基于机器学习方法预测妇科手术麻醉恢复时间

Prediction of anesthesia recovery time for gynecological surgery by machine learning models

目的构建妇科手术中麻醉恢复时间预测模型并验证。方法回顾性收集来自湖南省妇幼保健院2022年9月至2023年8月495例妇科手术,数据预处理后,构建逻辑回归(LR)模型、支持向量机(SVM)模型、多层感知器(MLP)模型、自适应提升(AdaBoost)模型、梯度提升决策树(GBDT)模型、随机森林(RF)模型、极限梯度提升(XGBoost)模型、类别提升(CatBoost)模型及引导聚集(Bagging)算法9个模型。为评估机器学习模型的预测准确性,绘制了受试者操作特征(ROC)曲线并比较曲线下面积(AUC)。针对最佳模型,探索各个特征在最佳模型中的重要性,并进行十折交叉验证最佳模型的适用性和稳定性。结果在495例患者中,全身麻醉恢复时间≥2 h者为A组,94例;<2 h者为B组,有401例。根据性能指标,9个模型中最佳模型为XGBoost(AUC为0.8701)。特征重要性分析中发现拔管时间与全身麻醉恢复时间关系密切。在十折交叉验证,该模型表现出了良好的普适性和稳定性。结论利用多种数据构建的XGBoost模型,有良好的预测表现和临床价值。

【Objective】To construct and verify the prediction models of anesthesia recovery time during gynecological surgery.【Methods】All patients undergoing general anesthesia from September of 2022 to August of 2023 were collected retrospectively.After data pre-processing,using these vital sign vectors,9 machine learning models were constructed:logistic regression(LR)Model,support vector machine(SVM)model,multilayer perceptron(MLP)model,adaptive boosting(AdaBoost)model,gradient boosting decision tree(GBDT)model,random forest(RF)model,extreme gradient boosting(XGBoost)model,categorical boosting(CatBoost)model,and bagging ensemble algorithm.The predictive accuracy was assessed by the receiver operating characteristic(ROC)curve and the area under the curve(AUC).In order to find the best model,the importance of each feature in the model was explored.Ten-fold cross-validation was conducted in the best model in order to explore the applicability and stability of the model.【Results】Among the 495 patients,401 patients had a recovery time of less than 2 hours from general anesthesia and 94 patients had recovered from general anesthesia for 2 hours or more.Based on the performance metrics,the optimal model was the XGBoost model,which achieved an AUC value of 0.87.According to the importance of each feature in the model,the extubation time was the most valuable feature.Through the ten-fold cross-validation,XGBoost model showed good applicability and stability.【Conclusion】XGBoost model is constructed using a variety of data,which has satiafying prediction preformance and clinical value.