Huang Xudong, Zhang Lifeng, Zhang Chenyang, Li Jing, Li Chenyang
Department of Science and Education, Shenyang Maternity and Child Health Hospital, Shenyang, China.
Department of Maternal, Child and Adolescent Health, School of Public Health, Shenyang Medical College, Shenyang, China.
Front Med (Lausanne). 2025 Aug 7;12:1565374. doi: 10.3389/fmed.2025.1565374. eCollection 2025.
Postpartum depression (PPD) is a common and serious mental health complication after childbirth, with potential negative consequences for both the mother and her infant. This study aimed to develop an explainable machine learning model to predict the risk of PPD and to identify its key predictive factors.
A retrospective analysis was conducted on 1,065 women who attended their 6-week postpartum follow-up visit at a tertiary maternal and child healthcare hospital in Shenyang, China, from January to December 2021. Feature selection was performed using LASSO regression and the Boruta algorithm. Eight machine learning algorithms were then employed to construct the prediction models. Model performance was evaluated according to the area under the receiver operator characteristic curve (AUC), sensitivity, specificity, recall, F1 score, and accuracy. Shapley additive explanations (SHAP) were used to visualize the features of the model and individual case predictions.
Among the 1,065 women, 251 (23.5%) developed PPD. An 11-variable prediction model was developed, with XGBoost showing the best performance on both training and validation sets. After optimizing the model parameters and applying 10-fold cross-validation, the model achieved an average accuracy of 0.95, an average AUC of 0.955, average precision of 0.945, and average specificity of 0.985, indicating excellent predictive performance. The key predictive factors included weight gain during pregnancy, relationship with the mother-in-law, sleep quality, marital relationship, planned pregnancy, fetal sex preference, pregnancy-related anxiety, pelvic-floor muscle endurance, cervix status, attendance at prenatal education classes, and postpartum care satisfaction.
The XGBoost model demonstrated optimal performance at predicting PPD and can aid healthcare professionals to identify high-risk individuals. The SHAP method enhanced the model's interpretability, facilitating a better understanding of the causes of PPD, how to prevent it, and how to improve patient outcomes.
产后抑郁症（PPD）是分娩后常见且严重的心理健康并发症，对母亲及其婴儿都可能产生潜在的负面影响。本研究旨在开发一种可解释的机器学习模型，以预测PPD风险并识别其关键预测因素。
对2021年1月至12月在中国沈阳一家三级妇幼保健院进行产后6周随访的1065名妇女进行回顾性分析。使用LASSO回归和Boruta算法进行特征选择。然后采用八种机器学习算法构建预测模型。根据受试者工作特征曲线下面积（AUC）、敏感性、特异性、召回率、F1分数和准确性评估模型性能。使用Shapley加法解释（SHAP）来可视化模型特征和个体病例预测。
在1065名妇女中，251名（23.5%）发生了PPD。开发了一个包含11个变量的预测模型，XGBoost在训练集和验证集上均表现出最佳性能。在优化模型参数并应用10倍交叉验证后，该模型的平均准确率为0.95，平均AUC为0.955，平均精确率为0.945，平均特异性为0.985，表明具有出色的预测性能。关键预测因素包括孕期体重增加、与婆婆的关系、睡眠质量、婚姻关系、计划妊娠、胎儿性别偏好、妊娠相关焦虑、盆底肌肉耐力、宫颈状况、参加产前教育课程以及产后护理满意度。
XGBoost模型在预测PPD方面表现出最佳性能，可帮助医疗保健专业人员识别高危个体。SHAP方法增强了模型的可解释性，有助于更好地理解PPD的成因、预防方法以及如何改善患者预后。
