文章摘要
李江南,蒲昱帆,李亚冉,等.首发急性缺血性脑卒中后认知障碍预测模型的构建与验证[J].中华物理医学与康复杂志,2026,48(10):895-904
首发急性缺血性脑卒中后认知障碍预测模型的构建与验证
  
DOI:10.3760/cma.j.cn421666-20251015-00909
中文关键词: 急性缺血性脑卒中  认知障碍  预测模型  机器学习  可解释性  列线图
英文关键词: acute ischemic stroke, cognitive impairment, predictive model, machine learning, interpretability, nomogram
基金项目:
作者单位
李江南 徐州医科大学附属淮安医院,淮安 223001 
蒲昱帆 徐州医科大学附属淮安医院,淮安 223001 
李亚冉 徐州医科大学附属淮安医院,淮安 223001 
张小兵 徐州医科大学附属淮安医院,淮安 223001 
李雪静 徐州医科大学附属淮安医院,淮安 223001 
摘要点击次数: 450
全文下载次数: 366
中文摘要:
  目的 构建并验证适用于首发急性缺血性脑卒中(AIS)后认知障碍的预测模型,并结合机器学习方法与列线图,为临床早期识别高危患者提供直观、可靠的风险评估工具。 方法 采用前瞻性设计,纳入在徐州医科大学附属淮安医院治疗的首发AIS患者,收集其人口学、实验室及影像学等多维度指标;最终共有627例患者于卒中后6个月时完成随访,采用简易精神状态检查量表(MMSE)评估其认知功能。将其中188例认知障碍患者纳入观察组,余439例认知正常患者纳入对照组,比较2组患者基线特征;随后按8∶2比例将上述患者随机划分为开发队列(n=502)与内部验证队列(n=125),另将在淮安市第一人民医院治疗的184例AIS患者纳入外部验证队列。通过Lasso回归筛选变量,并采用多因素Logistic回归确定独立预测因子。基于筛选所得变量构建7种机器学习模型并比较其预测性能,采用SHAP等方法进行可解释性分析,最终基于性能最优模型构建列线图。 结果 通过Lasso回归共筛选出7个预测因子,多因素Logistic回归分析显示脑梗死灶数量[OR=2.08,95%CI(1.58,2.72),P<0.001]、C反应蛋白水平[OR=1.02,95%CI(1.01,1.03),P<0.001]、额叶梗死[左侧或双侧:OR=5.19,95%CI(2.19,12.27),P<0.001]及中性粒细胞百分比[OR=1.06,95%CI(1.03,1.09),P<0.001]等是AIS后认知障碍的独立危险因素。通过对比各机器学习模型,发现Logistic Regression模型性能最优,其内部验证曲线下面积(AUC)为0.913(准确率为0.876),外部验证AUC为0.902(准确率为0.862)。可解释性分析显示脑梗死灶数量、中性粒细胞百分比、额叶梗死是最关键的预测变量。基于该模型构建的列线图具有较好的区分性能及临床参考价值。 结论 本研究构建的预测模型整合了多维度临床指标与机器学习方法,展现出较好的预测效能与可解释性;据此构建的列线图有助于临床快速评估个体风险,从而为AIS后认知障碍的早期识别与个体化干预提供了实用性工具。
英文摘要:
    Objective To develop and validate a prediction model for post-stroke cognitive impairment following first-episode acute ischemic stroke (AIS), integrating machine learning methods and a nomogram to provide an intuitive and reliable risk assessment tool for early identification of high-risk patients. Methods First-episode AIS patients were prospectively enrolled from the Affiliated Huai′an Hospital of Xuzhou Medical University, and multidimensional data including demographic, laboratory, and imaging indicators were collected. Follow-up at 6 months after stroke was completed for 627 patients, and cognitive function was assessed using the Mini-Mental State Examination (MMSE). Among them, 188 patients with cognitive impairment were assigned to the observation group and 439 cognitively normal patients to the control group, which were used for comparison of baseline characteristics;the patients were then randomly divided into a development cohort (n=502) and an internal validation cohort (n=125) at an 8∶2 ratio, while 184 patients from Huai′an First People′s Hospital served as an external validation cohort. Predictors were screened using LASSO regression, and independent predictors were further identified by multivariable Logistic regression. Based on the selected variables, seven machine learning models were constructed and their predictive performances were compared. Interpretability analysis was conducted using methods such as SHAP, and a nomogram was ultimately developed based on the optimal model. Results Lasso regression identified a total of 7 predictors. Multivariable Logistic regression analysis revealed that number of infarcts[OR=2.08, 95%CI(1.58-2.72), P<0.001], C-reactive protein [OR=1.02, 95%CI(1.01-1.03), P<0.001], left or bilateral frontal lobe infarct [OR=5.19, 95%CI(2.19-12.27), P<0.001], and neutrophil percentage[OR=1.06, 95%CI(1.03-1.09), P<0.001], among others, were independent risk factors for post-AIS cognitive impairment. Among the machine learning models, Logistic regression demonstrated the best performance, with an internal validation AUC of 0.913 (accuracy 0.876) and an external validation AUC of 0.902 (accuracy 0.862). Interpretability analysis indicated that number of infarcts, neutrophil percentage, and frontal lobe infarct were the most critical predictive variables. The nomogram constructed based on this model exhibited good discrimination performance and clinical reference value. Conclusions The prediction model developed in this study integrates multidimensional clinical indicators and machine learning methods, demonstrating good predictive performance and interpretability. The constructed nomogram facilitates rapid individual risk assessment in clinical practice, providing a practical tool for early identification and personalized intervention of cognitive impairment after AIS.
查看全文   下载PDF阅读器  HTML全文
关闭