Graph Neural Networks for Heart Failure Prediction on an EHR-Based Patient Similarity Graph

• Heloisa Oss Boll UFRGS / Halmstad University
• Stefan Byttner Halmstad University
• Mariana Recamonde-Mendoza UFRGS

DOI: https://doi.org/10.5753/sbcas_estendido.2025.7013

Resumo

Accurate disease prediction is critical in healthcare. This study introduces Graph Neural Networks (GNNs) and Graph Transformer (GT) models to predict heart failure (HF) incidence at the next hospital visit using a patient similarity graph. Using MIMIC-III Electronic Health Records (EHR) data, we constructed a patient similarity graph via K-Nearest Neighbors (KNN) with embeddings from diagnoses, procedures, and medications. We implemented GraphSAGE, GAT, and GT to predict HF, evaluating performance with F1, AUROC, and AUPRC metrics against baseline models. A three-axed interpretability analysis explored model decision-making. GT achieved the highest performance (F1: 0.5361, AUROC: 0.7925, AUPRC: 0.5168). While Random Forest (RF) showed a similar AUPRC, GT provided superior interpretability by leveraging patient relational information. Analyzing attention weights, graph structure, and clinical features offered deeper insights into classification decisions.

Referências

Choi, E., Xu, Z., Li, Y., Dusenberry, M., Flores, G., Xue, E., and Dai, A. (2020). Learning the Graphical Structure of Electronic Health Records with Graph Convolutional Transformer. Proceedings of the AAAI Conference on Artificial Intelligence, 34(01):606–613.

Choi, Y., Chiu, C. Y.-I., and Sontag, D. (2016). Learning low-dimensional representations of medical concepts. AMIA Summits on Translational Science Proceedings, 2016:41–50.

Johnson, A., Pollard, T., and Mark, R. (2015). MIMIC-III Clinical Database.

Pieroni, A., Cabroni, A., Fallucchi, F., and Scarpato, N. (2021). Predictive modeling applied to structured clinical data extracted from electronic health records: An architectural hypothesis and a first experiment. Journal of Computer Science, 17(9):762–775.

Sharma, V., Davies, A., and Ainsworth, J. (2021). Clinical risk prediction models: The canary in the coalmine for artificial intelligence in healthcare? BMJ Health & Care Informatics, 28(1):e100421.

Tang, S., Tariq, A., Dunnmon, J. A., Sharma, U., Elugunti, P., Rubin, D. L., Patel, B. N., and Banerjee, I. (2023). Predicting 30-day all-cause hospital readmission using multi-modal spatiotemporal graph neural networks. IEEE Journal of Biomedical and Health Informatics, pages 1–12.

Tariq, A., Lancaster, L., Elugunti, P., Siebeneck, E., Noe, K., Borah, B., Moriarty, J., Banerjee, I., and Patel, B. (2023). Graph convolutional network-based fusion model to predict risk of hospital acquired infections. Journal of the American Medical Informatics Association, 30(6):1056–1067.