基于深度学习的全身麻醉患者围术期非计划低体温预测模型的构建与应用

Construction and application of inadvertent perioperative hypothermia prediction model for patients under general anesthesia based on deep learning

目的构建全身麻醉患者围术期非计划低体温预测模型并应用于临床,验证其性能。方法纳入2016年1月至2020年9月浙江省某三级甲等医院19068例手术患者数据,运用基于深度学习的人工智能技术构建模型,采用受试者操作特征曲线下面积和决策曲线检验模型的预测效果。于2020年10月至2021年3月纳入2157例手术患者对模型的预测准确率进行检验。结果建模组的手术患者中低体温发生率为13.89%(2649/19068),验证组手术患者低体温发生率为14.18%(306/2157),预测模型的受试者操作特征曲线下面积为0.724(95%CI:0.707~0.741),灵敏度为0.516,特异度为0.823,截断值为0.175,实际应用的准确率为79.54%。结论本研究模型能够稳定的预测全身麻醉患者围术期非计划低体温的发生率,可为临床预防围术期非计划低体温提供参考。

Objective To construct a prediction model of inadvertent perioperative hypothermia in patients under general anesthesia,and to apply to clinic to verify its performance.Methods The data of 19068 surgical patients in a GradeⅢClass A hospital in Zhejiang Province from January 2016 to September 2020 were included.The model was constructed by using artificial intelligence technology based on deep learning,and the prediction effect of the model was tested by using the area under the subject operating characteristic curve and decision curve.Totally 2157 surgical patients were included from October 2020 to March 2021 to test the prediction accuracy of the model.Results The incidence of hypothermia was 13.89%(2649/19068)in the modeling group and 14.18%(306/2157)in the validation group.The area under the subject operating characteristic curve of the prediction model was 0.724(95%CI:0.707-0.741),the sensitivity was 0.516,the specificity was 0.823,the cut-off value was 0.175,and the accuracy of practical application was 79.54%.Conclusions This model can stably predict the incidence of perioperative inadvertent hypothermia in patients under general anesthesia,and provide reference for clinical prevention of inadvertent perioperative hypothermia.