Adaptation of Bayesian decision analysis for clinical trial design to binary outcomes: illustration for Ebola virus disease treatment.

Abstract

When designing trials for high-mortality diseases with limited therapies, the conventional 5% type I error rate ($\alpha$) may be unsuitable. A higher $\alpha$ could benefit seriously ill patients lacking therapeutic alternatives. Bayesian decision analysis (BDA) integrates multiple health consequences into trial design. We extend BDA for trials with binary outcomes to calculate optimal sample sizes (n) and α in an Ebola outbreak context. We consider fixed, two-arm randomized trials with binary outcomes and two types of clinical trial losses: post-trial loss, for not approving an effective treatment or approving an ineffective one; in-trial loss, for not administrating an effective treatment to control-arm patients or for administrating an ineffective treatment to experimental-arm patients. The model incorporates treatment side effects and disease severity into a single loss function, from which BDA-optimal n and α are derived. Using mortality as the outcome, we modeled various epidemic scenarios by varying target population size, mortality rate, and treatment efficacy. In most cases, BDA-optimal α exceeded the conventional one-sided 2.5% and sample sizes are smaller. Simulations comparing BDA-optimized two-arm trials (fixed/sequential) to conventional designs (two-arm/single-arm, fixed/sequential) across outbreak scenarios show comparable statistical power across designs, except when sample size assumptions are incorrect, or when trials started after the outbreak peak; in these situations, BDA-optimized trials show superior powers. This BDA framework provides a promising approach for designing trials with binary outcomes, particularly in high-mortality disease outbreaks, where declining case numbers and uncertain mortality and efficacy data challenge conventional methods.