Introduction:
Previous research by NALab has demonstrated that preterm birth (PTB) is a leading cause of mortality and morbidity for children under five. Discoveries in prevention measures have been based on maternal information such as previous PTB, socioeconomic background, quality of care visits, and environmental factors. Utilizing multi-omics and machine learning (ML) techniques, NALab aims to enhance precision medicine and healthcare by assessing PTB associations, predicting neonatal outcomes, and understanding the impact of sleep and physical activity (PA) during pregnancy. This study leverages smartwatch data to track sleep and PA, employing ML models to develop non-medication-based preventative measures.

Objective:
● To use smartwatch data to track and analyze sleep and physical activity during pregnancy.
● To create predictive algorithms using XGBoost and LASSO machine learning methods to assess the risk of preterm birth from proteomic pregnancy data.
● To evaluate the performance and significance of these models across different trimesters and their changes.

Outcomes:
● Developed and implemented algorithms using XGBoost and LASSO to predict preterm birth form proteomic data, with XGBoost showing better performance in the third trimester and delta (T3-T1) measurements.
● Identified that sleep and PA significantly impact PTB risk, with major improvements in prediction accuracy when analyzing the difference between trimesters.
● Found that dimension reduciton via LASSO did not significantly contribute to model accuracy, indicating the need for comprehensive data integration.
● Demonstrated the value of longitudinal data collection beyond a single trimester or visit, emphasizing the dynamic nature of sleep and PA throughout pregnancy.
● Concluded that future research should include more extensive data collection across all trimesters and explore additional data types to enhance predictive capabilities and develop more effective preventative measures.

Download Presentation File:

Download