C-index regression for recurrent event data.

Area of Science:

• Biostatistics
• Survival Analysis
• Statistical Modeling

Background:

• Recurrent event data analysis is crucial in medicine, social science, and economics.
• Existing proportional rates or mean models perform poorly with model misspecification.

Purpose of the Study:

• To propose a novel, model-free approach for recurrent event data analysis.
• To introduce a lower bound on the concordance index (C-Index) for improved model performance.
• To develop variable selection procedures for high-dimensional data.

Main Methods:

• Developed a model-free estimation method using a continuous lower bound on the C-Index.
• Utilized the log-sigmoid function to derive the C-Index lower bound.
• Implemented a variable selection procedure for high-dimensional settings.

Main Results:

• The proposed methods outperform the gamma frailty recurrent event model when proportional mean assumptions are violated.
• Simulation results demonstrate superior performance in both low and high dimensional settings.
• Application to hospital readmission data yielded results consistent with previous studies, with a higher C-Index.

Conclusions:

• The novel model-free approach offers a robust alternative for recurrent event data analysis.
• The C-Index lower bound method provides better predictive accuracy, especially when models are misspecified.
• The approach is effective in both low and high dimensional scenarios and applicable to real-world datasets.