基于随机森林算法对首发缺血性脑卒中患者出院90 d复发的预测作用评价

Evaluation of the predictive effect of randomized forest algorithm on the recurrence of first ischemic stroke at 90 days after discharge from hospital

目的评价基于随机森林算法对首发缺血性脑卒中患者出院90 d复发的预测作用。方法回顾性分析2019年1月至2019年7月于本院就诊的580例首发缺血性脑卒中患者的临床资料,根据出院90 d是否复发分为复发组(n=110)和正常组(n=470),并采用随机森林模型与多因素Logistci回归模型筛选患者出院后90 d内复发的影响因素,比较两种方法的准确度、灵敏度、特意度、约登指数,并评价随机森林的预测效果。结果首发缺血性脑卒中患者出院90 d的复发率为18.96%(110/580)。多因素Logistic回归分析显示,饮酒、糖尿病、高脂血症、舒张压、载脂蛋白A是复发的独立危险因素(P<0.05)。随机森林模型显示,排名前6位的复发的影响因素分别为载脂蛋白A、天门冬氨酸氨基转移酶、白蛋白、红细胞压积、糖尿病、乳酸脱氢酶(重要程度分别为6.091、5.045、4.531、4.492、4.346、4.331、4.251、4.135、4.086、3.976)。随机森林模型准确性、灵敏度、约登指数均高于多因素Logistic回归分析模型。结论基于随机森林算法构建的首发缺血性脑卒中患者出院90 d复发的预测模型的预测效果较传统的多因素Logistic回归分析模型有显著优势,可用于首发缺血性脑卒中患者出院90 d复发的预测,具有一定的临床应用价值。

Objective To evaluate the predictive effect of randomized forest algorithm(RFSA)on recurrence of first-episode ischemic stroke patients 90 days after discharge from hospital.Methods The clinical data of 580 patients with first-episode ischemic stroke admitted to our hospital from January 2019 to July 2019 were retrospectively analyzed,according to the 90 days whether relapse they were divided into recurrence group(n=110)and normal group(n=470),and used the random forest model with multi-factor Logistci regression model to screen the influence factors of recurrence in patients after discharge 90 d,compared two methods of accuracy,sensitivity,specifically,Youden index,and evaluated the predicted effect of random forests.Results The recurrence rate of first-episode ischemic stroke was 18.96%(110/580)90 days after discharge.Multifactor Logistic regression analysis showed that alcohol consumption,diabetes,hyperlipidemia,diastolic blood pressure and apolipoprotein A were independent risk factors for recurrence(P<0.05).According to the random forest model,the top 6 influencing factors for recurrence were apolipoprotein A,aspartate aminotransferase,albumin,hematocrit,diabetes,and lactate dehydrogenase(the importance was 6.091,5.045,4.531,4.492,4.346,4.331,4.251,4.135,4.086,3.976,respectively).The accuracy,sensitivity and Youden index of the random forest model were higher than those of the multi-factor Logistic regression analysis model.Conclusion The prediction effect of the model based on random forest algorithm for the recurrence of first-episode ischemic stroke patients 90 days after discharge is significantly superior to the traditional multi-factor Logistic regression analysis model,which can be used for the prediction of the recurrence of first-episode ischemic stroke patients 90 days after discharge,and has certain clinical application value.