Ask Mallard

Adaptive and group-sequential trials

An adaptive trial allows changes to the design while it runs, according to rules written down before it starts, using accumulating data without invalidating the conclusions.
NHMRC EVIDENCE LEVEL · 2009Level II · InterventionProvisional design classification. Study quality and certainty of findings require separate appraisal.
Why this level?Participants or clusters are randomly allocated to concurrent comparison groups. NHMRC 2009 source

Adaptive trials with genuine random allocation retain the randomised design category. Prespecification and control of error rates require separate appraisal; the NHMRC level does not certify them.

What question does this help answer?

The same question a fixed trial answers, but with the possibility of stopping early for benefit or futility, dropping an arm, or re-estimating the sample size, so that fewer participants are exposed to a treatment that is not working.

The distinction that matters

PRESPECIFIED adaptation preserves validity. Unplanned change does not. The difference is not a technicality: a rule written before the data exist constrains the analysis, while a change made after seeing results has already used the data twice. An adaptive trial's protocol must therefore state what will be looked at, when, and what each result triggers.

Planned and unplanned changes

Two ways a trial can change midway. Prespecified: an interim analysis leads to options written into the protocol in advance, stopping for benefit, stopping for futility, or continuing with an allowed modification, and the final analysis accounts for the look. Unplanned: someone looks at the results and makes a change the protocol did not plan, and the error rate is no longer controlled.Prespecified adaptationFirst stageInterim analysis, planned in the protocolStop forbenefitStop forfutilityContinue, asallowedFinal analysis accounts for the lookUnplanned changeFirst stageResults looked atA change the protocol did not planError rate no longer controlledStopping early for benefit tends to overstate the effect even when the rule was prespecified, so theestimate needs a method that accounts for the stopping rule.
Conceptual diagram. It shows the structure of the idea, not data from any study.

When to use it

Use it when there is genuine uncertainty about the effect size or the right dose, when accumulating data could reasonably change the design, and when a statistician can build and simulate the design in advance.

When not to use it

Avoid it when the outcome takes so long to observe that interim data would be uninformative by the time they exist, and when the added complexity cannot be supported by the trial team.

Why Mallard may recommend it

Mallard may raise an adaptive design when the plan describes substantial uncertainty about the effect size, but it will not design one: the operating characteristics require simulation specific to the trial.

Key assumptions and design requirements

  • Every adaptation rule is prespecified, including the timing and number of interim looks.
  • Type I error is controlled across looks with an alpha-spending function or equivalent.
  • Operating characteristics are established by SIMULATION before the trial opens.
  • An independent data monitoring committee applies the rules; the trial team stays blinded to interim results.
  • An effect estimated from a trial stopped early for benefit is biased upward and is reported with that stated.

What to watch out for

Prespecified adaptation preserves validity; unplanned change does not, and the distinction is the entire point. Interim looks spend type I error and must be accounted for, and stopping early for benefit systematically OVERSTATES the effect size, so an early-stopped trial's estimate should be read as an upper bound rather than the answer.

Clinical example

Question: Which of three doses is best, and does the best one beat control? Design: A prespecified two-stage trial that drops the least promising doses at an interim analysis and continues the rest, with alpha spending across looks and operating characteristics established by simulation before enrolment.

How to report it

ACE, the Adaptive designs CONSORT Extension, is the guideline for this design. It asks you to report the adaptations that were planned and the rules for making them, when interim analyses were run, which changes were actually made and whether any were unplanned, and how the analysis kept the conclusions valid. A reporting guideline is not a design manual: use it to make the study transparent, not as a substitute for careful design. Check the target journal's own requirements as well, which this page cannot know.

Recent studies that used this
Searching PubMed…
Related Learn pages
Sources & further reading
  1. The Adaptive designs CONSORT Extension (ACE) statement: a checklist with explanation and elaboration guideline for reporting randomised trials that use an adaptive design (opens in a new tab) · BMJ · doi 10.1136/bmj.m115
  2. Overview of clinical study designs (opens in a new tab) · Clin Exp Emerg Med · doi 10.15441/ceem.23.036
ACE is a reporting guideline and is explicit about distinguishing prespecified from unplanned changes. It does not tell you how to construct an adaptive design, which requires trial-specific simulation and a statistician.
Last source review: 2026-09-06
AI-generated educational content
Mallard uses reputable methodological and reporting sources, but this material may contain errors or omit important context. Verify important methodological decisions with the cited sources and, when appropriate, a qualified research methodologist or statistician.