产时Ⅱ度及以上会阴裂伤风险预测模型的构建和验证

目的 构建产时Ⅱ度及以上会阴裂伤风险预测随机森林算法模型，并初步评价模型的预测性能。方法 采用方便抽样法，选取经阴道分娩的1 366例产妇为研究对象，将其按照7∶3的比例随机分为训练集和验证集。采用LASSO回归分析筛选产时Ⅱ度及以上会阴裂伤的风险因素，采用随机森林算法构建预测模型，计算ROC曲线下面积、预测准确率、灵敏度和特异度等评价模型的性能。结果 共计8个预测因子被纳入随机森林模型中，分别为孕前BMI、孕期体质量增加、初产妇、剖宫产史、硬膜外麻醉、催产、引产和胎儿估计体质量，其中胎儿估计体质量对产时Ⅱ度及以上会阴裂伤的影响最大，其次是初产妇和催产。随机森林模型在验证集中的ROC曲线下面积为0.698（95%CI：0.645~0.751），预测准确率为80.0%（95%CI：75.8%~83.8%），灵敏度和特异度分别为50.5%和89.1%。结论 基于随机森林算法构建的产时Ⅱ度及以上会阴裂伤风险预测模型具有一定的预测价值，但预测性能仍有待提高。

onstruction and validation of a risk prediction model for second-degree and above perineal laceration during delivery

Objective To construct a random forest algorithm model for predicting the risk of second-degree and above perineal laceration during delivery and to preliminarily evaluate the predictive performance of the model. Methods A total of 1,366 parturients who underwent vaginal delivery were selected as the study subjects using convenient sampling method. They were randomly divided into a training set and a validation set in a 7∶3 ratio. LASSO regression analysis was employed to screen the risk factors for second-degree and above perineal lacerations. A random forest algorithm was then used to build the prediction model, and various performance metrics such as the area under the ROC curve, predictive accuracy, sensitivity, and specificity were calculated to evaluate the model. Results A total of 8 predictive factors were included in the random forest model, namely pre-pregnancy BMI, weight gain during pregnancy, primiparity, history of Cesarean section, epidural anesthesia, induction of labor, artificial labor, and estimated fetal weight. Among them, estimated fetal weight had the greatest impact on second-degree and above perineal lacerations during delivery, followed by primiparity and induction of labor. The area under the ROC curve of the random forest model in the validation set was 0.698 (95% CI:0.645~0.751), with a predictive accuracy of 80.0% (95% CI:75.8%~83.8%), and sensitivity and specificity of 50.5% and 89.1%, respectively. Conclusion The risk prediction model for second-degree and above perineal laceration during delivery, based on the random forest algorithm, has certain predictive value. However, the predictive performance still needs improvement.