Bayesian Adaptive Design

A trial design using Bayesian statistics to formally update hypothesis probabilities as data accumulate, allowing pre-specified adaptive decisions such as early stopping.

Source: Berry 2006

A Bayesian adaptive design treats the treatment effect as a probability distribution that is updated as each new patient's data arrive, giving a continuously current estimate of what the evidence shows. This running update naturally supports adaptive decisions, since the design can, at pre-specified points, use the current probability that a treatment is working to decide whether to stop, continue, or shift allocation. The Bayesian framework is well matched to learning as the trial proceeds. It updates belief with every result. Berry (2006) describes this approach.

Source: Berry 2006

A Bayesian adaptive design works by specifying prior distributions and, as data accumulate, updating them to posterior distributions that quantify the current evidence about treatment effects, then using these posteriors, through pre-specified rules, to make adaptations such as stopping for efficacy or futility, reallocating participants toward better-performing arms, or dropping arms. The Bayesian updating provides a continuous, coherent measure of evidence to drive the decisions. Simulation is typically used in planning to check the design's operating characteristics, ensuring the adaptive rules control error rates while gaining efficiency from the flexibility.

Source: Berry 2006

Bayesian methods are used in adaptive designs because they naturally update the probability of hypotheses as data accumulate, providing a coherent framework for the continual learning that adaptation requires, and they yield intuitive posterior probabilities to guide decisions such as stopping or reallocating. Bayesian approaches can incorporate prior information and handle complex adaptations flexibly, and they support response-adaptive randomisation that directs more participants to better treatments. This fit between Bayesian updating and the sequential, learning nature of adaptive trials makes Bayesian methods well suited to designing and running adaptive designs.

Source: Berry 2006

A Bayesian adaptive design can guide pre-specified decisions based on the updated evidence, including stopping the trial early for demonstrated efficacy or for futility; adjusting the allocation of participants toward arms that appear more effective, through response-adaptive randomisation; dropping arms that are unlikely to succeed; and adjusting sample size. The posterior probabilities of treatment effects, computed as data accumulate, inform these decisions through the rules set in advance. This allows the trial to respond to emerging evidence in a principled way, improving efficiency and the ethical allocation of participants.

Source: Berry 2006

The challenges of Bayesian adaptive designs include the need to specify prior distributions, which can influence results and must be justified; the complexity of designing, simulating, and analysing the trial to ensure the adaptive rules control error rates; the computational demands of updating and simulation; and the requirement for rapid data handling to support interim decisions. Regulatory acceptance requires demonstrating good operating characteristics. These challenges mean Bayesian adaptive designs are planned rigorously, with careful prior choices, extensive simulation, and appropriate oversight, so their flexibility improves efficiency without compromising validity.

Source: Chow & Chang 2008