Developing and enabling efficient hypothesis test for response-adaptive design with patient benefit goals

Before therapies are made available for general use in the population, they are typically evaluated in
clinical trials to determine that they are safe and effective. A main driver in the statistical design of
such trials is to ensure they can provide definitive answers for decision-making. Clinical trials are
usually expensive, and the full developmental process can take several years before a new successful
therapy is able to reach most patients. In many settings, such as life-threatening rare diseases, there
is a strong desire to allocate patients to a potentially superior intervention as soon as possible (i.e.,
during the trial itself). A useful approach to incorporate this additional goal into a clinical trial in such
settings is to use a response-adaptive design. These designs skew the allocation of patients in favour
of new interventions as long as they are showing promise during the trial. However, by possibly
assigning more patients to an intervention during the trial, the study could also result in a lower level
of evidence collected on all other interventions, which in turn could hinder the delivering definitive
answers to the efficacy question.
Response-adaptive designs are not new and have been proposed with the aim to deliver patient
benefit within a trial while preserving integrity of the final evidence. However, key statistical and
practical questions remain over the best approach to ensure that a trial using a response-adaptive
design has a high probability of definitively answering if an intervention is effective (without requiring
unrealistically large sample sizes to do so). Additionally, any new method that increases the chances
of finding a definitive answer after a response-adaptive design would still need to ensure statistical
integrity when no intervention is effective. The latter challenge is even greater if trials last for a long
time period and important variables in the patients' characteristics change over time (as is the case in
platform trials).
This project will develop novel statistical methods to maximise the probability to identify efficacious
interventions when using a response-adaptive design that offers patients in the trial a higher chance
of receiving the superior intervention. This will provide valid analysis methods for clinical trials that
offer the flexibility needed to enable patients within trials to expect a better outcome than in a
traditional design with fixed allocations per intervention. We will also ensure these methods preserve
validity even under changing temporal conditions. To ensure that the methods we develop are widely
disseminated and have maximum impact on clinical trial practice, we will provide open-source
software and recommendations for the use in practice of the produced methods. The
recommendations and guidance will take input from a workshop with key stakeholders including
statisticians with expertise in adaptive trial designs, clinicians, clinical trialists, relevant regulatory
bodies and patient representatives