Adaptive Clinical Trial Designs

Adaptive Clinical Trial Designs

1. Definition & core concept

• An adaptive clinical trial design is a trial that allows pre-planned modifications to one or more aspects of the design (sample size, allocation, treatment arms, study population, endpoints) based on interim analyses of the accumulating data, while preserving statistical validity and integrity (Browner 5e Ch.12, pp.7–8 — "adaptive designs that allow changes to the trial protocol based on interim analyses of the results may be preferable").
• Contrast with the classic ("fixed") trial: conducted according to a protocol that does not change during the study — sample size, allocation ratio, doses and analysis are all fixed in advance (Browner 5e Ch.12, p.7; Ch.11, pp.2–3).
• The defining feature is that the adaptation rules are prospectively specified in the protocol/statistical analysis plan before the trial begins and before any data are examined — adaptation is "by design," not a reactive, post-hoc protocol amendment (Browner 5e Ch.12, p.8 — "rules for how such changes may be made should be established before the trial begins"). [FDA]
• FDA working definition: an adaptive design is "a clinical trial that allows for prospectively planned modifications to one or more aspects of the design based on accumulating data from participants in the trial." The two operative phrases are prospectively planned and based on accumulating data. [FDA]
• Adaptive designs are feasible only when outcomes are measured and analysed early enough to allow design changes to be made while the trial is still recruiting (Browner 5e Ch.12, pp.7–8). A long-latency outcome (e.g. dementia incidence over 5 years) leaves no window to adapt; a short-term continuous or surrogate endpoint does.
• Adaptation is information-driven, not calendar-driven in principle — the decision to adapt is triggered by reaching a pre-set quantity of accrued outcome data (the "information fraction"), although interim looks are often scheduled at planned calendar/enrollment milestones for logistical reasons (Browner 5e Ch.11, pp.28–29). [FDA]
• Two governing goals that every adaptive design must satisfy simultaneously: (i) preserve the chance of an erroneous positive conclusion (the overall Type I error probability, conventionally ≤ 0.05 two-sided); and (ii) yield treatment-effect estimates with minimal bias and adequate reliability. Both can be threatened by the act of looking at and acting on interim data. [FDA] (Browner 5e Ch.11, pp.28–29)