Not too big, not too small: a goldilocks approach to sample size selection

Area of Science:

• Clinical Trials
• Biostatistics
• Bayesian Methods

Background:

• Optimizing sample size is crucial for clinical trial efficiency and ethical considerations.
• Traditional trial designs often require fixed sample sizes determined pre-trial, which can lead to underpowered or overpowered studies.
• Adaptive designs offer flexibility by allowing modifications based on accumulating data.

Purpose of the Study:

• To present a Bayesian adaptive design for confirmatory trials that dynamically selects the optimal sample size.
• To introduce the "Goldilocks trial design" concept for efficient sample size determination.
• To provide a framework for choosing design parameters and illustrate with examples.

Main Methods:

• A Bayesian adaptive sample size selection algorithm is developed.
• Frequent interim analyses use predictive probabilities to assess futility or sufficiency of the current sample size.
• The design ensures complete patient follow-up before the primary analysis.

Main Results:

• The proposed algorithm enables data-driven sample size adjustments during trial accrual.
• Predictive probabilities guide decisions on continuing accrual or stopping for futility.
• Demonstrated applicability for both dichotomous and time-to-event endpoints.

Conclusions:

• The Bayesian adaptive "Goldilocks trial design" offers a robust method for optimizing sample size in confirmatory trials.
• This approach enhances trial efficiency by avoiding unnecessary resource allocation.
• The methodology is adaptable to various endpoint types, supporting broader clinical trial applications.