thebench.bio #1 Ecosystem for the Bio-Tech Industry
All Clinical Trial Design questions

Question 19 of 20 Advanced Scenario

When should an adaptive design be preferred, and what operational biases must it control?

Short answer

Prefer an adaptive design when real uncertainty, such as the effect size, the best dose or the responsive subgroup, could be resolved at an interim look and the outcome is known quickly enough to act on. It must keep type I error controlled, estimates reliable, adaptations fully pre-specified and interim results confidential.

Why

FDA's final guidance on adaptive designs for drugs and biologics, issued in November 2019, sets four principles: control the chance of erroneous conclusions, estimate treatment effects reliably, pre-specify the design completely, and maintain trial integrity. Adaptive designs include group sequential stopping, sample size re-estimation, dropping arms, enrichment of a subgroup and response adaptive randomization. They pay off when an early look can answer a question that would otherwise need a second trial.

They pay off less when the outcome takes longer to observe than enrollment takes, because most participants are enrolled before anything can change. The operational risk is leakage. If trial staff infer interim results, from a dropped arm or a changed sample size, their behavior can shift in ways no analysis can correct.

So access to comparative results is limited to an independent group, and communication of decisions is designed to reveal as little as possible. FDA lists limitations too: complex methods often need simulation, the maximum sample size may rise, and results before and after an adaptation may differ. ICH E20 on adaptive designs was still a draft when this was reviewed.

How to work it through

  1. State the uncertainty the adaptation is meant to resolve.
  2. Check that the outcome will be observed early enough for an interim decision to matter.
  3. Pre-specify the number and timing of interim analyses and the decision rules.
  4. Simulate the design to show type I error control, power and bias in estimates.
  5. Set up an independent committee and firewalls for comparative results.
  6. Plan what decisions will be communicated and how.

An example

A phase 2 and 3 study plans three doses against placebo with a 12 week endpoint and 18 months of enrollment. At an interim look after 240 participants have reached week 12, the committee keeps the best dose by a pre-specified rule. Simulations of 10,000 trials show the overall one sided type I error stays at 0.025 under the null, with 88 percent power for the expected effect.

Sites are told only that two arms have closed, not which dose looked best or by how much.

Where the answer stops holding

When the endpoint takes a year to observe and enrollment takes a year, an adaptation arrives too late to help. Small sponsors may lack the infrastructure for firewalls. Regulators may still question a design whose simulations rely on optimistic assumptions.

Guidance may change once ICH E20 is final.

A strong answer covers

  • Four FDA principles for adaptive designs
  • Benefit depends on outcome timing relative to enrollment
  • Pre-specification and simulation
  • Firewalls against operational bias
  • ICH E20 status as a draft

Common mistakes

  • Treating an adaptive design as permission to change the trial whenever the data suggest it.
  • Assuming adaptive always means a smaller trial.
  • Letting the sponsor team see interim comparisons to plan the next study without firewalls.