Modeling and validating Bayesian accrual models on clinical data and simulations using adaptive priors.

Area of Science:

• Biostatistics
• Clinical Trial Design
• Health Services Research

Background:

• Slow patient recruitment significantly increases clinical trial costs and resource use.
• Effective planning and monitoring of accrual are crucial to prevent resource waste.

Purpose of the Study:

• To introduce two hierarchical extensions to the Bayesian constant accrual model: the accelerated prior and the hedging prior.
• To enable adaptive estimation of clinical trial completion times by integrating prior researcher experience and current accrual data.

Main Methods:

• Developed and evaluated two novel Bayesian priors (accelerated and hedging) for constant accrual models.
• Assessed model performance using actual clinical trial data from a cancer center and simulations, focusing on prediction precision, coverage probability, and decision accuracy.

Main Results:

• Strongly informative priors offer high accuracy and efficiency when accrual is on target but are biased when off target.
• Weakly informative priors protect against off-target accrual but are less efficient on target.
• The hedging prior balances performance, acting like weak priors when accrual is extremely off-target and strong priors when on-target.

Conclusions:

• The hedging prior demonstrates robust performance across various accrual scenarios, offering a valuable improvement over existing models.
• These adaptive Bayesian models enhance the prediction of clinical trial completion times, aiding in resource optimization.
• Further research into model enhancements and new Bayesian approaches for clinical trial recruitment is recommended.