Original title: Bayesian Power‐Based Sample Size Determination for Diagnostic Accuracy Studies
Shunsuke Shiota, Go Horiguchi, Akari Naito, Mitsuko Nakata +1 · September 2026
When researchers test new medical tools or drugs, they need to check diagnostic measures like sensitivity, which spots true sickness, and specificity, which spots true health. Before starting, they must figure out the right sample size, or how many people to test. While standard math rules are common, Bayesian methods, which use prior beliefs, are gaining interest. The researchers built a new way to pick this sample size using Bayesian power and two types of priors. One prior guides the final analysis, while another generates possible future study data. For every potential dataset, the method updates the analysis belief to test if a success goal is met. The overall power is the predicted chance of meeting that goal. They found that less uncertainty in the data-generating priors generally reduced the needed sample size when prior averages topped target goals. Also, picking a more optimistic analysis prior lowered the required sample size.
Imagine you are a coach deciding how many free throws a player should attempt in practice to reliably tell if they've improved. You start with some belief about their current skill level — maybe you've seen them shoot before. Now, you want to know: if they attempt 50 shots, how often will their performance actually convince you they've improved, given that your belief might be off? The less certain you are about your initial belief, the more shots you will need to be convinced. In this paper, the coach's initial belief is the design prior, the standard used to judge improvement is the success criterion, and the number of free throws is the sample size.
Diagnosing a disease correctly matters — a test that misses cases or gives false alarms can hurt people. Before you run a study to check how good a diagnostic test is, you need to know how many patients to include. Get it wrong and the study is either too small to be reliable or wastes resources. This paper gives researchers a Bayesian method to plan the sample size more flexibly, using prior knowledge about the test and the disease, which can lead to smaller or more efficient studies.
The abstract's numerical results are limited to the specific parameter settings considered — they do not claim general rules for all diagnostic studies. The method requires specifying analysis priors, design priors, and a success criterion, which may be subjective or difficult to elicit in practice. The comparisons with frequentist methods are mentioned but not detailed here, so the relative advantages in real-world settings are not fully established. Prevalence uncertainty is only handled via a design prior, not an analysis prior.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Bayesian Power‐Based Sample Size Determination for Diagnostic Accuracy Studies — Shunsuke Shiota, Go Horiguchi, Akari Naito et al. — Pharmaceutical Statistics — doi:10.1002/pst.70122
Abstract licence: CC BY
Explanation generated by Synthesis from the abstract — it is not the original paper.