Scholarlink

Background <p>Early-onset preeclampsia (PE) poses significant risks for maternal and fetal outcomes, particularly when emergency delivery is required. This study aimed to develop machine learning models to predict emergency delivery within 48&#xa0;h of PE diagnosis, facilitating timely clinical interventions.</p> Methods <p>We analyzed a retrospective cohort of 648 singleton pregnancies diagnosed with PE at Fujian Maternal and Child Health Hospital from 2014 to 2024, with gestational ages ranging from 28 to 34 weeks. Patients were stratified into emergency delivery (≤ 48&#xa0;h post-diagnosis, <i>n</i> = 174) and non-emergency groups (<i>n</i> = 474). Feature selection was performed via univariate analysis, collinearity testing, and logistic regression, yielding 16 predictors. Eight machine learning models (logistic regression, naïve Bayes, XGBoost, LightGBM, SVM, GBDT, MLP, elastic net) were trained and evaluated for discriminative power and calibration. SHAP analysis was employed to interpret model predictions.</p> Results <p>XGBoost demonstrated superior performance (testing AUROC: 0.908) with excellent calibration, while GBDT achieved high AUROC (0.931) but poorer calibration. SHAP analysis identified CRP, D-dimer, and hypoproteinemia as the most influential predictors.</p> Conclusions <p>Machine learning models effectively predict emergency delivery in early-onset PE using clinically interpretable features. Integration into obstetric practice may enhance risk stratification, though prospective validation is warranted.</p> Graphical Abstract <p></p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Risk stratification and prediction of emergency delivery in early-onset preeclampsia using machine learning

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年11月6日

查看全文
  • Yanhong Xu,
  • Xinying Liu,
  • Ying Zhang,
  • Xingyi Qi,
  • Chengcheng Jin,
  • Zewei Liang,
  • Xia Xu,
  • Jianying Yan

摘要

Background

Early-onset preeclampsia (PE) poses significant risks for maternal and fetal outcomes, particularly when emergency delivery is required. This study aimed to develop machine learning models to predict emergency delivery within 48 h of PE diagnosis, facilitating timely clinical interventions.

Methods

We analyzed a retrospective cohort of 648 singleton pregnancies diagnosed with PE at Fujian Maternal and Child Health Hospital from 2014 to 2024, with gestational ages ranging from 28 to 34 weeks. Patients were stratified into emergency delivery (≤ 48 h post-diagnosis, n = 174) and non-emergency groups (n = 474). Feature selection was performed via univariate analysis, collinearity testing, and logistic regression, yielding 16 predictors. Eight machine learning models (logistic regression, naïve Bayes, XGBoost, LightGBM, SVM, GBDT, MLP, elastic net) were trained and evaluated for discriminative power and calibration. SHAP analysis was employed to interpret model predictions.

Results

XGBoost demonstrated superior performance (testing AUROC: 0.908) with excellent calibration, while GBDT achieved high AUROC (0.931) but poorer calibration. SHAP analysis identified CRP, D-dimer, and hypoproteinemia as the most influential predictors.

Conclusions

Machine learning models effectively predict emergency delivery in early-onset PE using clinically interpretable features. Integration into obstetric practice may enhance risk stratification, though prospective validation is warranted.

Graphical Abstract