Deep representation learning identifies associations between physical activity and sleep patterns during pregnancy...

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Maternal and Child Health

Background:

• Preterm birth (PTB) is a leading global cause of infant mortality.
• Existing PTB predictive models often overlook cost-effective interventions suitable for low- and middle-income populations (LMICs).
• Objective measurement of physical activity and sleep is crucial but challenging; self-reported data lacks accuracy.

Purpose of the Study:

• To develop a deep learning model for tracking healthy pregnancy dynamics using wearable physical activity data.
• To create interpretability algorithms for understanding model predictions related to sleep, activity, and clinical variables.
• To assess the association between deviations from predicted pregnancy progression and preterm birth risk.

Main Methods:

• Utilized over 181,944 hours of physical activity data from 1083 patients via wearable devices.
• Developed a deep learning time-series classification architecture to model pregnancy progression using gestational age (GA) as a surrogate.
• Implemented novel interpretability algorithms including clustering, error analysis, and feature attribution.

Main Results:

• The deep learning model outperformed seven other machine learning methods in modeling pregnancy progression.
• Deviations from the predicted 'clock' of physical activity and sleep significantly correlated with PTB risk.
• Model underestimation of GA was linked to fewer PTBs, while overestimation indicated higher PTB risk (P < 1.01e-67 and P < 2.82e-39, respectively).
• Model error correlated negatively with interdaily stability, and sleep patterns were important predictors of GA estimation.

Conclusions:

• Deviations in physical activity and sleep dynamics during pregnancy, as detected by the model, are strong indicators of preterm birth risk.
• The model's interpretability allows for identifying when behavioral changes impact PTB likelihood.
• Advocates for clinical decision support systems using passive monitoring and personalized recommendations for exercise and sleep, particularly in LMICs.