(1170) Machine Learning Algorithm for the Point of Care Prediction of Preterm Labor

Friday, January 31, 2025

4:00 PM – 6:00 PM

Submitting Author and Presenting Author(s)

Reetam Ganguli, BS

Elythea
San Jose, CA, United States

Coauthor(s)

SW

Stephen Wagner, MD

Beth Israel Deaconess Medical Center
Boston, Massachusetts, United States

Objective: Preterm birth is the leading cause of neonatal mortality and is 25-fold more expensive than term deliveries to payors. Relevant literature has demonstrated that clinical intervention indicated by early risk screening tests can reduce preterm birth occurrence by 20% and can save over $800 per patient. Current screening methods to detect patients at risk of preterm labor are limited with some studies suggesting established protocols such as cervical length screening have a sensitivity as low as 11%. Our goal was to develop a machine learning model to non-invasively predict risk of preterm birth at the point of care.

Study Design:
We conducted a cohort study utilizing sociodemographic and clinical data obtainable in the 1st trimester from the CDC Vital Statistics System. Deliveries with missing preterm labor data and deliveries in hospitals not reporting preterm labor were excluded. The primary outcome was spontaneous preterm birth < 37 weeks.
Models were trained on years 2018-2020 of the CDC Vital Statistics System and were tested on a hold-out set of year 2021.

Results:
40 clinical variables for 12,476,992 obstetric patients were included in the training data. 2,685,095 (21.5%) patients had a preterm birth.
The model was tested on 3,666,801 patients, of which 450,318 (12.3%) patients went into preterm labor. The developed model had an area under the receiver operating characteristic curve of 0.73, sensitivity of 0.65, and F1 score of 0.68.
The highest weighted factors in the model were: number of prenatal visits, trimester that prenatal care was initiated in, and interval since last live birth.

Conclusion:
Our developed ML model leveraging data available at early stages of pregnancy harbors the potential to identify women at increased risk for spontaneous preterm birth at nearly 6 times the sensitivity than existing cervical length measurement tests as soon as the point of care and may allow for targeted interventions to reduce spontaneous preterm labor incidence/lower healthcare expenditures related to preterm birth.