Adaptive designs allow for data-driven changes to the future course of the trial following one or more interim looks at the data without compromising for the robustness and integrity of the trial or the interpretability of trial results. In general, and especially for confirmatory trials, the number and timing of interim looks and the proposed adaptations and statistical rules (guidelines) for making such adaptations have to be pre-specified in the trial protocol.

Sample size calculation for RCTs are commonly based on assumptions on treatment effect size and variability. These assumptions may come from prior trial data and/or literature published data. However, more often than not, there is great deal of uncertainty in these quantities which increases the risk of an under-powered or even sometimes an over-powered trial. Adaptive sample size re-estimation (SSR) designs including blinded sample size reassessment along with Group Sequential Designs (GSD) are ideal strategies for such risk mitigation. The main difference between SSR and GSD is that in the latter the sponsor commits to a larger sample size based on conservative assumptions in the hope of an early stop to save on sample size and/or time while for the former the sponsor has a smaller upfront commitment based on ambitious assumptions while being ready to invest more if the interim results are promising.

Risk of trial failure may also come from situations like heterogeneity in treatment effect, ill-informed dose selection at an earlier phase and delayed treatment effect for trials or inadequate follow-up times for trials with time-to-event endpoints. These risks can also be mitigated using appropriate adaptive designs.

Adaptive designs can also greatly enhance trial efficiency through seamless (both operationally and inferentially) approaches. Master protocols where multiple therapies are investigated for the same indication (Umbrella design) or where a single therapy is investigated for multiple diseases (Basket design) are also usually conducted in an adaptive fashion.

Adaptive designs can be based on both the traditional frequentist framework as well as the Bayesian framework. The Bayesian framework in general allows for more flexibility and robust interim decision making which includes incorporation of prior data in certain situations. It is not uncommon that a trial is designed in a hybrid fashion where the interim decision making is carried out in the Bayesian framework while the final analysis is carried out in the traditional framework.

Adaptive designs do demand more rigorous planning and communication with the regulatory bodies for confirmatory trial. This includes extensive simulations to be able to optimize the designs with respect to one or more of sample size, study duration and/or power (See Simulation guided designs). Most of these design would fall under the Complex Innovative Trial Design program described by the FDA (FDA guidance on CID initiative) . Guidance on adaptive designs can also be found in FDA Guidance on Adaptive Designs and in EMA’s Guidance for Adaptive Designs.

Our statistical experts with experience ranging from 12 to 20+ years with adaptive trial designs can help during the planning phase for optimizing your trial design and to identify and mitigate potential risks of trial failure. Their support will also extend to regulatory defence of protocols with adaptive designs and other communications with the regulatory bodies. During execution phase our team can support by carrying out the blinded/unblinded analysis and reporting to the Data Monitoring Committee (DMC) and helping them in taking reliable interim adaptive decisions (See DMC related service).

The following are some complex adaptive designs we have worked on in the past:

• Adaptive Group Sequential Sample Size Re-estimation designs, both in the traditional frequentist and Bayesian frameworks. This includes several therapeutic areas and endpoint types (Binary, count, continuous and time-to-event).
• Hybrid SSR design: For a large Phase-3 CVD trial we have also implemented a Bayesian interim analysis based on predictive power to determine increase in sample size and/or minimum follow-up time, while the final analysis is to be carried out in the traditional framework using a logrank statistic.
• Adaptive Biomarker threshold and enrichment designs in oncology.
• Adaptive inferentially seamless Phase-2/3 dose selection and confirmation design in oncology and rare diseases.
• Adaptive and operationally seamless augmented training and threshold optimisation followed by validation of diagnostic devices based on machine learned algorithms. Areas include Neurology, Obstetrics and Mental health. Also see Diagnostic Devices.
• Bayesian group sequential design for a large vaccine trial. Also see Bayesian Designs.
• Adaptive SSR designs with complex endpoints such as Negative Binomial count data and Win-Ratio statistic.
• Adaptive design with delayed endpoints while interim decision were based on intermediate endpoints. Therapeutic areas included Neurology and Oncology.
• Adaptive SSR design for confirmatory trial for validating measurement device with respect to a gold-standard.