Estimation of predictive accuracy in survival analysis using R and S-PLUS

Area of Science:

• Biostatistics
• Statistical Computing

Background:

• Predictive accuracy evaluation is essential for survival regression models before practical application.
• Existing measures for survival data lack standardization and are often omitted from statistical software.
• There is a need for accessible tools to evaluate the predictive performance of survival models.

Purpose of the Study:

• To develop and introduce the surev library for R and S-PLUS.
• To provide functions for evaluating predictive accuracy measures for survival data, specifically those proposed by Schemper and Henderson.
• To facilitate the assessment of predictive accuracy in parametric and Cox regression models.

Main Methods:

• Development of the surev library in R and S-PLUS.
• Implementation of functions to calculate predictive accuracy measures for parametric regression models.
• Inclusion of capabilities to evaluate Cox models, including those with time-dependent covariates and Bayesian model averaging.

Main Results:

• The surev library provides a comprehensive tool for assessing predictive accuracy in survival analysis.
• The library supports evaluation for both parametric and Cox models.
• It accommodates complex scenarios such as non-proportional hazards and Bayesian model averaging in Cox models.

Conclusions:

• The surev library enhances the practical application of survival regression models by enabling robust predictive accuracy evaluation.
• It addresses the gap in standardized measures and software implementation for survival data.
• The library's utility is demonstrated through real-world data examples.