Randomised trials should always report results by randomised treatment group to transparently display the available evidence on the causal treatment effects. This approach is a challenge for adaptive trials because crude treatment comparisons are not randomised evidence and might have confounding originating from changes in design during the study. Statistical models can be used to deal with this confounding, but the results of modelled analyses do not provide a clear summary of the core data underlying the randomised comparisons. This method prevents transparency and hinders the reuse of trial results by other researchers in systematic reviews and meta-analyses. Having guidance on the transparent display of core results from adaptive trials is therefore important.
We have introduced a framework for transparent reporting of results from adaptive clinical trials based on the fundamental concept of a concurrently randomised cohort, a cohort of participants that all had the same treatments available and were allocated to these treatments in the same way. We have presented this framework in the context of individually randomised parallel studies, the most common design used in adaptive trials. We have also introduced a range of terms that are useful additions and clarifications to the lexicon of adaptive trials. Box 1 provides a summary of the key terminology for describing our proposed reporting framework. This terminology includes a clear distinction between the concepts of concurrent and contemporaneous cohorts, the former denoting randomisation within the same embedded fixed design and the latter denoting randomisation within the same stage. Our reporting framework is motivated by concurrent randomisation being the primary basis for unbiased treatment comparisons.
Box 1Key terminology
Adaptation: change to available treatments or the way participants are assigned to treatments
Stage: time period between adaptations when the design is fixed
Participant randomisation scheme: individual listing that captures, for each participant, the treatments that were available to them and their chance of receiving these treatments
Embedded fixed design: one of the unique randomisation configurations occurring within the participant randomisation scheme, which is equivalent to a fixed design embedded within the overall adaptive trial
Concurrently randomised cohort: cohort of participants who had the same treatments available and the same chance of receiving these treatments
Contemporaneous cohort: cohort of participants who were randomised within the same stage, although not necessarily within the same concurrently randomised cohort. Participants might be contemporaneous but not concurrent, and vice versa
Our primary recommendation is that all adaptive trials should provide concurrently randomised cohort reporting. This approach requires identification of the fixed designs that are embedded within the overall adaptive design and construction of all of the participant cohorts that were randomised under the same fixed design. Concurrently randomised cohort reporting then involves display of baseline characteristics and outcomes by treatment group for each of these cohorts. Box 2 summarises our key recommendations.
Box 2Key recommendations
All adaptive trials should report key results by concurrently randomised cohort, termed concurrently randomised cohort reporting, and have the following elements:
For transparent reporting of the randomised design, each embedded fixed design within an adaptive trial should be identified and reported. The elements of an embedded fixed design are the available treatment arms, randomisation probabilities for each treatment arm, and any factors used to stratify the randomisation. The collection of embedded fixed designs defines the available concurrently randomised cohorts.
For transparent reporting of baseline balance, key pre-randomisation characteristics and sample sizes should be reported by treatment arm for each concurrently randomised cohort.
For transparent reporting of randomised comparisons, post-randomisation outcomes should be reported by treatment arm for each concurrently randomised cohort. At a minimum, this reporting should include the primary outcome but might also include other efficacy or safety outcomes. Depending on the types of outcomes, this reporting could include event numbers for binary outcomes, mean changes for continuous outcomes, or Kaplan-Meier estimates for time-to-event outcomes.
For transparent data sharing, the participant randomisation scheme should be considered as an integral component of the individual participant data of an adaptive trial. The participant randomisation scheme should be made available as part of the individual participant data in any data sharing arrangements.
Guidelines for adaptive trials, including reporting standards, have previously been published.11 13 These guidelines have not explicitly recommended concurrently randomised cohort reporting but have implicitly alluded to some of the problems that we have looked at more fully here. For example, the ACE checklist recommends that baseline characteristics should be reported by stage for each treatment group.11 Also recommended is that treatment effects should be reported from each stage of an adaptive trial.7 Although these stage specific recommendations are influenced by the same principles that we examined here, an important distinction exists between reporting by stage and reporting by concurrently randomised cohort. This distinction comes from the difference between contemporaneous and concurrent participants (box 1). We believe that we have examined these questions in detail and we propose that our recommendations should be added to existing guidelines. In particular, we hope that future updates to the reporting standards in the ACE checklist and other guidelines can explicitly include the added requirement for concurrently randomised cohort reporting of adaptive trials.
We have avoided recommending primary analysis methods for adaptive trials, which will often involve complex analytical techniques.14 15 Bayesian and frequentist statistical models are common and valid for primary analyses.16–18 Furthermore, the preferred approach to incorporating (or not incorporating) non-concurrent controls might differ from trial to trial.6 7 12 We do not provide any general recommendations on these topics here. Regardless of the methodology used in the primary analysis, however, presenting transparent summaries of the available randomised evidence in an easily interpretable format is important. Concurrently randomised cohort reporting provides these summaries and should be considered standard practice in the reporting of adaptive clinical trials.