Related Experiment Video
Updated: May 10, 2026
A Novel Use of Three-dimensional High-frequency Ultrasonography for Early Pregnancy Characterization in the Mouse
Published on: October 24, 2017
Machine learning approaches for predicting fetal macrosomia at different stages of pregnancy: a retrospective study
Qingyuan Liu1,2, Simin Zhu1, Meng Zhao1
1Department of Obstetrics and Gynecology, Beijing Tsinghua Changgung Hospital, School of Clinical Medicine, Tsinghua University, Beijing, China.
Background:
Macrosomia presents significant risks to both maternal and neonatal health, however, accurate antenatal prediction remains a major challenge. This study aimed to develop machine learning approaches to enhance the prediction of fetal macrosomia at different stages of pregnancy.
Methods:
This retrospective study involved 500 pregnant women who delivered singleton infants at Beijing Tsinghua Changgung Hospital between December 2019 and July 2024. The training set comprised 208 cases of macrosomia and 208 non-macrosomia cases, with 84 additional cases used for external validation. A total of 23 candidate variables, including maternal characteristics, physical measurements, and laboratory tests were collected for feature selection. Seven algorithms were applied in combination with three sets of selected features, resulting in 21 fitted models. Model performance was evaluated via the area under the receiver operating characteristic curve (AUC), accuracy, precision, sensitivity, specificity, and F1-score.
Results:
Maternal height, pre-pregnancy weight, first-trimester weight, pre-labor weight, gestational age at birth, gestational weight gain, and the proportion of male neonates were significantly greater in the macrosomia group compared to non-macrosomia group in the training set (p < 0.05). The top five predictors for macrosomia were pre-labor weight, gestational weight gain, the Pre-labor Hb/First-trimester Hb ratio, first-trimester Hb, and maternal height. Logistic regression yielded the highest AUC values in the pre-pregnancy (0.790) and first-trimester (0.815) periods in the validation set, whereas the ensemble model achieved the highest AUC value of 0.930 before labor. SHapley Additive exPlanations (SHAP) analysis highlighted pre-labor weight, gestational age, gestational weight gain, first-trimester Hb, and neonatal sex as important factors for the prediction of macrosomia.
Conclusion:
This is the first study to utilize machine learning with data from the pre-pregnancy, first-trimester, and pre-labor periods to predict macrosomia. The logistic regression model and the final ensemble model demonstrated strong predictive performance, offering valuable insights to improve pre-pregnancy counseling, antenatal assessment, and intrapartum decision-making.
