摘要
目的利用决策树算法构建重症颅脑损伤(STBI)术后脑疝形成的风险预测模型。方法回顾性分析郑州市第七人民医院2016-01—2023-12收治的472例STBI患者的临床资料,根据术后是否发生脑疝将其分为脑疝组(62例)和非脑疝组(410例)。采用多因素Logistic回归分析STBI术后脑疝形成的危险因素,以卡方自动交互检验算法构建风险预测决策树模型。采用受试者工作特征(ROC)曲线和混淆矩阵评估决策树模型的预测效能。结果STBI患者术后脑疝形成的发生率为13.14%(62/472)。脑疝组年龄≥60岁、开放性损伤、格拉斯哥昏迷量表(GCS)评分3~5分、急性生理和慢性健康状态评价Ⅱ(APACHEⅡ)评分≥10分、低血压、蛛网膜下腔出血、脑积水占比均高于非脑疝组(P<0.05)。多因素Logistic回归分析显示年龄≥60岁、开放性损伤、GCS评分3~5分、APACHEⅡ评分≥10分、低血压、蛛网膜下腔出血、脑积水均是STBI术后脑疝形成的危险因素(P<0.05)。STBI术后脑疝形成的风险预测决策树模型共生长5层,共计13个节点、7个终端节点,最终筛选6个解释变量,即年龄、损伤类型、GCS评分、APACHEⅡ评分、蛛网膜下腔出血、脑积水,其中脑积水是最重要的危险因素,决策树模型共提取7条分类规则,筛选出3类高危人群。ROC曲线分析显示,决策树模型、多因素Logistic回归模型预测STBI术后脑疝形成的曲线下面积分别为0.959(95%CI:0.936~0.975)、0.948(95%CI:0.924~0.966)。混淆矩阵分析显示,决策树模型的精确度79.17%,F值0.85。结论STBI术后脑疝形成的风险预测决策树模型共生长5层,包括年龄、损伤类型、GCS评分、APACHEⅡ评分、蛛网膜下腔出血、脑积水共6个危险因素,其中脑积水是最为重要的危险因素,且该模型具有良好的预测效能,有助于临床筛选高风险患者以预防脑疝形成。
Objective To construct a risk prediction model for postoperative brain herniation formation in severe craniocerebral injury(STBI)by using decision tree algorithm.Methods Clinical data of 472 STBI patients admitted to the Seventh People’s Hospital of Zhengzhou from January 2016 to December 2023 were retrospectively analyzed,and they were divided into brain herniation group(n=62)and non-brain herniation group(n=410)based on whether postoperative brain herniation occurred.The risk factors for postoperative brain herniation formation in STBI was analyzed by multivariate Logistic regression method,and the risk prediction decision tree model was constructed by using the chi-square automatic interaction test(CHAID)algorithm.The predictive efficiency of the decision tree model was evaluated by the receiver operating characteristic(ROC)curve and confusion matrices.Results The incidence rate of postoperative brain herniation in STBI patients was 13.14%(62/472).The proportions of age≥60 years old,open injury,Glasgow coma scale(GCS)score of 3-5 points,acute physiological and chronic health evaluation Ⅱ(APACHEⅡ)score of≥10 points,hypotension,subarachnoid hemorrhage and hydrocephalus in the brain herniation group were higher than those in the non-brain herniation group(P<0.05).Multivariate Logistic regression analysis showed that the age≥60 years,open injury,GCS score 3-5,APACHEⅡ score≥10 points,hypotension,subarachnoid hemorrhage and hydrocephalus were all risk factors for the postoperative brain herniation formation in STBI(P<0.05).The risk prediction decision tree model for the postoperative brain herniation formation in STBI growed 5 layers,including 13 nodes and 7 terminal nodes,ultimately screened 6 explanatory variables,namely age,injury type,GCS score,APACHEⅡ score,subarachnoid hemorrhage and hydrocephalus,and hydrocephalus was the most important risk factor.A total of 7 classification rules were extracted and 3 high-risk groups were screened out from the decision tree model.ROC curve analysis results showed that the area under the curve of the decision tree model and the multivariate logistic regression model predicting the postoperative brain herniation formation in STBI were 0.959(95%CI:0.936-0.975)and 0.948(95%CI:0.924-0.966),respectively.The confusion matrix results showed that the accuracy of the decision tree model was 79.17%,and the F-value was 0.85.Conclusion The risk prediction decision tree model for the postoperative brain herniation formation in STBI grows of 5 layers,including 6 risk factors of age,injury type,GCS score,APACHEⅡ score,subarachnoid hemorrhage and hydrocephalus,and hydrocephalus is the most important risk factor among them,and the model has good predictive performance,which helps clinical screening of high-risk patients to prevent brain herniation formation.
作者
刘光辉
周春鹏
张国栋
LIU Guanghui;ZHOU Chunpeng;ZHANG Guodong(The Seventh People’s Hospital of Zhengzhou,Zhengzhou 450016,China;Xinxiang Central Hospital,Xinxiang 453000,China)
出处
《中国实用神经疾病杂志》
2025年第12期1528-1533,共6页
Chinese Journal of Practical Nervous Diseases
基金
2020年河南省医学科技攻关计划项目(编号:LHGJ20200953)。
关键词
重症颅脑损伤
脑疝
决策树算法
蛛网膜下腔出血
脑积水
危险因素
预测模型
Severe craniocerebral injury
Brain herniation
Decision tree algorithm
Subarachnoid hemorrhage
Hydrocephalus
Risk factors
Predictive model
