Article in the latest issue of Science Translational Medicine co-authored by Peter Bacchetti, Director of the CTSI Biostatistics Consultation Service and and CTSI Board Director Joseph M. McCune.
Abstract
Innovative clinical and translational research is often delayed or prevented by reviewers’ expectations that any study performed in humans must be shown in advance to have high statistical power. This supposed requirement is not justifiable and is contradicted by the reality that increasing sample size produces diminishing marginal returns. Studies of new ideas often must start small (sometimes even with an n of 1) because of cost and feasibility concerns, and recent statistical work shows that small sample sizes for such research can produce more projected scientific value per dollar spent than larger sample sizes. Renouncing false dogma about sample size would remove a serious barrier to innovation and translation.
CONCLUSIONS
As we have argued in this Perspective, the demand for conventional sample size calculations raises substantial barriers to the conduct and completion of innovative, bench-to-bedside translational research. Enforcement of the requirement for at least 80% power has no valid justification and is especially inappropriate for early translational studies investigating new ideas. Practical considerations, including the costs of larger versus smaller sample sizes, inevitably drive sample size choices, and the reality of diminishing marginal returns implies that this is scientifically valid. Indeed, the use of nroot determined only by costs is reasonable for innovative studies. Acceptance of this and other alternatives to current conventions would remove a formidable barrier to the conduct of innovative translational research.
Keep reading
