Z-residual diagnostic tool for assessing covariate functional form in shared frailty models

Area of Science:

• Biostatistics
• Survival Analysis
• Statistical Modeling

Background:

• Shared frailty models are crucial for analyzing clustered survival data by accounting for unobserved heterogeneity.
• Assessing the functional form of covariates in these models is challenging using traditional methods like Martingale and deviance residuals.
• Existing residual methods lack objective numerical tests and rely on subjective visual interpretation.

Purpose of the Study:

• To introduce Z-residuals, a novel diagnostic tool for shared frailty models.
• To provide both graphical and numerical tests for assessing covariate functional form.
• To address the limitations of existing residual diagnostics in shared frailty models.

Main Methods:

• Development of Z-residuals based on randomized survival probability.
• Implementation of Z-residuals computation within an R package.
• Conducting extensive simulation studies to evaluate the power of the numerical test.
• Application to a real-world dataset of acute myeloid leukemia (AML) survival times.

Main Results:

• Z-residuals offer a powerful numerical test for covariate functional form assessment.
• Simulation studies confirm the high power of the proposed numerical test.
• The Z-residual analysis identified the inadequacy of log-transformation for a specific covariate in the AML dataset.
• Demonstrated limitations of traditional residuals in effectively assessing covariate functional form.

Conclusions:

• Z-residuals provide a robust and objective method for diagnosing covariate functional form in shared frailty models.
• The new diagnostic tool enhances the reliability of survival data analysis.
• This approach offers significant advantages over existing methods, particularly in complex clustered survival data settings.