Early termination in single-parameter model phase II clinical trial designs using decreasingly informative priors

Area of Science:

• Biostatistics
• Clinical Trial Design
• Statistical Modeling

Background:

• Subjective Bayesian prior selection can be replaced with assumptions tied to statistical decision-making in clinical studies.
• The decreasingly informative prior (DIP) is introduced as an alternative to standard Bayesian early termination methods.
• DIPs are designed to mitigate premature trial adaptations by incorporating skepticism proportional to the unobserved sample size.

Purpose of the Study:

• To expand standard Bayesian early termination methods for Phase II clinical trials using DIPs.
• To evaluate the efficiency and error control of DIPs in one-parameter statistical models.

Main Methods:

• Parameterization of DIPs using effective prior sample size for Bernoulli, Poisson, and Gaussian distributions.
• Simulation studies to determine optimal total sample sizes and termination thresholds for admissible designs (≥80% power, ≤5% type I error rate).

Main Results:

• The DIP approach requires fewer patients to achieve admissible designs for common distributions.
• When admissible designs are not met, DIPs offer comparable power and better type I error control than existing Bayesian priors.
• DIPs demonstrated similar or improved performance compared to Thall and Simon's Bayesian priors.

Conclusions:

• DIPs effectively control type I error rates, particularly in preventing erroneous early trial terminations.
• The DIP methodology allows for comparable or reduced patient numbers while maintaining desirable statistical properties.