Model evaluation based on the negative predictive value for interval-censored survival outcomes.

Area of Science:

• Biostatistics
• Survival Analysis
• Clinical Epidemiology

Background:

• Time-to-event data in cohort studies, such as disease recurrence, are often interval-censored.
• Predicting patient outcomes and identifying predictive covariates are crucial objectives in clinical research.
• Existing prediction rules may not adequately address survival endpoints with interval-censored data.

Purpose of the Study:

• To propose a novel model evaluation strategy for interval-censored survival outcomes.
• To leverage negative predictive functions for enhanced predictive accuracy.
• To provide a simple, flexible, and minimally implemented approach for biomedical data analysis.

Main Methods:

• Developed a model evaluation strategy utilizing negative predictive functions.
• Employed a nonparametric estimation of predictive accuracy.
• Required only a working model to obtain regression coefficients.

Main Results:

• The proposed method offers a flexible approach for evaluating predictive models with interval-censored survival data.
• The strategy makes minimal assumptions, enhancing its applicability.
• Simulation studies and a breast cancer trial demonstrated the practical advantages.

Conclusions:

• The developed method provides a valuable tool for assessing predictive accuracy in interval-censored survival data.
• Its minimal implementation effort facilitates immediate use in biomedical research.
• This approach improves the prediction of patient outcomes in clinical studies.