Multilevel joint frailty model for hierarchically clustered binary and survival data

Area of Science:

• Biostatistics
• Clinical Research Methodology
• Health Data Science

Background:

• Hierarchical data structures are common in medical research, often involving nested data (e.g., patients within hospitals).
• Existing multilevel models struggle with simultaneous analysis of mixed multivariate outcomes within these hierarchical structures.
• Multicenter studies frequently present complex data requiring advanced statistical approaches.

Purpose of the Study:

• To develop a novel multilevel joint frailty model for analyzing hierarchical data with both binary and survival outcomes.
• To simultaneously estimate regression parameters and model within-patient and within-hospital correlations.
• To provide a computationally efficient estimation method for complex hierarchical models.

Main Methods:

• Introduction of a multilevel joint frailty model accommodating binary and survival outcomes.
• Simultaneous analysis of outcomes to jointly estimate regression parameters.
• Application of a residual maximum likelihood (REML) method for efficient estimation and prediction of cluster-specific frailties.
• Modeling of within-patient correlation between outcomes and within-hospital correlation separately for each outcome.

Main Results:

• The proposed multilevel joint frailty model effectively handles hierarchical data with mixed multivariate outcomes.
• The residual maximum likelihood method provides a computationally efficient estimation procedure.
• Simulation studies demonstrate the robust performance of the model and estimation technique.
• The model's practical utility is confirmed in analyzing disease-free survival and platelet recovery in a bone marrow transplantation dataset.

Conclusions:

• The developed multilevel joint frailty model offers a powerful tool for analyzing complex hierarchical data in medical research.
• The efficient estimation method overcomes challenges associated with multidimensional integration in traditional likelihood-based approaches.
• This approach facilitates a more comprehensive understanding of disease progression and treatment outcomes in multicenter studies.