Conventional randomised controlled trials follow a fixed design: the sample size, endpoints, and analytical approach are determined at the outset and remain unchanged. This rigidity was designed to protect against bias, but it carries a cost — it makes trials slow to respond to new information emerging during the study itself. Adaptive designs offer a principled alternative, allowing pre-specified modifications to the trial based on accumulating data while preserving statistical validity.
An adaptive trial is any trial that uses pre-planned rules to modify one or more aspects of the study design after enrolment begins, based on interim analysis of the trial's own data. The modifications are pre-specified in the protocol and statistical analysis plan — the key word is pre-specified. Unplanned changes made after examining unblinded outcome data are not adaptive designs; they are protocol deviations.
Common adaptations include:
A platform trial is a specific type of adaptive design that evaluates multiple treatments simultaneously against a shared control arm, within a single overarching protocol. The defining feature is that treatments can be added or dropped over time based on pre-specified criteria, without the trial stopping. This infrastructure — the platform — persists while the experimental arms turn over.
The efficiency gains are substantial. A shared control arm reduces the total number of participants needed compared to separate trials, accrual is faster because the platform is continuously open, and operational costs are distributed across sponsors. The RECOVERY trial, which evaluated multiple COVID-19 treatments simultaneously using a single protocol and shared infrastructure, demonstrated the speed at which a well-designed platform can generate practice-changing evidence.
Master Protocol terminology: The term "master protocol" describes the overarching protocol governing the platform. Individual experimental treatments are typically described in sub-protocols or appendices. Basket trials (multiple tumour types, single target) and umbrella trials (single tumour type, multiple targets) are related designs that share infrastructure but are typically not perpetual platforms.
The principal statistical challenge in adaptive designs is protecting the overall Type I error rate — the probability of incorrectly concluding that an ineffective treatment works. Unplanned interim analyses inflate Type I error; pre-specified analyses with appropriate alpha-spending functions do not.
Alpha-spending functions determine how much of the total available Type I error is "spent" at each interim look. Common choices include the O'Brien-Fleming boundary (conservative at early looks, liberal at later looks) and the Pocock boundary (uniform spending across looks). The choice should be specified before the first interim and reviewed by an independent DSMB.
Response-adaptive randomisation introduces additional complexity. While it can increase the proportion of participants receiving a superior treatment — an ethical advantage in some contexts — it can also reduce statistical power and introduce selection bias if poorly implemented. Stratified randomisation and careful monitoring of balance across arms are essential.
The FDA has issued guidance on adaptive designs for drugs and biologics (2019) that explicitly endorses well-designed adaptive trials while emphasising the importance of pre-specification and operational integrity. The guidance is notably supportive of seamless designs and master protocols, and signals FDA's willingness to engage at early-stage meetings on complex adaptive designs. Sponsors who plan adaptive elements should include them in pre-IND or Type B meeting requests to ensure alignment with regulatory expectations before committing to a design.
Operationally, adaptive trials require robust firewall procedures — typically involving an unblinded Statistical Analysis Centre (SAC) and an independent DSMB — to ensure that the adaptation decisions are made without compromising the blinding of the investigators and sponsor.
Participating in an adaptive trial as a site investigator requires familiarity with how your enrolment decisions may change over the course of the trial. Eligibility criteria may narrow if the trial enriches for a biomarker subgroup; your randomisation ratio may shift if response-adaptive allocation is used; and the trial may stop early for efficacy or futility in one arm while continuing in others.
At KCLEAGENICS MEDICAL, we provide comprehensive site training on the adaptive elements of any trial we run, with clear documentation of how and when adaptations may occur and what information will be shared with sites at each stage. For clinicians interested in discussing co-investigator participation in our current platform studies, please contact our research team.
Pre-specification in Clinical Trials
For CliniciansBiomarker-Driven Patient Selection in Precision Oncology
For CliniciansMetabolic Risk and Haematological Malignancy
Subscribe to our monthly Clinical Intelligence Briefing
