Identifying High-Risk Patients in Hospitals to Prevent Complications Using Graph Neural Networks (GNNs) and the COVID-19 Open Research Dataset (CORD-19)

The COVID-19 pandemic highlighted the urgent need for predictive models that can identify high-risk patients in hospitals to prevent severe complications. Traditional machine learning models rely on structured patient data, but recent advancements in Graph Neural Networks (GNNs) provide a novel approach to analyzing complex patient relationships. In this study, we propose a GNN-based framework to model patient interactions, symptoms, and underlying conditions using the COVID-19 Open Research Dataset (CORD-19). Our model constructs a heterogeneous patient-disease graph and applies Graph Convolutional Networks (GCN) and Graph Attention Networks (GAT) to predict the likelihood of severe complications. We perform extensive hyperparameter tuning to optimize performance, using evaluation metrics such as AUC-ROC, precision-recall, and SHAP for interpretability. Our results demonstrate that GNNs outperform traditional models in predicting high-risk patients, offering a scalable and effective solution for hospital decision-making.