Machine Learning Algorithm Using Clinical Data and Demographic Data for Preterm Birth Prediction

We sought to develop and validate a machine learning algorithm for preterm birth (PTB) using clinical, demographic, laboratory data. performed cohort study of pregnant women delivering at single institution prospectively collected information on clinical conditions, patient demographics, data, health care utilization from 2019 - 2021. Thirty-three elements previously identified in the literature increase risk PTB were included supervised model. Primary outcome prediction was < 37 weeks. The dataset randomly divided into derivation (70%) validation (30%). In both logistic regression (XG-Boost) models used derive best fit (C-Statistic) then validated cohort. measured model discrimination with C-Statistic assessed performance calibration determine whether provided decision-making benefits. To address issues related interpretability predictive models, we decision tree based feature importance quantification algorithm, provide further visibility influence various features Cohort 12,440 deliveries among 12,071 women. Derivation 8,708 3,732 occurred 2,037 births (16.4%). Using XG-Boost, following (by importance) are multiple gestation, number ED visits prior pregnancy, initial unknown BMI, gravidity, PTB, gestational hypertension, inpatient admissions during first 20 weeks’ being African-American, BMI >30.0 kg/m2 (Figure 1). Test characteristics similar between groups (derivation AUC =0.7 vs. AUC= 0.63). Our integrates pre-pregnancy pregnancy history, acute demographic factors moderate precision.