• Offer opportunity to make changes to the design of an ongoing trial as patient outcome data is accrued.|
• Can improve efficiency of trial (more power for same sample size, or reduced sample size for same power), make trial more robust to design assumptions, and/or improve patient benefit provided by trial.
• Benefit relies on the primary endpoint (or informative intermediate endpoint) being observed relatively quickly compared to recruitment.
|Adaptive signature design||
• Uses high-dimensional data to form a ‘sensitive’ subgroup of patients who experience higher benefit from an intervention in comparison to the overall population.|
• Allows forming, and confirmatory testing, of a predictive signature in the same trial.
• May be difficult to interpret the resulting signature.
|Augmented analysis of composite responder outcomes||
• Efficiently analyse responder endpoints, which classify patients as responders or non-responders on the basis of a combination of binary and continuous measurements.|
• Can substantially improve the power of trials using responder endpoints whilst maintaining the clinically relevant outcome.
• More complex analysis that makes extra assumptions compared to the traditional analysis approach.
|Basket and umbrella designs||
• Use an overarching protocol to test interventions in related disease conditions or patient subgroups, simultaneously.|
• Allow operational and statistical efficiencies; with the latter realised by using advanced statistical approaches that can e.g., share information between the different arms of the trial.
• Generally requires assuming the same endpoint and control group, despite various sub-studies, in the trial.
|Emulation of trials||
• A method for using large retrospective datasets to predict what it would have been if yielded by a randomised controlled trial.|
• Exploits the value of data that is already collected.
• Analysis makes strong assumptions and can only compare interventions in current use.
|Sequential Multi Assignment Randomised Trials (SMART)||
• Allow multiple randomisations of patients at different stages of the study.|
• Allow separate research questions to be answered and for the optimal ‘adaptive intervention’ to be found.
• For a specific AI, they allow to improve individual outcomes by further tailoring treatment by baseline or time-varying characteristics.