产后抑郁风险预测模型的构建和验证

Development and validation of postpartum depression risk prediction model

目的构建产后抑郁风险预测模型,并识别预测因子。方法选取住院分娩产妇835人为研究对象,按照时间段分为训练集722人及测试集113人,以产后6周是否发生产后抑郁为结局指标。利用logistic回归、支持向量机和随机森林3种监督学习算法建立风险预测模型,采用序列前向选择法筛选特征,通过网格搜索法调整模型参数。将训练好的模型在训练集上进行十折交叉验证,在测试集上进行外部验证。结果产妇产后6周抑郁发生率为22.6%(189/835)。经筛选,最终纳入14个预测因子。3种监督学习模型中,随机森林模型预测性能最佳,在测试集上的受试者工作特征曲线下面积、Brier得分、准确率、精确度、召回率和F1得分分别为0.943、0.073、0.903、0.684、0.722和0.703。结论基于随机森林的产后抑郁风险模型预测性能最佳,能够辅助医护人员识别高风险人群。

Objective To develop a risk prediction model for postpartum depression(PPD),and to identify the predictors.Methods A total of 835 women who gave birth in hospital were selected,and divided into a training set of 722 women and a test set of 113 ones according to the time period.The outcome variable was defined as the occurrence of PPD at 6 weeks.Three supervised machine learning algorithms,namely logistic regression,support vector machine and random forest,were used to build risk prediction models,and the features were screened by using the sequence forward selection method,and the model parameters were adjusted by using the grid search method.The trained model was subjected to ten-fold cross-validation on the training set and external validation on the test set.Results The overall incidence of PPD at 6 weeks was 22.6%(189/835).Fourteen predictors were eventually included.Among the three supervised learning models,the random forest model had the best prediction performance,with the area under the receiver operator characteristic curve,Brier score,accuracy,precision,recall and F1 values of 0.943、0.073、0.903、0.684、0.722 and 0.703.Conclusion The prediction model based on random forest algorithm can help health care workers to identify women at high risk of PPD.