Bayesian decision-theoretic group sequential clinical trial design based on a quadratic loss function: a frequentist...

Area of Science:

• Biostatistics
• Clinical Trial Design
• Decision Theory

Background:

• Traditional frequentist interim analysis plans (e.g., O'Brien-Fleming) do not inherently compare the value of additional information against enrollment costs.
• Decisions to terminate clinical trials at interim analyses often involve complex trade-offs not explicitly modeled in standard frequentist approaches.

Purpose of the Study:

• To develop a two-armed Bayesian decision-theoretic clinical trial design for diseases with binary outcomes.
• To quantify the cost of future enrollment and optimize early termination decisions.
• To compare the performance of the proposed Bayesian design against frequentist group-sequential designs.

Main Methods:

• A Bayesian decision-theoretic framework was developed using a quadratic decision loss function and backward induction.
• Monte Carlo simulations were employed to compare Bayesian designs with frequentist designs (O'Brien-Fleming, Pocock).
• The study evaluated frequentist error rates and mean required sample sizes.

Main Results:

• Bayesian designs, when optimized for frequentist error rates, required smaller mean sample sizes compared to O'Brien-Fleming designs.
• Frequent early stopping for equivalence and more frequent interim analyses contributed to the reduced sample size in Bayesian designs.
• Incorporating stochastic curtailment into frequentist designs with an equivalent number of interim analyses yielded comparable trial performance.

Conclusions:

• The developed Bayesian design offers a framework for optimizing clinical trial termination decisions by explicitly considering costs and benefits.
• Bayesian designs can achieve well-characterized frequentist error rates while potentially reducing mean sample size.
• The proposed methodology allows for flexible interpretation of results in both Bayesian and frequentist frameworks.