Abstract: Background　Large-artery atherosclerotic acute ischemic stroke (AIS) accounts for a substantial proportion of clinical stroke cases and is associated with poor prognosis. Stroke recurrence remains a major threat to long-term survival and quality of life in these patients. However, the characteristics of 1-year recurrence risk and its associated risk factors in this population remain unclear. Objective　To identify factors associated with 1-year stroke recurrence in patients with large-artery atherosclerotic AIS and to evaluate the advantages of a random survival forest (RSF) model in predicting recurrence risk. Methods　Patients diagnosed with large-artery atherosclerotic AIS in four hospitals in Nanchang from March 2019 to March 2021 were included. Baseline data within 48 hours of admission were collected using a standardized case report form. The follow-up outcome was stroke recurrence within 1 year. The dataset was randomly divided into a training set and a test set at a ratio of 7:3. An RSF model was constructed to predict AIS recurrence, and model parameters were optimized using five-fold cross-validation. The model was compared with a Cox regression model. Model performance was evaluated using the concordance index (C-index) and calibration curves. Results　A total of 1 186 patients were included in the final analysis. During the 12-month follow-up period after discharge, 102 patients experienced recurrence, with a recurrence rate of 8.6% (102/1 186), and were classified as the recurrence group; 1,084 patients had no recurrence and were classified as the non-recurrence group. Seventy patients died, with a mortality rate of 5.9% (70/1 186). Significant differences were observed between the recurrence and non-recurrence groups in glycated hemoglobin (GHb), creatinine (Cr), high-density lipoprotein (HDL), and National Institutes of Health Stroke Scale (NIHSS) score on day 3 after admission (P<0.05). Model performance evaluation showed that the overall C-index of the RSF model was 0.815 in the training set and 0.704 in the validation set, both higher than those of the Cox regression model (0.784 and 0.678, respectively). Receiver operating characteristic (ROC) curve analysis showed that, at 12 months of follow-up, the area under the ROC curve (AUC) of the RSF model for predicting recurrence was 0.833 (95%CI=0.788-0.872) in the training set and 0.718 (95%CI=0.634-0.795) in the validation set. In contrast, the AUC of the Cox regression model was 0.800 (95%CI=0.748-0.849) in the training set and 0.690 (95%CI=0.602-0.765) in the validation set. Calibration curves and decision curve analysis showed that the RSF model achieved favorable clinical incremental net benefit, and that the predicted recurrence probabilities were consistent with the observed recurrence probabilities. SHAP value ranking showed that the top three risk factors for recurrence were NIHSS score, alanine aminotransferase (ALT), and Cr. Conclusion　The RSF model developed in this study can effectively predict the 1-year risk of stroke recurrence in patients with large-artery atherosclerotic AIS. The top three factors associated with recurrence were NIHSS score at admission, ALT, and Cr. These findings may help clinicians perform individualized risk assessment and implement targeted interventions.

Key words: Acute ischemic stroke, Large artery atherosclerotic, Randomized survival forest, Recurrence, Risk prediction

摘要： 背景　大动脉粥样硬化型急性缺血性脑卒中（AIS）临床占比高、预后差，复发是威胁患者长期生存与生活质量的重要问题。然而，该类患者发病后1年的卒中复发风险特征及相关危险因素尚不明确。目的　探究大动脉粥样硬化型AIS患者1年内复发的影响因素，分析随机生存森林（RSF）模型对AIS复发风险预测的优势。方法　纳入2019年3月—2021年3月南昌四家医院确诊为大动脉粥样硬化型AIS患者，使用统一制订的病例报告表收集患者入院48 h内的基线资料，随访结局为1年内脑卒中复发情况。按照7∶3的比例将数据随机划分成训练集和测试集，构建预测AIS患者复发的RSF模型，采用5折交叉验证对模型参数进行优化，并与构建的Cox回归模型进行比较，利用一致性指数（C-index）和校准曲线评估模型性能。结果　研究最终纳入1 186例患者进行分析，在出院后12个月的随访期间，102例（8.6%，102/1 186）患者复发（作为复发组），1 084例未复发（记为未复发组）；70例（5.9%，70/1 186）患者死亡。复发组和未复发组患者糖化血红蛋白（GHb）、肌酐（Cr）、高密度脂蛋白（HDL）、入院第3天美国国立卫生研究院脑卒中量表（NIHSS）评分比较，差异有统计学意义（P<0.05）。模型性能评估显示，RSF模型的整体一致性指数（C-index）在训练集中为0.815，在验证集中为0.704，均高于Cox回归模型（0.784、0.678）。受试者工作特征（ROC）曲线结果显示，随访12个月时，在训练集中RSF模型预测患者复发的ROC曲线下面积（AUC）为0.833（95%CI=0.788~0.872），在验证集中AUC为0.718（95%CI=0.634~0.795），而Cox回归模型在训练集中AUC为0.800（95%CI=0.748~0.849），验证集AUC为0.690（95%CI=0.602~0.765）。校准曲线和决策曲线分析结果显示，RSF模型可获得较好的临床增量净收益且预测概率与实际复发概率具有一致性。SHAP值排序显示，影响患者复发的前3位风险因素依次为NIHSS评分、丙氨酸氨基转移酶（ALT）和Cr。结论　本研究构建的RSF模型能够有效预测大动脉粥样硬化型AIS患者1年内的脑卒中复发风险，影响患者复发的前3位风险因素依次为入院时NIHSS评分、ALT和Cr，有助于临床医师进行个体化风险评估与干预。

关键词: 急性缺血性脑卒中, 大动脉粥样硬化型, 随机生存森林, 复发, 风险预测