Multilevel joint model of longitudinal continuous and binary outcomes for hierarchically structured data

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Clinical Research Methodology

Background:

• Joint models commonly link longitudinal and survival data but infrequently address hierarchical structures or binary outcomes.
• Multilevel modeling is crucial for multicenter studies but less explored within joint modeling frameworks.
• Existing joint models often neglect nested effects, limiting their application in complex clinical data like that from cystic fibrosis registries.

Purpose of the Study:

• To propose a novel multilevel joint model integrating linear mixed-effects (LME) and generalized linear mixed-effects (GLMM) submodels using a Bayesian approach.
• To apply this model to analyze lung function and pulmonary exacerbations in cystic fibrosis (CF) patients across multiple centers.
• To evaluate the bias introduced by non-hierarchical models and the benefits of incorporating center-specific effects.

Main Methods:

• Developed a Bayesian multilevel joint model combining LME for continuous longitudinal outcomes and GLMM for binary outcomes.
• Incorporated nested effects (center effects) within the LME submodel to account for hierarchical data structures.
• Utilized a symmetric power exponential power (spep) link function in the GLMM to model center-specific parameters.

Main Results:

• The multilevel joint model significantly improved estimation and prediction accuracy by including center effects and individual variation in the LME submodel.
• Simulation studies confirmed the bias reduction and improved performance of the hierarchical model compared to non-hierarchical approaches.
• Application to US CF Foundation Patient Registry data demonstrated the importance of accounting for center-specific variations in lung function.

Conclusions:

• Multilevel joint models are essential for accurately analyzing hierarchical longitudinal data, particularly in multicenter clinical studies.
• Incorporating center-specific effects and appropriate link functions (like spep) enhances the reliability of findings in populations like cystic fibrosis.
• This approach offers improved clinical insights for managing diseases with complex, multi-level data structures.