基于Relief特征选择算法的癌症患者缺血性脑卒中高危因素分析

Analysis of High Risk Factors of Ischemic Stroke in Cancer Patients by Relief Feature Selection Algorithm

目的:运用数据挖掘Relief特征选择算法,探讨癌症患者缺血性脑卒中(cancer-associated ischemic stroke,CAIS)发生的高危因素。方法:收集巴中市中心医院神经内科2016年1月至2021年12月连续收治的1 260例急性缺血性脑卒中(acute ischemic stroke, AIS)患者的临床资料,按照AIS发病前有无非神经系统活动性癌症分为癌症患者缺血性脑卒中组(CAIS组)和非癌症患者缺血性脑卒中组(NCAIS组),运用Relief特征选择算法对全部数据资料进行分析,计算每一个特征(危险因素)的权重,并对权重进行排序。结果:相对于NCAIS人群,按权重系数排序后有15个危险因素与CAIS发病有关,权重系数大于0.1的危险因素依次为:脑卒中疾病史、饮酒和高血脂症,是其最主要的危险因素;其次有血尿素氮、血尿酸、总胆固醇、同型半胱氨酸、高密度脂蛋白、甘油三酯、低密度脂蛋白、性别、谷丙转氨酶、谷草转氨酶、年龄、血肌酐是CAIS发生的危险因素;吸烟、高血压、糖尿病、短暂性脑缺血发作疾病史、房颤史在区分NCAIS及CAIS人群时的权重值为负值。结论:利用数据挖掘Relief特征选择算法可以挖掘CAIS发病密切相关的危险因素,主要包括脑卒中疾病史、饮酒和高血脂症。

Objective:To assess the high risk factors of ischemic stroke in cancer patients through data mining with Relief feature selection algorithm(RFSA).Methods:Clinical data of 1,260 patients with acute ischemic stroke(AIS)in the department of neurology in Bazhong Central Hospital from January 2016 to December 2021 were collected.Patients with or without non nervous system active cancer before the onset of AIS were assigned to the cancer-associated ischemic stroke(CAIS)group and the non-cancer-associated isch-emic stroke(NCAIS)group.RFSA was used to analyze all data,calculate the weight of each feature(risk factor),and rank the weight.Results:15 risk factors ranked by weight coefficients were related to the incidence of stroke.The top risk factors of CAIS were the history of stroke,drinking and hyperlipidemia,with weight coefficient greater than 0.1,followed by blood urea nitrogen,blood uric acid,total cholesterol,homocysteine,high-and low-density lipoprotein,triglycerides,sex,alanine aminotransferase,aspartate aminotransferase,age and blood creatinine.Weight values of smoking,hypertension,diabetes,the history of transient ischemic attack and the history of atrial fibrillation in differentiating CAIS patients were negative values.Conclusion:Data mining with RFSA can be used to mine the risk factors closely related to the onset of CAIS,mainly including the history of stroke,drinking and hyperlipidemia.