Do we need to adjust for interim analyses in a Bayesian adaptive trial design?

Area of Science:

• Biostatistics
• Clinical Trial Design

Background:

• Bayesian adaptive methods offer advantages in clinical trial design.
• Decisions in these trials typically rely on posterior distributions.
• A key debate concerns the necessity of controlling type I error in Bayesian designs, a frequentist concept.

Purpose of the Study:

• To examine the impact of interim analyses on error rates in Bayesian adaptive trials.
• To discuss adjustments for multiplicity and alternative decision-making strategies.
• To address regulatory requirements for type I error control.

Main Methods:

• Analysis of two case studies involving Bayesian adaptive designs with interim analyses.
• Evaluation of type I and type II error rates under different stopping rules (efficacy, futility, or both).
• Exploration of methods to control type I error and alternative Bayesian decision frameworks.

Main Results:

• Bayesian adaptive designs with early stopping for efficacy inflated type I error without multiplicity adjustments.
• Early stopping for efficacy sometimes increased statistical power.
• Interim analyses for futility only decreased type I error but also reduced power.
• Multiple interim analyses allowing both efficacy and futility stopping generally increased type I error and decreased power.

Conclusions:

• Regulators mandate type I error control for Bayesian adaptive designs, necessitating boundary adjustments for early efficacy stopping.
• Adjustments are not required for futility-only stopping rules.
• Strict Bayesian approaches, suitable for exploratory trials, may allow focus on posterior probabilities, potentially disregarding type I error control.