2025 年第 12 期 第 20 卷

基于SMOTE算法的心搏骤停复苏后癫痫风险预测模型

A risk prediction model for epilepsy after cardiac arrest and resuscitation based on SMOTE algorithm

作者：张晶卢生桂蔡宝裕郑伊静李奕鑫

英文作者：Zhang Jing Lu Shenggui Cai Baoyu Zheng Yijing Li Yixin

单位：中国人民解放军联勤保障部队第九一〇医院急诊医学科，泉州362000

英文单位：Department of Emergency Medicine the 910th Hospital of Joint Logistics Support Force of the Chinese People′s Liberation Army Quanzhou 362000 China

关键词：心搏骤停复苏成功；癫痫；SMOTE算法；预后模型

英文关键词：Successfulresuscitationofcardiacarrest;Epilepsy;SMOTEalgorithm;Prognosticmodel

• 摘要：

• 目的 探究基于SMOTE算法在心搏骤停复苏成功后癫痫风险预测模型。方法 回顾性收集2016年10月至2023年10月中国人民解放军联勤保障部队第九一〇医院收治的心搏骤停复苏成功患者200例作为观察对象，根据复苏成功后是否发生癫痫分为癫痫组（50例）和非癫痫组（150例）。收集和分析受试者资料，用单因素及多因素Logistic回归分析筛选心搏骤停复苏成功后并发癫痫的危险因素，再通过SMOTE算法重建影响因素原始数据集，得出风险预警模型并验证其预测效能。结果 200例心搏骤停复苏成功后并发癫痫者50例，其发生率为25.0%。多因素Logistic回归分析结果显示，年龄、癫痫史、颅内感染、心脑血管疾病、缺氧缺血性脑损伤是心搏骤停复苏成功后并发癫痫的危险因素（均P＜0.05），得到原始预警模型P1=1.107X1+1.221X2+1.025X3+1.227X4+1.162X5-3.909，经Hosmer-Lemeshow检验结果 表明，模型拟合度良好（决定系数R2＝0.418，P＝0.599）。基于SMOTE算法的预警模型P2=1.212X1+1.351X2+1.208X3+1.357X4+1.271X5-4.241，经Hosmer-Lemeshow检验结果表明，模型拟合度良好（决定系数R2＝0.617，P＝0.833）。受试者工作特征曲线分析结果显示，原始预警模型P1的曲线下面积为0.791，预警模型P2的曲线下面积为0.825。结论 基于SMOTE算法构建的预警模型较原始预警模型更优，能准确预警心搏骤停复苏成功后并发癫痫。

• Objective To explore the SMOTE algorithm based epilepsy risk prediction model after successful resuscitation of cardiac arrest. Methods A total of 200 patients with successful resuscitation of cardiac arrest admitted to the 910th Hospital of Joint Logistics Support Force of the Chinese People′s Liberation Army from October 2016 to October 2023 were retrospectively collected as observation objects. According to whether epilepsy occurred after successful resuscitation, the patients were divided into epilepsy group (50 cases) and non-epilepsy group (150 cases). Data of the subjects were collected and analyzed. Univariate and multivariate Logistic regression analysis were used to screen the risk factors of epilepsy after successful resuscitation of cardiac arrest, and then the SMOTE algorithm was used to reconstruct the original data set of influencing factors to obtain the risk early warning model and verify its predictive efficiency. Results There were 50 cases of epilepsy after successful resuscitation in 200 cases of cardiac arrest, and the incidence was 25.0%. Multivariate Logistic regression analysis showed that age, history of epilepsy, intracranial infection, cardiovascular and cerebrovascular diseases, and hypoxic-ischemic brain injury were risk factors for epilepsy after successful resuscitation of cardiac arrest (all P＜0.05). The original early warning model P1=1.107X1+1.221X2+1.025X3+1.227X4+1.162X5-3.909 was obtained. Hosmer-Lemeshow test results showed that the model had a good fit (coefficient of determination R2=0.418, P=0.599). The early warning model based on the SMOTE algorithm was P2=1.212X1+1.351X2+1.208X3+1.357X4+1.271X5-4.241. The Hosmer-Lemeshow test results indicated that the model had a good fit (coefficient of determination R2=0.617, P=0.833). The results of receiver operating characteristic curve analysis showed that the area under the curve of the original warning model P1 was 0.791, and the area under the curve of the early warning model P2 was 0.825. Conclusion The SMOTE early warning model is better than the original warning model, and can accurately predict epilepsy after successful resuscitation of cardiac arrest.