Artificial intelligence–supported probabilistic modeling showed that maternal characteristics and glucose data from wearable sensors can help predict neonatal hypoglycemia in pregnancies affected by gestational diabetes mellitus (GDM). Published in the Journal of Diabetes Science and Technology, the study demonstrated how integrating clinical variables with CGM-derived features may improve perinatal risk assessment.
The analysis included 118 women with GDM enrolled in the STEADY SUGAR clinical trial. Bayesian network modeling was used to map associations among maternal health factors, CGM metrics collected during the 90 days before delivery, and neonatal outcomes.

Neonatal hypoglycemia showed direct associations with maternal hypertension (odds ratio [OR] 2.13, 95% CI 1.02–4.46), family history of diabetes (OR 1.43, 95% CI 0.57–3.57), and higher maternal BMI (OR 3.59, 95% CI 1.42–9.08). Cesarean delivery also increased NH risk (OR 2.05, 95% CI 0.98–4.28). The model identified indirect associations involving medication use and delivery type. Ethnic differences in glycemic control emerged, with Afro-American participants showing higher hyperglycemia (OR 2.91, 95% CI 1.19–7.11).

CGM-derived features demonstrated meaningful links to several neonatal outcomes, positioning wearable glucose data as a potentially valuable predictor of short-term neonatal risk.

These findings suggest that Bayesian network modeling can characterize complex maternal–neonatal interactions in GDM. Expanding this framework with larger datasets may support earlier, more personalized interventions to reduce neonatal complications.