Table 1.

Brief summary of FDA guidance document adaptive design elements

Adaptive design element	Brief description	Advantages	Challenges
Group sequential designs	Designs that allow for one or more prospectively planned interim analyses of trial data with prespecified criteria for stopping the trial, generally based upon evidence of efficacy/effectiveness or futility	Trial can stop early, reducing overall sample size Results can be disseminated more quickly if trial stops early	May need a much larger maximum sample size than a trial without interim monitoring Early stopping may reduce safety data on a new intervention
Adapting the sample size	When uncertainty exists around the estimates used to power a study, an interim analysis can use accumulating data to re-estimate the sample size to ensure a trial has high power if the true magnitude was less than hypothesized but is still clinically meaningful	Reduces chance of a negative trial with a meaningful effect by increasing the sample size to have sufficient power Blinded approaches and re-estimation of nuisance parameters minimally impact type I error rates	Sample size increase may be so large as to be infeasible for continued enrollment Certain approaches may inflate the type I error rate without special consideration
Adaptive enrichment	A design which may adapt the patient population to a targeted subgroup (usually through demographic characteristics or by genetic/pathophysiologic markers believed to be related to the mechanism of action) or continue to enroll the participants from the originally specified trial population	Can refine eligibility criteria to enroll subgroups most likely to benefit from the intervention Approaches exist for both prognostic (identifying high-risk patients) and predictive (identifying more responsive patients) enrichment	Subgroups may be small (i.e., rare or hard to enroll) and challenging to determine benefit Choice of demographic characteristics or genetic markers may lead to different subgroups
Adaptations to treatment arm selection	Modification to the trial design that could add or terminate study arms, present in both early phase studies (e.g., dose-finding) and later phase studies (e.g., seamless designs and platform trials)	Extremely flexible to terminate study arms (for futility or efficacy) and add new arms Shared control arm, if used, may increase allocation to study interventions Can use a single master protocol versus multiple standalone trials	Multiple comparisons may lead to issues with type I error rates Criteria for terminating arms or adding new arms may be complex Overall resource use may be challenging to plan for given unknown number of arms and sample size
Adapting patient allocation	Also known as adaptive randomization (AR), the incorporation of methods to modify the randomization process that may be based on baseline covariates (i.e., to achieve “balance” in select covariates across study arms), response/outcome AR (i.e., attempting to randomize more participants to “effective” arms), or maintaining equal amounts of information when incorporating historic/supplemental data sources	Covariate-AR only uses baseline characteristics to promote balance across groups Response-AR increases probability of allocation to better-performing study arms AR when incorporating external control arm data can increase allocation to the intervention to maintain a desired allocation (e.g., 1:1)	Need to account for Covariate-AR in the statistical analysis plan Response-AR has special challenges with temporal trends, two-arm studies, & potential unblinding of study arms To incorporate external data advanced statistical approaches are needed and this supplemental data may introduce bias
Adapting endpoint selection	The ability to select one endpoint from a collection of potential primary endpoints when there is uncertainty about effect sizes across outcomes at an interim analysis, when done in FDA trials it involves extensive discussion and the review with the FDA Review Division	Can select a primary endpoint while the trial is ongoing based on uncertainty in the design stage Recent statistical research explores new designs to incorporate this adaptation	Not recommended by the EMA FDA recommends review and discussion with regulatory authorities May be seen as cherry picking an outcome that makes the intervention look good
Adapting multiple features	The above elements can be utilized individually or may be combined within a single adaptive trial design (at the expense of increasing complexity that needs to be carefully and thoroughly evaluated)	Provides the greatest flexibility by combining multiple adaptive features May result in designs with greater appeal to participants	Combining multiple adaptive features requires extensive statistical work and simulation studies May result in overly complex studies when a traditional design may have sufficed