[edit]
Towards a personalized pregnancy experience: Forecasting symptoms using graph neural networks and digital health technologies
Proceedings of the 4th Machine Learning for Health Symposium, PMLR 259:1104-1120, 2025.
Abstract
Pregnancy is an intricate process involving substantial physiological changes that impact both maternal and fetal health. In this study, we demonstrate the ability to predict two common symptoms during the third trimester of pregnancy—edema and fatigue—using physiological measures from wearable devices and self-reported daily surveys from mobile apps. Our approach employs a Graph Neural Network (GNN) framework, enhanced with a modified weighted Cross-Entropy loss to improve prediction performance. The model achieved AUC scores of 77.27% for edema and 69.70% for fatigue. Additionally, we aligned data from self-reported pregnancy symptoms with clinical examinations to carefully select participant cohorts for our experiments. We also explored how various features identified by the GNN are linked to these symptoms, gaining deeper insights into the relationship between physiological measures and pregnancy symptoms. Our findings indicate that heart rate variability plays a significant role in predicting symptoms of edema and fatigue, and features related to low-intensity activity also have a notable impact. Some of our findings closely align with previous studies on pregnancy. Our research serves as a proof of concept that symptoms can be predicted using wearable data, which may enhance the immediate well-being of expectant mothers and potentially personalize the overall pregnancy experience.