Survival analysis with change-points in covariate effects

Area of Science:

• Biostatistics
• Statistical modeling
• Survival analysis

Background:

• Cox regression models are widely used for survival data analysis.
• Identifying changes in covariate effects over time is crucial for accurate modeling.
• Existing methods may not effectively detect multiple change-points.

Purpose of the Study:

• To develop a novel statistical method for detecting multiple change-points in covariate effects within Cox regression.
• To provide a robust approach for inferring the number and location of these change-points.
• To construct confidence intervals for model parameters using a bootstrap method.

Main Methods:

• Application of a maximal likelihood ratio test for multiple change-point detection.
• Sequential inference for determining the number of change-points.
• Bootstrap method utilizing Bernstein polynomials for confidence interval construction.
• Assessment via simulation studies and application to real-world datasets.

Main Results:

• The proposed maximal likelihood ratio test effectively identifies multiple change-points in covariate effects.
• The sequential approach accurately infers the number of change-points.
• Bootstrap-based confidence intervals provide reliable estimates for regression and change-point parameters.
• Simulations confirm the method's performance.

Conclusions:

• The developed method offers a powerful tool for analyzing time-dependent covariate effects in Cox regression.
• This approach enhances the accuracy and interpretability of survival models with multiple structural breaks.
• The technique is validated and applicable to diverse biomedical datasets.