目的 构建基于患者临床信息的机器学习模型,用于预测药物难治性额叶癫痫(FLE)患者术后复发风险,为临床干预提供参考依据。方法 回顾性分析2020-01至2023-12陆军军医大学第二附属医院神经外科和武警重庆总队医院神经外科收治的药物难治性FLE患者181例的临床资料,收集其临床基线信息及术后复发情况。采用logistic回归分析筛选与术后复发相关的危险因素,选取逻辑回归(LR)、线性判别分析(LDA)、支持向量机(SVM)3种机器学习算法,按8:2比例将患者分为训练集与测试集,构建术后复发预测模型。采用受试者操作特征(ROC)曲线、校准曲线、决策曲线及诺模图评估模型预测效能。结果 经logistic回归分析筛选出6个术后复发相关危险因素,分别为癫痫首发年龄、癫痫发作持续时间、癫痫发作频率、癫痫发作类型、服用抗癫痫药物种类数量、ICU住院天数(均P＜0.05)。基于训练集构建的3个预测模型中,测试集分析结果为:LR模型AUC为0.786,LDA模型AUC为0.782,SVM模型AUC为0.761;校准曲线显示3个模型预测概率与实际复发概率一致性良好,决策曲线提示模型具有较好的临床应用价值,诺模图可直观呈现患者术后复发概率。结论 基于癫痫首发年龄等6个危险因素构建的LR、LDA、SVM机器学习模型,均可有效预测药物难治性FLE患者术后复发风险,其中LR模型预测效能最优,可为临床制定个体化干预策略提供科学依据。

Objective To construct a machine learning model based on clinical information of patients for the prediction of the risk of postoperative recurrence in patients with drug-resistant frontal lobe epilepsy (FLE), and to provide reference for clinical intervention. Methods A total of 181 patients with drug-resistant FLE who were admitted to the Second Hospital Affiliated to PLA Army Medical University and Chongqing Municipal Corps Hospital of Chinese People's Armed Police Force from January 2020 to December 2023 were enrolled, and their clinical baseline information and postoperative recurrence status were collected. Logistic regression analysis was used to screen the risk factors related to postoperative recurrence, and three machine learning algorithms, namely logistic regression (LR), linear discriminant analysis (LDA), and support vector machine (SVM), were selected to construct a postoperative recurrence prediction model by dividing the patients into a training set and a test set at a ratio of 8:2. ROC curve, calibration curve, decision curve analysis (DCA) and nomogram were used to evaluate the predictive efficacy of the model. Results Six risk factors related to postoperative recurrence were screened out through logistic regression analysis, including the age of the first epilepsy onset, duration of epilepsy attacks, frequency of epilepsy attacks, type of epilepsy attacks, number of antiepileptic drug taken, and length of ICU stay (all P＜0.05). Among the three prediction models constructed based on the training set, the test set analysis results showed that the AUC of the LR model was 0.786, the AUC of the LDA model was 0.782, and the AUC of the SVM model was 0.761. The calibration curve showed that the prediction probabilities of the three models were consistent with the actual recurrence probabilities, the decision curve suggested that the models had good clinical application value, and the nomogram could intuitively present the postoperative recurrence probability of patients. Conclusions The LR, LDA, and SVM machine learning models constructed based on 6 risk factors such as the age of the first epilepsy onset can effectively predict the risk of postoperative recurrence in patients with drug-resistant FLE, among which the LR model has the best predictive efficacy, which can provide a scientific basis for the clinical formulation of individualized intervention strategies.