基于D-二聚体、BUN/Alb、MHR、CSVD负荷评分的机器学习构建急性脑梗死后神经功能恢复的临床研究

doi:10.19854/j.cnki.1008-2425.2026.03.0004

摘要/Abstract

摘要： 目的分析急性脑梗死后神经功能恢复的危险因素,并基于D-二聚体(D-D)、尿素氮(BUN)与白蛋白(ALB)比值(B/A)、单核细胞计数和高密度脂蛋白胆固醇比值(MHR)、脑小血管病(CSVD)影像负荷评分,采用机器学习算法构建风险预测模型。方法选取2022年5月~2025年9月在我院收治的218例急性脑梗死患者为对象,根据患者治疗7d后NIHSS评分变化情况将其分为恢复良好组(177例)和恢复不良组(41例)。收集患者一般临床资料、血清D-D、B/A、MHR及CSVD影像负荷评分。采用单因素和多因素分析急性脑梗死后神经功能恢复不良的相关因素,并基于筛选出的危险因素分别应用决策树、随机森林及合成少数类过采样技术(SMOTE)算法构建风险预测模型,比较不同机器学习模型对急性脑梗死后神经功能恢复不良的预测价值。结果恢复良好组与恢复不良组在年龄、基线NIHSS评分、D-D、B/A、MHR、CSVD影像负荷评分方面比较,差异均有统计学意义(P＜0.05)。Logistic回归分析筛选出D-D、基线NIHSS评分、B/A、MHR、CSVD影像负荷评分4个变量。模型预测效能比较显示,AUC值由高到低依次为随机森林模型、SMOTE模型、决策树模型。结论基于D-二聚体、基线NIHSS评分、BUN/Alb、MHR及CSVD负荷评分构建的风险预测模型,能够有效整合多个危险因素,对急性脑梗死后神经功能恢复不良具有良好的预测价值。

关键词: 急性脑梗死, 神经功能恢复不良, 机器学习, 风险预测

Abstract: Objective To analyze the risk factors for neurological function recovery after acute cerebral infarction,and to construct a risk prediction model using machine learning algorithms B/Ased on D-dimer (D-D),【blood urea nitrogen/albumin ratio (B/A)】,【monocyte-to-high-density lipoprotein cholesterol ratio (MHR)】,and cerebral small vessel disease (CSVD) imaging load score. Methods 218 patients with acute cerebral infarction who were admitted to our hospital from May 2022 to September 2025 were selected as the subjects.They were divided into the good recovery group (177 cases) and the poor recovery group (41 cases) according to the changes in NIHSS score 7 days after treatment.General clinical data,serum D-D,B/A,MHR and CSVD imaging load score of the patients were collected.Univariate and multivariate analyses were conducted to identify the related factors for poor neurological function recovery after acute cerebral infarction,and risk prediction models were constructed using decision tree,random forest and synthetic minority oversampling technique (SMOTE) algorithms B/Ased on the selected risk factors.The predictive value of different machine learning models for poor neurological function recovery after acute cerebral infarction was compared. Results There were statistically significant differences between the good recovery group and the poor recovery group in terms of age,baseline NIHSS score,D-D,B/A,MHR,and CSVD image burden score (P<0.05.Logistic regression analysis identified four variables:D-D,baseline NIHSS score,B/A,MHR,and CSVD imaging load score.The comparison of model predictive efficacy showed that the AUC values from high to low were the random forest model,the SMOTE model,and the decision tree model. Conclusion The risk prediction model constructed based on D-dimer,baseline NIHSS score,BUN/Alb,MHR,and CSVD load score can effectively integrate multiple risk factors and has good predictive value for poor neurological recovery after acute cerebral infarction.

Key words: Acute cerebral infarction, Poor neurological function recovery, Machine learning, Risk prediction

中图分类号:

R743.3

张亚平, 任小雨, 吴升华, 龙苏生. 基于D-二聚体、BUN/Alb、MHR、CSVD负荷评分的机器学习构建急性脑梗死后神经功能恢复的临床研究[J]. 立体定向和功能性神经外科杂志, 2026, 39(3): 152-159.

Zhang Yaping, Ren Xiaoyu, Wu Shuhua, Long Susheng. A clinical study on the assessment of neurological recovery after acute cerebral infarction based on D-dimer,BUN/Alb,MHR,and CSVD load score using machine learning[J]. Chinese Journal of Stereotactic and Functional Neurosurgery, 2026, 39(3): 152-159.

