A SEQUENTIAL SIGNIFICANCE TEST FOR TREATMENT BY COVARIATE INTERACTIONS

Area of Science:

• Biomedical research
• Clinical research
• Statistics

Background:

• Biomedical research is shifting towards personalized medicine.
• Identifying treatment-covariate interactions is crucial for personalized medicine.
• Existing machine learning methods lack formal hypothesis testing for treatment selection.

Purpose of the Study:

• To develop a novel hypothesis testing procedure for identifying treatment-covariate interactions.
• To address the limitations of existing machine learning methods in formal hypothesis testing.
• To enable sequential identification of variables interacting with treatments.

Main Methods:

• A novel testing procedure based on the m-out-of-n bootstrap.
• Theoretical analysis of the method's properties.
• Simulation studies to compare performance against competing methods.

Main Results:

• The proposed m-out-of-n bootstrap method demonstrates superior control of type-I error rates.
• The method achieves satisfactory statistical power in identifying interactions.
• The procedure effectively identifies variables that interact with treatments.

Conclusions:

• The novel m-out-of-n bootstrap testing procedure is a valuable tool for personalized medicine.
• The method offers improved performance over existing approaches for treatment-covariate interaction identification.
• The procedure's utility is confirmed through real-world data from clinical trials and observational studies.