Evaluating survival model performance: a graphical approach

Area of Science:

• Biostatistics
• Survival Analysis
• Statistical Modeling

Background:

• Numerous statistics evaluate survival model performance, but often overlook temporal variations.
• Existing methods provide a single performance metric, failing to capture dynamic changes over time.
• Assessing time-varying covariate effects in survival models remains a challenge.

Purpose of the Study:

• To propose a graphical method for evaluating survival model performance that accounts for time-varying effects.
• To extend existing binary regression measures for application in survival analysis.
• To provide a tool for detecting time-varying covariate effects within the Cox proportional hazards model framework.

Main Methods:

• Developed a graphical method extending measures from binary regression for survival data.
• Applied the method to estimate model performance at specific time points.
• Utilized Cox proportional hazards models and rank statistics for illustration.

Main Results:

• The graphical method effectively depicts survival model performance across different time intervals.
• The approach allows for the identification of time-varying covariate effects.
• Demonstrated utility on both simulated and real-world datasets.

Conclusions:

• The proposed graphical method offers a dynamic assessment of survival model performance.
• This technique enhances the evaluation of Cox proportional hazards models by revealing time-dependent patterns.
• The method serves as a valuable tool for understanding covariate influence over the duration of survival.