Cornelius, Victoria R; Elkes, Jack; White, Ian R; Turner, Rebecca M; Clements, Michelle; Quartagno, Matteo; Tweed, Conor D; ... Cro, Suzie; + view all Cornelius, Victoria R; Elkes, Jack; White, Ian R; Turner, Rebecca M; Clements, Michelle; Quartagno, Matteo; Tweed, Conor D; Saglani, Sejal; Cro, Suzie; - view fewer (2025) A Bayesian framework to evaluate non-inferiority in randomised controlled trials of uncommon conditions. Journal of Clinical Epidemiology , Article 112002. 10.1016/j.jclinepi.2025.112002. (In press).

Text Quartagno_Designing trials for NI revision FINAL.pdf Access restricted to UCL open access staff until 30 September 2026. Download (511kB)

Abstract

BACKGROUND: Non-inferiority (NI) trials typically require larger sample sizes than superiority comparisons. This is problematic for uncommon conditions where recruitment is restricted. When a power calculation results in an unfeasibly large sample size, there is a need to justify whether a trial in an uncommon condition is worth undertaking. We propose reversing the question to demonstrate what can be shown with a feasible maximum sample size using a Bayesian framework in place of a traditional power calculation. METHODS: We propose using the posterior probability of non-inferiority to demonstrate the value of undertaking a NI trial. We describe a Bayesian framework and its five inputs: the data generation/analysis model; maximum feasible sample size; different potential outcomes; primary NI margin; and plausible priors. We illustrate the framework in an NIHR funded NI trial of mepolizumab compared to omalizumab in children with severe therapy resistant asthma. We examine the trial operating characteristics when mepolizumab is inferior, identical, and superior to omalizumab, under four differing prior distributions on the treatment effect (3 informative, 1 vague), and suggest suitable interpretation. RESULTS: Our case study had a maximum feasible sample size of 150 severe therapy resistant asthmatic children. Using the proposed Bayesian framework we demonstrated that if mepolizumab was truly identical or superior to omalizumab then the average posterior probability of non-inferiority would be reassuringly high, from 0.87 to >0.99. The probabilities were reassuringly low when mepolizumab was inferior (≤ 0.22). The framework provided a comprehensible summary for reviewers to judge the value of undertaking the trial. CONCLUSION: A Bayesian approach using posterior probability for NI trials can offer a practical way to assess the value of undertaking a trial in an uncommon condition with a fixed sample size as an alternative to a power calculation.

Download activity - last month

Export as CSVXMLJSON

Download activity - last 12 months

Export as JSONXMLCSV

Downloads by country - last 12 months

Export as CSVXMLJSON

Archive Staff Only

View Item