初产妇经阴道分娩产后盆底功能障碍风险预测模型的构建与验证

Construction and Validation of a Risk Prediction Model for Pelvic Floor Dysfunction after Vaginal Delivery in Primiparous Women

摘要: 目的：探究初产妇产后盆底功能障碍(PFD)相关危险因素，基于LASSO回归、支持向量机(SVM)、随机森林3种机器学习算法构建产后PFD风险预测模型，对比各模型预测效能与临床应用价值。方法：选取2025年1月~2026年4月本院315例经阴道分娩初产妇，根据产后6~8周PFDI量表结果分为PFD组215例、非PFD组100例；按7:3随机拆分训练集236例、验证集79例。收集产妇人口学、妊娠分娩相关指标，单因素分析组间基线差异；LASSO回归筛选独立危险因素，分别构建Logistic回归(LR)、SVM、随机森林预测模型，以AUC、准确率、精确率、召回率、F1分数评价模型区分度与稳定性。结果：315例初产妇产后PFD发生率68.3%；LASSO回归筛选出6个核心危险因素：年龄、孕期体重增加、新生儿体重、足月分娩、妊娠期高血压、PFD家族史。训练集LR模型性能最优，AUC = 0.885，准确率86.6%，精确率95.1%，召回率88.7%，F1 = 0.918；验证集LR模型AUC降至0.649，泛化能力下降。SVM、随机森林模型训练、验证集区分度均低于LR模型，且验证集性能均出现衰减。结论：高龄、孕期增重过多、巨大儿、足月分娩、妊娠期高血压、PFD家族史是初产妇阴道分娩后PFD高危因素；基于6项易获取临床指标构建的Logistic回归模型在建模样本中预测效果良好，可辅助临床对产后PFD开展分层预防干预，但模型存在过拟合，仍需扩大样本、多中心外部验证优化。

Abstract: Objective: To explore the risk factors of postpartum pelvic floor dysfunction (PFD) in primiparous women, construct risk prediction models based on LASSO regression, support vector machine (SVM) and random forest, and compare the predictive efficacy of each model. Methods: A total of 315 primiparous women with vaginal delivery from January 2025 to April 2026 were enrolled, divided into PFD group (215 cases) and non-PFD group (100 cases). All samples were randomly split into training set (236 cases) and validation set (79 cases) at a ratio of 7:3. Maternal demographic, pregnancy and delivery data were collected. Univariate analysis was used to compare baseline differences, LASSO regression was applied to screen independent risk factors, then Logistic regression, SVM and random forest models were established. AUC, accuracy, precision, recall and F1-score were used to evaluate model performance. Results: The overall incidence of PFD was 68.3%. Six independent risk factors were screened out by LASSO regression: maternal age, gestational weight gain, neonatal birth weight, full-term delivery, gestational hypertension and family history of PFD. Logistic regression model achieved the best performance in training set (AUC = 0.885, accuracy = 86.6%, precision = 95.1%, recall = 88.7%, F1 = 0.918), while its AUC decreased to 0.649 in validation set with obvious overfitting. The predictive performance of SVM and random forest was inferior to Logistic regression both in training and validation sets. Conclusion: Advanced age, excessive gestational weight gain, macrosomia, full-term delivery, gestational hypertension and family history of PFD are critical risk factors for postpartum PFD. The Logistic regression model based on six routine clinical indicators shows favorable discrimination in training data, which can support stratified prevention of postpartum PFD in clinic. Further multi-center external validation with larger sample size is required to improve its generalization ability.