Dynamic borrowing methods for basket trials with order restrictions

Area of Science:

• Biostatistics
• Clinical Trial Design
• Pharmacology

Background:

• Basket trials evaluate new therapies across multiple patient subgroups (baskets) under a single protocol.
• Subgroups can be defined by disease type, biomarkers, or patient characteristics.
• Prior knowledge of response order between baskets can inform trial design.

Purpose of the Study:

• To modify existing dynamic borrowing methods for single-arm basket trials with binary endpoints.
• To incorporate known response order information between baskets.
• To assess the impact of order-restricted methods on sample size and statistical performance.

Main Methods:

• Modification of two previously published dynamic borrowing methods.
• Incorporation of known response order assumptions between baskets.
• Evaluation of statistical properties including sample size, type 1 error rate, and power.

Main Results:

• Order-restricted methods require marginally smaller sample sizes (approx. 10% reduction for a five-basket trial) when order assumptions are met.
• When order assumptions hold and the study is adequately powered, these methods are efficient.
• Violations of order assumptions lead to inflated type 1 error rates and reduced power.

Conclusions:

• Incorporating known response orders into dynamic borrowing methods can improve sample size efficiency in basket trials.
• Careful consideration of order assumptions is crucial, as violations can negatively impact trial validity.
• These modified methods offer a potential advantage when prior biological or clinical information supports response order.