Score test of homogeneity for survival data

Area of Science:

• Biostatistics
• Survival Analysis
• Statistical Modeling

Background:

• Assessing homogeneity in grouped survival data (e.g., familial, spatial) is crucial for accurate analysis.
• Traditional methods may not adequately account for random group effects or adjust for covariates.

Purpose of the Study:

• To develop and validate a score test for homogeneity in grouped survival data within a frailty proportional hazards framework.
• To provide a computationally simple test that can also assess overdispersion and proportional hazards model fit.

Main Methods:

• Utilized a frailty proportional hazards model to incorporate random group effects and adjust for covariates.
• Derived a score test for homogeneity from the marginal partial likelihood.
• Employed counting process arguments to determine asymptotic variance.

Main Results:

• The derived score test is a sum of a pairwise correlation term of martingale residuals and an overdispersion term.
• The test is computationally simple and demonstrated effective use with simulated and real data.
• Proposed a decomposition of the score statistic for robustness and model fit testing.

Conclusions:

• The proposed score test offers a robust method for assessing homogeneity in grouped survival data.
• The test's components facilitate checks for proportional hazards model departures and overdispersion.
• This approach enhances the reliability of survival analyses involving clustered data.