A Weighted Survival Regression Framework for Incorporating External Prediction Information

Area of Science:

• Biostatistics
• Survival Analysis
• Machine Learning in Healthcare

Background:

• Accurate estimation of time-to-event data is crucial in medical research.
• External prediction models offer valuable supplementary information.
• Existing methods may not fully leverage external predictions for right-censored data.

Purpose of the Study:

• To develop a novel weighted estimation approach for right-censored time-to-event data.
• To integrate predictions from external models into survival data analysis.
• To address the challenges in statistical inference associated with this new methodology.

Main Methods:

• A weighted estimation technique is proposed for time-to-event data.
• The method accommodates arbitrary external prediction models.
• Subject-specific weights are utilized, compatible with standard statistical software.
• New theoretical results and a perturbation-based inference method are developed.

Main Results:

• The weighted approach allows flexible incorporation of external predictions.
• The methodology is computationally feasible using existing software.
• The proposed inference method provides reliable results for complex scenarios.
• The approach was successfully applied to three diverse public datasets.

Conclusions:

• The developed weighted method offers a powerful tool for survival data analysis.
• It effectively leverages external predictions, enhancing estimation accuracy.
• The methodology facilitates robust statistical inference in the presence of censoring and external model information.