参考文献

[1] Li C,Xu BF,Zhang M,et al.Severe thrombocytopenia with acute cerebral infarction:a case report and literature review[J].Niger J Clin Pract,2023,26(7):1040-1044.
[2] Zheng DL,Li XM,Fu YP.Risk factors of acute cerebral infarction in patients with primary hypertension[J].Ir J Med Sci,2023,192(5):2441-2445.
[3] 韦韬,高文,陈红.长链非编码RNA在急性脑梗死中作用研究进展[J].中国老年学杂志,2023,43(24):6134-6137.
[4] 陈婷,蒋小英,林燕妮.阿替普酶与尿激酶溶栓治疗急性脑梗死的效果比较[J].临床合理用药,2025,18(19):45-48.
[5] 孙同.脑梗死后患者神经功能恢复不良的危险因素分析[J].中国实用神经疾病杂志,2017,20(1):82-83.
[6] 闫海成,任君浩,武强,等.髓鞘碱性蛋白、S100钙结合蛋白B与脑梗死介入治疗后早期神经功能恶化风险的关联性研究[J].四川大学学报(医学版),2025,56(5):1387-1391.
[7] Jackson S,Mercer C,Singer BJ.An exploration of factors influencing physical activity levels amongst a cohort of people living in the community after stroke in the south of England[J].Disabil Rehabil,2018,40(4):414-424.
[8] 刘莉,周依林,邢雪梅.尿素氮与白蛋白比值对老年急性脑梗死患者病情、预后的评估意义[J].中国老年学杂志,2023,43(7):1560-1563.
[9] 李海云,王耀虹,杜梦琪.急性脑梗死患者纤维蛋白原降解产物、D-二聚体/纤维蛋白原比值和载脂蛋白(a)水平变化与早期神经功能恶化的相关性[J].微循环学杂志,2026,36(1):85-91.
[10] 孔玉娟,武钟毅,闫彩萍,等.纤维蛋白原/白蛋白比值与脑小血管病总负荷评分的相关性及对急性缺血性脑卒中患者早期神经功能恶化的预测价值[J].中风与神经疾病杂志,2026,43(1):52-59.
[11] 金建,鞠新越,郭洋君,等.老年缺血性脑卒中患者纤维蛋白原与白蛋白比值对脑小血管病MRI负荷和功能预后的关系[J].中华老年心脑血管病杂志,2025,27(12):1709-1713.
[12] Mathew M,John A,Vazhakkalayil Ramachandran R.Nurse stress and patient safety in the ICU:physician-led observational mixed-methods study[J].BMJ Open Qual,2025,14(2):3109.
[13] 朱慧明,洪明阳,王守卫,等.基于多算法改进机器学习结核性胸膜炎的风险预测模型研究[J/OL].中华医院感染学杂志,1-7[2026-03-27].
[14] 王大伟,郑正,方媛,等.基于九种机器学习算法构建肛周脓肿多重耐药菌感染风险预测模型[J/OL].中华医院感染学杂志,1-7[2026-03-27].
[15] 李光硕,赵性泉.《中国急性缺血性卒中诊治指南2023》解读[J].中国卒中杂志,2024,19(8):956-961.
[16] Kim JM,Park KY,Kim HR,et al.Association of bone mineral density to cerebral small vessel disease burden[J].Neurology,2021,96(9):1290-1300.
[17] Roushdy T,Abdel Nasser A,Nasef A,et al.A clinical comparative analysis between expanded NIHSS and original NIHSS in posterior circulation ischemic stroke[J].J Clin Neurosci,2023,114(1):77-80.
[18] Naess H,Kurtz M,Thomassen L,et al.Serial NIHSS scores in patients with acute cerebral infarction[J].Acta Neurol Scand,2016,133(6):415-420.
[19] Chen C,Ma F,Wen XL,et al.Clinical efficacy of the Huo Xue Hua Yu method combined with aspirin in the Treatment of acute cerebral infarction:a systematic evaluation and meta-analysis[J].Curr Pharm Des,2023,29(25):2009-2017.
[20] 黄秀婷,李培.2型糖尿病合并急性脑梗死患者神经功能恢复状态影响因素的研究[J].中国糖尿病杂志,2025,33(5):345-349.
[21] 崔明明.影响急性脑梗死患者静脉溶栓后神经功能的相关因素分析[J].临床医学,2019,39(7):43-45.
[22] 王俞洁.急性脑梗死患者Fbg、D-D、FDP、Hcy水平的变化分析及其与早期神经功能的关系[J].中外医学研究,2025,23(12):49-52.
[23] 刘莉,周依林,邢雪梅.尿素氮与白蛋白比值对老年急性脑梗死患者病情、预后的评估意义[J].中国老年学杂志,2023,43(7):1560-1563.
[24] 邵敏洁,毛友兵,周辰珩,等.MHR、LMR与急性缺血性脑卒中患者颈动脉内膜增厚及早期神经功能恶化的关系[J].中国实用医药,2023,18(17):1-5.
[25] 苏春贺,王铭,周东杨,等.脑小血管病总负荷与急性脑梗死患者早期神经功能恶化的相关性[J].中国实用神经疾病杂志,2022,25(2):145-149.
[26] 柯四水,廖静.基于决策树算法构建急性胰腺炎肠外营养患者静脉导管相关血流感染预测模型[J/OL].中国医院统计,1-7[2026-03-27].
[27] Nguyen T,Mengersen K,Sous D,et al.SMOTE-CD:SMOTE for compositional data[J].PLoS One,2023,18(6):287705.
[28] Jin YD,Lan AL,Dai YR,et al.Development and testing of a random forest-B/Ased machine learning model for predicting events among breast cancer patients with a poor response to neoadjuvant chemotherapy[J].Eur J Med Res,2023,28(1):394.