Currently submitted to: JMIR Medical Informatics
Date Submitted: May 4, 2026
Open Peer Review Period: May 18, 2026 - Jul 13, 2026
(currently open for review)
Warning: This is an author submission that is not peer-reviewed or edited. Preprints - unless they show as "accepted" - should not be relied on to guide clinical practice or health-related behavior and should not be reported in news media as established information.
Nodes selection in Federated Learning for Hypoglycemia Prediction Hypoglycemia is an acute diabetic condition in which blood glucose drops below 70 milligrams per deciliter. Consequences of hypoglycemia include seizures, coma, and death. Hypoglycemia is often avoidable if emerging episodes are identified early enough, so accurate prediction can be highly beneficial to patients with diabetes. Prior work shows that timely prediction of hypoglycemic events is feasible using continuous glucose monitoring (CGM) time-series data and deep learning; however, deep learning typically requires substantial amounts of training data, motivating aggregation of CGM data across many patients, while CGM data itself is highly sensitive and may not be easil
ABSTRACT
Background:
Background:
Hypoglycemia, defined as blood glucose below 70 mg/dL, can lead to seizures, coma, and death. Timely prediction using continuous glucose monitoring (CGM) data is clinically important, but deep learning models often require data from many patients, while CGM data are sensitive and difficult to share.
Objective:
Objective:
This study aimed to develop and evaluate a privacy-preserving federated node-selection framework (Fed-node-selection) for 30-minute-ahead hypoglycemia prediction that improves predictive performance and reduces communication and computational overhead.
Methods:
Methods:
Each patient was treated as a federated node that trained a local neural-network predictor using the previous 2 hours of CGM data (24 samples) to predict hypoglycemia 30 minutes ahead, sharing only model parameters with a central server. During training, nodes ranked locally trained models on temporally partitioned validation data, and consistently top-ranked influential nodes were selected for downstream aggregation or ensemble inference. We evaluated two implementations: FedAvg-node-selection, which aggregates selected node updates, and FedEns-node-selection, which ensembles selected local models. Performance was assessed using temporal validation in an 89-patient type 1 diabetes cohort, a patient-disjoint holdout cohort of 22 unseen patients, and an external AZT1D cohort.
Results:
Results:
In temporal validation, FedEns-node-selection achieved balanced accuracy of 85.98% (SD 8.21%), outperforming FedEnsemble with all nodes (83.42%, SD 7.48%) and baseline FedAvg (82.57%, SD 7.84%). FedEns-node-selection also reduced missed hypoglycemia events, with a false negative rate of 14.26% (SD 11.43%) compared with 21.4% (SD 10.6%) for FedEnsemble. In zero-shot evaluation on 22 unseen patients, FedEns-node-selection achieved balanced accuracy of 90.77% (SD 12.23%) and reduced the false negative rate to 11.01% (SD 20.46%). On the external AZT1D cohort, it maintained balanced accuracy of 88.37% (SD 4.36%) without retraining.
Conclusions:
Conclusions:
Selectively using influential federated nodes improved the accuracy, efficiency, and portability of privacy-preserving hypoglycemia prediction. FedEns-node-selection was particularly effective for reducing missed hypoglycemia events while avoiding raw CGM data sharing.
Citation
Request queued. Please wait while the file is being generated. It may take some time.
Copyright
© The authors. All rights reserved. This is a privileged document currently under peer-review/community review (or an accepted/rejected manuscript). Authors have provided JMIR Publications with an exclusive license to publish this preprint on it's website for review and ahead-of-print citation purposes only. While the final peer-reviewed paper may be licensed under a cc-by license on publication, at this stage authors and publisher expressively prohibit redistribution of this draft paper other than for review purposes.