A joint learning framework for analyzing data from national geriatric centralized networks: A new toolbox deciphering...

Area of Science:

• Geriatric Medicine
• Health Services Research
• Biostatistics

Background:

• Analyzing large-scale healthcare data presents challenges like patient heterogeneity, missing data, and confounding factors.
• National centralized networks, such as Medicare Claims, contain valuable information for understanding patient outcomes.
• Existing methods may struggle with the complexity of real-world healthcare datasets.

Purpose of the Study:

• To introduce JLNet, a joint learning framework with an R package for analyzing geriatric centralized network data.
• To address challenges including hospital clustering, patient variability, and loss to follow-up in healthcare data analysis.
• To support data-driven decision-making in geriatric care.

Main Methods:

• JLNet employs a three-step process: dynamic propensity score modeling for missing data, projection-based regularized regression for feature selection, and hospital clustering using residuals.
• The framework is designed for scalability and interpretability, handling high-dimensional covariates and hospital-level confounding.
• Applied to Medicare claims data (2010-2018) to study post-hip fracture recovery in older adults with Alzheimer's disease and related dementias (ADRD).

Main Results:

• JLNet identified key patient variables (e.g., age, weight loss) and distinct hospital clusters influencing post-discharge recovery (days at home) for ADRD patients.
• The framework demonstrated superior performance in variable selection and hospital clustering compared to existing methods.
• Numerical experiments confirmed JLNet's effectiveness in settings with high-dimensional data and unmeasured confounding.

Conclusions:

• JLNet provides a scalable and interpretable solution for analyzing complex centralized health data.
• The framework facilitates the identification of high-risk patient subgroups and hospital clusters, enabling personalized care.
• Findings support optimized resource allocation and the development of targeted interventions for older adults, particularly those with ADRD.