Early Prediction of Gestational Diabetes Using Deep Learning and Electrocardiography: Results from the MERGD Study
Year: 2026
Type: preprint
Abstract: Background: Current guidelines by the American College of Obstetrics and Gynecology require a two-step procedure for diagnosis of gestational diabetes (GD). This two-step procedure is challenging in resource limited settings. We determined if electrocardiography (ECG) done combined with artificial intelligence techniques can predict GD to provide a non-invasive, early gatekeeper strategy for GD risk stratification. Methods: Data for this study came from the Markers of Early Risk-stratification of Gestational Diabetes (MERGD) study and included 10s, 12-lead ECGs collected at the first antenatal visit. Patients were invited to undergo GCT at 24-28 weeks and, if required, an OGTT was conducted. GCT result was considered positive if the 1-hour, 50g glucose load blood glucose level was ≥130 mg/dL. ECGs were preprocessed, filtered, and a synthetic dataset was generated by using features from autoencoders, device-reported variables and heart rate variability parameters. A random forest model (RF-ECG) was trained on the synthetic data and validated first by 10-fold cross-validation, then on a held-out test of the synthetic data and finally on the original data not seen by the model. Results: Total 1,036 quality checked, preprocessed ECGs were used in the study. A total of 123 (11.87%) women developed GD. The fully trained RF-ECG model provided validation performance as follows: 78.28% accuracy, 87.80% sensitivity, 77.00% specificity and a likelihood ratio of positive test of 3.82. The RF-ECG model outperformed existing models of GD risk stratification. Importance metric identified 40 features as contributory to the model’s diagnostic logic. Conclusions: ECGs can accurately predict the likelihood of GD ahead of the final diagnosis. If supported by external validation in future studies, our approach can help noninvasively identify GD risk early during pregnancy in resource limited settings.
Source: Preprints.org
Language: English
Topic: ECG Monitoring and Analysis
Subfield: Cardiology and Cardiovascular Medicine
Field: Medicine
Domain: Health Sciences
Open Access status: green
OpenAlex ID: W7204254418