absolute change, risk metrics, exponential and proportional hazards models. Findings underscore the distinct nature of these metrics, reinforcing the necessity of an effect size measure as a standardized framework. Notably, analysis suggests it is possible to translate continuous and binary outcomes through a common effect size metric, which could facilitate meta-analyses involving heterogeneous outcome types. However, extending such translations to time-to-event outcomes presents additional complexities warranting advanced modeling techniques and hazard-based metrics. Through critical examination of effect size-based power calculations, this study contributes insights into efficient sample size estimation. It highlights the importance of standardized effect sizes as a unifying measure and the potential for outcome translation across endpoints.