Deep learning for the change-point Cox model with current status data.

Area of Science:

• Biostatistics
• Machine Learning
• Survival Analysis

Background:

• Traditional Cox proportional hazards models often use linear assumptions for covariate effects.
• Linear models may fail to capture complex relationships, impacting change-point detection accuracy.
• Current status data presents unique challenges for survival analysis.

Purpose of the Study:

• To develop novel estimation methods for a deep partially linear Cox proportional hazards model with a change point.
• To address limitations of linear models in capturing complex covariate effects for change-point analysis.
• To improve the accuracy of change-point detection in survival data.

Main Methods:

• Utilized a deep neural network to model covariate effects within the Cox proportional hazards framework.
• Proposed a maximum likelihood estimation procedure for the deep partially linear Cox model with a change point.
• Established asymptotic properties of the estimators, including consistency and semiparametric efficiency.

Main Results:

• The proposed deep learning-based Cox model effectively handles complex covariate relationships.
• Simulation studies demonstrated good performance of the inference procedure in finite samples.
• The methodology was successfully applied to a real-world breast cancer dataset.

Conclusions:

• The developed deep partially linear Cox model offers a powerful tool for survival data analysis with change points.
• This approach enhances the ability to accurately detect and model complex change-point effects.
• The findings have implications for survival analysis in biomedical research, particularly for time-to-event data with covariates.