Discovering wearable sensor-derived temporal digital phenotypes and their association with hypertensive disorders in pregnancy

Hypertensive disorders of pregnancy (HDP) remain a leading cause of maternal morbidity, and episodic clinic measurements often miss early physiologic changes. We investigated whether longitudinal wearable data could identify interpretable physiological clusters (digital phenotypes), and trimester-to-trimester changes associated with subsequent HDP. We conducted a retrospective cohort study using pregnant participants in the NIH All of Us Research Program with wearable (Fitbit) and electronic health record data. Wearable measurements were obtained from 14-day windows preceding the end of each trimester and postpartum. We trained an attention-based recurrent autoencoder separately for physical activity, sleep, and pulse rate variability, and then clustered the latent embeddings using Gaussian mixture models to derive modality-specific digital phenotypes. We examined whether transitions in cluster membership between adjacent periods were associated with HDP by the next boundary. Activity clusters were externally validated in the Washington University Prematurity Research Cohort. In the discovery cohort (272 participants; 1,088 windows), we identified three activity, two sleep, and two PRV clusters with moderate-to-strong latent separation (silhouette 0.63–0.65; area under the curve 0.69–0.72). The clusters reflected gradients in activity and sedentary behavior, sleep quantity and architecture, and autonomic tone. Transitions toward lower activity, poorer sleep, and reduced autonomic variability were associated with increased odds of subsequent HDP. In external validation (131 pregnancies; 393 windows), activity clusters retained similar structure and discrimination (silhouette 0.54; area under the curve 0.68), and assignments were stable under resampling. These wearable sensor-derived digital phenotypes support the potential of machine learning-enabled continuous monitoring to complement routine prenatal care.

Verbatim fromthe publisher's DOI record· Publisher licence: https://creativecommons.org/licenses/by/4.0/