Icon Legend

This session is not in your schedule.

This session is in your schedule. Click again to remove it.

Presentation Icons

Additional registration fee required

Faculty have requested this content not be shared on social media

CME Credit Offered

Poster Icons

Foundation Awardee

No Social Media

Back

Q&A

Prematurity

Poster Session 3

(736) Identifying women at high risk for preterm birth based on clinical data using machine learning

Wednesday, February 14, 2024

8:30 AM – 10:00 AM

Location: Prince George's Exhibit Hall CDE, Lower Atrium Level

Primary & Presenting Author(s)

Gal Cohen, MD (she/her/hers)

OBGYN Resident
Meir Medical Centerartment of Obstetrics and Gynecology, Meir Medical Center
Kfar Saba, Israel, Israel

Coauthor(s)

TB

Tal Biron-Shental, MD

Meir, Tel-Aviv University
Sdeh Warburg, HaMerkaz, Israel
TM

Tal Meshali, N/A

Bar Ilan University
Ramat Gan, HaMerkaz, Israel
YL

Yoram Louzoun, PhD

Bar Ilan University
Ramat Gan, HaMerkaz, Israel

Objective: Existing models for predicting spontaneous preterm birth (delivery< 37 weeks, sPTB) are limited, relying mostly on a history of sPTB and multiple gestations, leaving the question of whether to screen the low-risk population for cervical dynamics during the 2^nd trimester a matter of debate. Another challenge arises upon admission of patients presenting with suspected sPTB, regarding the need for corticosteroids or magnesium treatment for reducing prematurity complications. We aimed to develop a bedside tool using machine learning, to assist clinicians identify patients who justify cervical length screening, as well as identify those in need of corticosteroids or magnesium upon admission for suspected sPTB.

Study Design: A data-based study of all live deliveries in a tertiary hospital between 2013-2023, of which 20% were used as a holdout test set, and 80% to train and validate a prediction model for sPTB using Xgboost. 1^st trimester parameters were used to predict sPTB at 12 weeks. Within the population admitted with suspected sPTB (presenting with PMC, pPROM or cervical length< 20 mm), data available upon admission were used to predict sPTB within 2 or 7 days.

Results: Included 63,937 deliveries, of whom 3,772 (5.9%) had a sPTB. Using 1^st trimester parameters, 2% of the population were identified with a probability of 50% for having a sPTB that represent 25% of the sPTB cases. Excluding multiple gestations allowed to recognize 20% of the population having a 10% risk for sPTB representing over 40% of the sPTB cases. Including only singleton nulliparas, still allowed for some stratification; women with the highest 20% scores had a 1.5 risk factor for sPTB.
Within 3,718 patients who were admitted with suspected sPTB, 17% were recognized as high-risk patients, having a probability of practically 100% to deliver within 7 days (recall 0.5) and 10% were recognized as having a probability of 100% to deliver within 2 days (recall 0.3).

Conclusion:

Implementation of machine learning can assist clinicians identify patients who are at higher risk for sPTB and justify cervical screening or medical interventions.