Construction and evaluation of a risk prediction model for low Apgar score in very preterm infants based on different machine learning algorithms
10.3760/cma.j.cn101451-20251124-00462
- VernacularTitle:基于不同机器学习算法的极早产儿低Apgar评分风险预测模型构建和评估
- Author:
Huiying FAN
1
;
Anran LIANG
1
;
Hao SONG
1
;
Jingyue XING
1
;
Yaxuan LIU
1
;
Jingjing ZHAO
1
;
Wenli LI
1
;
Falin XU
1
Author Information
1. Department of Neonatology, Third Affiliated Hospital of Zhengzhou University, Zhengzhou 450052, China
- Publication Type:Journal Article
- Keywords:
Apgar score;
Extremely preterm infants;
Machine learning;
Predictive model;
Prenatal risk assessment
- From:
Chinese Journal of Neonatology
2026;41(11):665-671
- CountryChina
- Language:Chinese
-
Abstract:
Objective To construct and evaluate the risk prediction models for low Apgar score in very preterm infants (VPIs) using different machine learning algorithms. Methods From June 2023 to December 2024, a retrospective case‑control study was conducted among VPIs born in the Department of Obstetrics of our hospital. The patients were randomly assigned into a training set and a validation set with a ratio of 3:1. Least absolute shrinkage and selection operator (LASSO) regression analysis was used to select the risk factors of low Apgar score in the training set. 1‑min Apgar score ≤7 was defined as the outcome indicator. Four machine learning algorithms: Logistic regression (LR), extreme gradient boosting (XGBoost), random forest (RF) and support vector machine (SVM) were used to construct the prediction model. Model performance was evaluated in the validation set using area under the receiver operating characteristic (ROC) curve (AUC), accuracy, precision, sensitivity, specificity and F1 score. The optimal model was further studied using Shapley additive explanations (SHAP) method. Results A total of 544 cases were enrolled, including 408 in the training set and 136 in the validation set. There were 126 cases with low Apgar scores and 282 cases with normal Apgar scores in the training set. LASSO regression identified 9 predictive factors: gestational age, maternal anemia, antenatal fever, cervical insufficiency, abnormal umbilical artery blood flow, fetal malpresentation, fetal distress, premature rupture of membranes (PROM) and placental abruption. There were 42 cases with low Apgar scores and 94 cases with normal Apgar scores in the validation set. The ROC curve analysis showed that AUCs of LR, XGBoost, RF and SVM were 0.804 (95%CI 0.759‑0.845), 0.795 (95%CI 0.715‑0.868), 0.817 (95%CI 0.730‑0.890) and 0.802 (95%CI 0.750‑0.840), respectively. RF had most balanced sensitivity (0.798, 95%CI 0.665‑0.845) and specificity (0.786, 95%CI 0.628‑0.891) with highest F1 score (0.843). SHAP analysis showed that the influencing factors in the model were ranked by importance as follows: fetal malpresentation, cervical insufficiency, fetal distress, maternal anemia, antenatal fever, gestational age, abnormal umbilical artery blood flow, PROM and placental abruption. Conclusions The model based on RF algorithm has a good predictive efficacy for low Apgar score in VPIs and maybe a useful tool for prenatal risk assessment.
