Graph convolutional networks (GCNs) had been advanced to study representations of graph-based records. This paper presents a novel approach to predicting cardiac arrest the use of GCNs on observational information from a nationwide clinical registry. The model combines demographic records (age, intercourse, comorbidities, and many others.), previous admissions to hospitals, and historic essential signs and symptoms. The resulting model’s performance became as compared towards conventional system-learning techniques the usage of receiver running characteristics and go-validation. Outcomes reveal that the GCN correctly predicts cardiac arrest occasions with a place below & the curve of zero. Seventy seven. Moreover, the GCN outperforms traditional device-mastering methods inclusive of logistic regression, random woodland, and gradient boosting. This has potential implications for actionable interventions for danger-discount and aid-allocation planning for healthcare systems.