Quality of life analysis in community pharmacy using deep learning and explainability methods.

Area of Science:

• Clinical Informatics
• Machine Learning in Healthcare
• Patient Quality of Life Research

Background:

• Chronic diseases pose a significant burden on patients' quality of life.
• Community pharmacies offer accessible settings for clinical data collection.
• Identifying key factors influencing quality of life is crucial for personalized chronic disease management.

Purpose of the Study:

• To develop and evaluate a deep learning model for processing clinical data from community pharmacies.
• To identify key variables influencing health-related quality of life in patients with chronic diseases.
• To utilize global and local explainability methods for model interpretability.

Main Methods:

• Analysis of data from 347 chronic patients with 257 variables.
• Comparison of five predictive models: Gradient Boosting, Random Forest, LightGBM, and fully connected neural networks (FCNNs).
• Application of SHapley Additive exPlanations (SHAP) for global variable importance and Local Interpretable Model-Agnostic Explanations (LIME) for local interpretation.

Main Results:

• A fully connected neural network (FCNN) ensemble achieved the highest performance (R² = 0.511 ± 0.126).
• Explainability analysis identified pain, mobility limitations, beta-blocker use, anxiety/depression, and daily living difficulties as key influential variables.
• Deep learning models successfully captured complex relationships among clinical and psychosocial variables.

Conclusions:

• Deep learning models, combined with SHAP and LIME, provide clinically interpretable insights for personalized chronic disease care.
• Community pharmacies are practical settings for collecting data and applying predictive models.
• Machine learning supports personalized decision-making in chronic disease management by identifying critical factors affecting quality of life.