Machine learning-based prediction model for spontaneous preterm birth in singleton pregnancies using mid-trimester cervical elastographic parameters

The aim of this study was to propose a prediction model based on machine learning algorithms for spontaneous preterm delivery (sPTD) prediction in singleton pregnancies using the cervical elastographic markers obtained at mid-trimester. Multicentre prospective study was performed between June 2018 to December 2020 by the Korean Research Group of Cervical Elastography (grant ID: HI18C1696). Women with singleton pregnancies who underwent ultrasound between 18+0 and 24+0 weeks of gestation were eligible for analysis. Cervical elastography were performed using E-cervixTM (WS80A and HERA, Samsung Medison). sPTD was defined as delivery < 37 weeks of gestation due to either spontaneous preterm labour or rupture of membranes. The final data set included data from 1,518 patients and this data set was split into training and validation data sets. Three machine learning classification algorithms (Lasso regression, random forest and XGBoost) were used to build sPTD prediction models. Model performance in predicting sPTD from elastographic data and clinical factors was assessed using validation data set. Among 1,518 patients, 78 (5.1%) had spontaneous preterm delivery. The mean area under the receiver operating characteristic curve (AUC) for Lasso regression, random forest, and Xgboost model were 0.695, 0.704 and 0.723, respectively. The sPTD prediction score based on Xgboost was developed based on 9 selected variables: cervix length at examination, maternal age, maternal weight, internal os strain (IOS), external os strain (EOS), IOS to EOS ratio, elasticity contrast index, hardness ratio 30. The prediction model had an AUC of 0.737, significantly higher than CL alone (AUC = 0.594). The prediction model using mid-trimester cervical elastographic parameters had a potential value in predicting spontaneous preterm birth in single pregnancies.