Non-parametric frailty Cox models for hierarchical time-to-event data

Area of Science:

• Biostatistics
• Health Services Research
• Survival Analysis

Background:

• Hierarchical time-to-event data, common in healthcare, often uses Cox models with parametric frailties.
• Existing models struggle with general frailty distributions and identifying provider-specific risk heterogeneity.
• Limited covariates in administrative data pose challenges for traditional survival models.

Purpose of the Study:

• To introduce a novel, flexible statistical model for hierarchical time-to-event data.
• To relax the parametric frailty assumption using a non-parametric discrete distribution.
• To improve the analysis of healthcare administrative data with limited covariates.

Main Methods:

• Developed a novel hierarchical survival model with non-parametric discrete frailties.
• Proposed a tailored Expectation-Maximization algorithm for parameter estimation.
• Compared model selection methods and assessed performance via simulation studies.

Main Results:

• The proposed model offers greater flexibility for diverse frailty distributions.
• It enables data clustering into groups of healthcare providers with similar unobserved risk factors.
• The model effectively handles healthcare administrative data with few covariates.

Conclusions:

• The novel non-parametric hierarchical model enhances survival analysis for healthcare data.
• It provides a flexible alternative to traditional Cox models with parametric frailties.
• The approach facilitates exploration of latent structures among healthcare providers and patient risk.