Adaptive Trial Design

An adaptive design is a trial whose conduct can change while it runs, according to rules written into the protocol before the first participant is enrolled. Permitted adaptations include dropping arms that perform poorly, re-estimating the sample size, narrowing enrolment to a responding subgroup, or moving seamlessly from dose-finding into a confirmatory stage. The defining feature is not flexibility but prespecification: every possible change and its statistical consequence is decided in advance, so the overall false-positive rate stays controlled.

Regulators treat this as mature methodology rather than experiment. The FDA issued final guidance on adaptive designs for drug and biologic trials in 2019. The best-known examples are platform trials: I-SPY 2 in breast cancer, which adds and drops agents against a shared control, and RECOVERY during the pandemic, which produced a definitive answer on dexamethasone faster than a series of separate trials could have.

The payoff is efficiency in a specific sense: fewer participants exposed to arms already looking futile, and less calendar time between a dose-finding signal and a confirmatory answer. The cost is complexity, since simulation is usually needed to characterise the operating characteristics, and interpretation is harder because the population and the allocation may have shifted mid-study.

The abuse to watch for is adaptation dressed up after the fact. A protocol amended once the data have been seen, an arm quietly dropped without a prespecified rule, or a subgroup chosen because it looked good at an interim is not an adaptive design; it is data-dependent analysis with a respectable name attached. The registry entry and the analysis plan, timestamped before the looks, are what separate the two.