Preterm birth prediction may be improved using wearable sleep trackers

Preterm birth prediction may be improved using wearable sleep trackers | Image Credit: © Africa Studio – © Africa Studio – stock.adobe.com.

Preterm birth (PTB) prediction may be improved through a wearable device assessing variability in sleep patterns, according to a recent study in npj Women’s Health.¹

Significance of PTB and the need for early prediction

The World Health Organization has stated nearly 75% of pediatric deaths because of PTB may be prevented through timely interventions, but the symptoms indicating PTB remain unknown. This data highlights a model that may be used to improve PTB prediction based on sleep patterns.

“Raw data from wearables can be very messy, but using a healthy combination of statistical methods, [artificial intelligence], and clinical knowledge, researchers can extract important clinical insights,” said Chenyang Lu, PhD, MS, BS, director of the AI for Health Institute at Washington University in St. Louis.

The study was conducted to evaluate the efficacy of binary-outcome PTB classification using engineered actigraphy features and patient history features.² Participants included women aged 18 years or older with a singleton pregnancy at under 20-weeks’ gestation planning to deliver at Barnes-Jewish Hospital.

Data collection across pregnancy

Baseline maternal demographics, medical history, antepartum data, and obstetric outcomes were obtained through case report forms by trained obstetric research staff. This data was collected from patients during scheduled study visits at each trimester and delivery.

Alongside standardized survey responses, investigators collected biological samples, imaging, and actigraphy from patients. Questions on the survey covered schedule, stress, physical activity, sleep quality, diet, postnatal depression, demographics, and overall lifestyle.

Definition of PTB and actigraphy collection timeline

PTB was defined as births occurring 3 full weeks before the estimated date of confinement, which was determined based on a patient’s last menstrual period or first ultrasound. Two-week periods were designated in each trimester for collecting actigraphy measurements.

In the model, day-level actigraphy features were aggregated to their mean and standard deviation throughout all of pregnancy. Gestational ages (GAs) below a set range were dropped from the model. Data about both domain knowledge and automatic techniques were selected to be included in the model.

PTB incidence and model performance

There were 665 patients with first- or second-trimester actigraphy data included in the analysis, with a mean 39.1 day-level samples throughout pregnancy. These patients were aged a mean 29.2 years and 55.34% were multiparous. A PTB outcome was reported in 14.18%.

Investigators noted that the performance of the models did not consistently change when increasing the GA upper-bound, though a noticeable increase was observed when using the full GA spectrum. Weak links were reported for sample trends with the area under the receiver-operator curve and area under the precision-recall curve.

When using all features in the model, investigators found that features linked to the number of complications in previous births had the greatest impacts on the model’s output. Race, ethnicity, and employment status were also consistently linked to PTB.

Actigraphy findings

Of actigraphy features, day-to-day variability between sleep start had the greatest impact on PTB. Similar effects were reported for sleep start time, the variance of sleep cycle start, and day-to-day variations in the duration of sleep cycles.

Increased ranks were reported for actigraphy features relating to sleep pattern variations vs those obtained from averages throughout pregnancy. A similar order of relevant features was found when evaluating the best-preforming actigraphy-only model.

“There is no intervention because we can’t predict who’s going to have a preterm birth,” said Sarah England, PhD, director of the Center for Reproductive Health Sciences.¹ “We’re hoping that this will be much more helpful in getting predictive power of women who are going to be at higher risk.”

References:

1. Sleep data from wearable device may help predict preterm birth. Washington University in St. Louis. June 25, 2025. Accessed June 30, 2025. https://www.eurekalert.org/news-releases/1088926.
2. Warner BC, Zhao P, Herzog ED, Frolova AI, England SK, Lu C. Validation of sleep-based actigraphy machine learning models for prediction of preterm birth. npj Women’s Health. 2025;3(40). doi:10.1038/s44294-025-00082-y