Federated learning predicts which ER patients need coronary revascularization without sharing hospital data

Researchers in South Korea have built a federated learning model that predicts which emergency room patients will receive coronary revascularization within a week of their visit, without any hospital having to hand over its patient records. The retrospective study, published in BMC Medical Informatics and Decision Making, drew on de-identified data from three hospitals: Korea University Anam Hospital, Ansan Hospital, and Guro Hospital.

The study included patients who underwent electrocardiography plus cardiac biomarker testing with troponin and CK-MB, and excluded those who were discharged, deceased, transferred, or had trauma. In total, the dataset comprised 26,906 cases from Anam, 27,194 from Ansan, and 21,397 from Guro, with revascularization rates ranging from 2.6% to 3.4% across the sites. Data from January 2017 to July 2020 formed the training set, while cases from August 2020 to July 2021 were held out for testing.

In this study, the team added Gaussian noise with a standard deviation of 0.01 to each site’s parameters before server aggregation as an additional privacy measure. The researchers compared three approaches: models trained locally at each hospital, a centralized model trained on pooled data, and the federated model.

Both the federated and centralized models outperformed the local models, each achieving an area under the receiver operating characteristic curve of 0.95 and an area under the precision-recall curve of 0.64. Performance was also assessed using precision, recall, and accuracy. The federated model maintained robust performance even with the added noise, suggesting that the privacy protection did not come at the cost of predictive accuracy in this setting.

To gauge clinical usefulness, the team measured agreement between the model’s predictions and physicians’ decisions to perform coronary angiography, using Cohen’s kappa. The federated model showed moderate-to-substantial agreement with those decisions. According to the authors, the approach may enable scalable decision support that reduces unnecessary coronary angiography while still identifying patients likely to require intervention, though the model predicts the receipt of revascularization rather than directly measuring patient outcomes, and the findings come from a retrospective analysis of three South Korean hospitals.

Subject of Research: Federated learning for predicting coronary revascularization in emergency room patients

Article Title: Federated learning approach to optimize coronary revascularization decisions with privacy-preserving data integration from multiple hospitals

Article References: Kim, S. M., Oh, S., Kim, J. H., Hong, S. J., Yu, C. W., Kim, Y. H., Kim, E. J., & Joo, H. J. (2026). Federated learning approach to optimize coronary revascularization decisions with privacy-preserving data integration from multiple hospitals. BMC Medical Informatics and Decision Making. https://doi.org/10.1186/s12911-026-03940-0

Image Credits: AI Generated

DOI: 10.1186/s12911-026-03940-0

Keywords: federated learning, coronary revascularization, emergency room, acute coronary syndrome, machine learning, privacy-preserving, coronary angiography, electrocardiography, troponin, multicenter study, decision support, Federated