A Laplacian regularized graph neural network for predictive modeling of multiple chronic conditions

Area of Science:

• Computational medicine
• Artificial intelligence in healthcare
• Network science

Background:

• Multiple chronic conditions (MCC) pose significant challenges to healthcare systems, increasing mortality and disease progression.
• Understanding the complex interplay of pre-existing conditions and patient-specific risk factors is crucial for managing MCC.
• Existing models struggle to capture the intricate relationships inherent in the evolution of multiple chronic diseases.

Purpose of the Study:

• To develop and evaluate a novel Graph Neural Network (GNN) model for analyzing the relationships between chronic diseases, patient risk factors, and comorbidities.
• To investigate the impact of graph structure on the performance of GNNs in a healthcare context.
• To improve the accuracy and robustness of predictive models for patients with multiple chronic conditions.

Main Methods:

• A Graph Neural Network (GNN) model was developed to analyze relationships among five chronic conditions: diabetes, obesity, cognitive impairment, hyperlipidemia, and hypertension.
• Graph Laplacian regularization was incorporated into the GNN's loss function to enhance parameter learning and model accuracy.
• The model was validated using historical data from the Cameron County Hispanic Cohort (CCHC), involving 600 patients.

Main Results:

• The Laplacian regularized GNN model demonstrated superior performance compared to a baseline GNN, achieving an average accuracy of ≥89% across various combinations of chronic conditions.
• The proposed model showed improved robustness as the number of chronic conditions increased, unlike the standard GNN whose performance declined.
• Laplacian regularization facilitated consistent predictions for interconnected nodes, proving beneficial for patients with shared attributes.

Conclusions:

• Laplacian regularization is vital for GNNs applied to graph-structured health data, enhancing node categorization and predictive accuracy by leveraging graph topology.
• This study highlights the importance of incorporating graph structure into neural network design for complex biomedical data.
• The developed regularization method shows promise for future applications in diverse graph-based machine learning tasks within healthcare and beyond.