The FDA just issued for comment the ICH E20 draft guideline on adaptive designs in confirmatory clinical trials. (Until finalized, existing regional guidances still govern.)
Here are a few of our key takeaways:
▪️ An “adaptive design” is a prospectively planned design that allows pre-specified modifications to one or more aspects of an ongoing trial based on interim analyses of accumulating data from that same trial. Unplanned amendments or changes driven only by external data are out of scope.
▪️ A few things can adapt: early stopping (group sequential), sample size, population selection, treatment/dose selection, and response-adaptive randomization.
▪️ Sponsors have to satisfy a few principles:
1/ Fit within the total program. Don’t use an adaptive confirmatory trial to substitute for earlier dose-finding or foundational learning. Limit the number and complexity of adaptations at the confirmatory stage and justify why an adaptation is needed given residual uncertainty.
2/ Adequate planning and pre-specification. Before the first patient in: define interim timing, which features may adapt, the anticipated adaptation rule(s), estimands and analysis methods for each possible path, and how trial integrity will be protected. If simulations are needed to understand operating characteristics, plan, execute, and report them rigorously.
3/ Control erroneous conclusions. Use analysis methods that control Type I error under the planned adaptations and any multiplicity across endpoints/populations. Evaluate power impacts and ensure sufficient information for safety and benefit-risk, especially if early stopping is possible.
4/ Reliable estimation. The primary estimator should align with the targeted estimand and have limited to no bias with accurate uncertainty (CIs with correct coverage). Account for selection/early-stop bias; use adjusted estimators where available. Time your interim such that the estimates used for decisions are stable enough to make good selections.
5/ Maintain trial integrity. Keep sponsors, investigators, and participants blinded to treatment-group summaries. Use an independent DMC to review unblinded interim data and make recommendations.