A Variance Estimator for Marginal Cox Regression Models Fit to Non-Nested Multilevel Data

Area of Science:

• Health Services Research
• Biostatistics
• Epidemiology

Background:

• Health services research frequently utilizes clustered data to analyze individual outcomes and cluster-level factors.
• Generalized estimating equation (GEE) and hierarchical regression models are standard for single or nested clustering.
• Existing methods for marginal regression lack development for multiple, non-nested clustering structures.

Purpose of the Study:

• To propose a novel variance estimator for marginal Cox regression models applied to non-nested multilevel data.
• To address limitations in analyzing complex health services data with multiple, independent clustering sources.
• To enhance statistical modeling capabilities for non-nested hierarchical data structures.

Main Methods:

• Developed a variance estimator combining Miglioretti and Heagerty's GEE-type approach with Lin and Wei's robust Cox model variance estimator.
• Employed extensive Monte Carlo simulations to evaluate the performance of the proposed variance estimator.
• Applied the estimator in a case study of acute myocardial infarction patients clustered by hospital and neighborhood.

Main Results:

• The proposed variance estimator demonstrated effective performance in Monte Carlo simulations for non-nested multilevel data.
• The method successfully addressed the challenges of analyzing data with multiple, non-nested clustering.
• The case study illustrated the practical application and utility of the new variance estimator.

Conclusions:

• A variance estimator, inspired by Miglioretti and Heagerty, is suitable for marginal Cox regression models with non-nested multilevel data.
• The proposed method offers a valuable tool for health services researchers dealing with complex data structures.
• This advancement improves the statistical rigor for analyzing outcomes in non-nested hierarchical settings.