The cross-validated adaptive signature design

Area of Science:

• Biostatistics
• Genomics
• Clinical Trial Design

Background:

• Many anticancer therapies benefit only a subset of patients, yet phase III trials often use broad eligibility criteria.
• Genomic data from microarrays can identify patient subgroups likely to respond to specific therapies.
• Developing reliable predictive classifiers for targeted therapies is challenging due to high-dimensional genomic data.

Purpose of the Study:

• To propose a cross-validation extension of the adaptive signature design.
• To optimize the efficiency of classifier development and validation within a single clinical trial.
• To improve the identification of patient subpopulations who benefit most from anticancer treatments.

Main Methods:

• Introduced the adaptive signature design combining classifier development and treatment effect testing.
• Proposed a cross-validation extension to enhance the adaptive signature design.
• Evaluated the new design using simulations and application to breast cancer trial data.

Main Results:

• The cross-validation approach significantly improves the performance of the adaptive signature design.
• Simulations demonstrated the effectiveness of the proposed method.
• The design was successfully applied to real-world breast cancer trial data.

Conclusions:

• The cross-validation extension offers a more efficient and reliable method for adaptive signature design.
• This approach enhances the ability to identify sensitive subpopulations for targeted anticancer therapies.
• Methods for estimating treatment effects in identified subpopulations were described.