基于随机森林算法的急性持续性眩晕患者脑卒中风险预测模型构建

Construction of risk prediction model for stroke in the patients with acute persistent vertigo based on random forest algorithm

目的构建基于随机森林算法的急性持续性眩晕患者脑卒中风险预测模型。方法回顾性选取2019年1月至2022年12月郑州市第七人民医院400例急性持续性眩晕患者为研究对象,采用6∶4比例将患者分为训练集(240例)及验证集(160例)。利用R 4.1.3软件包将最终获得的训练集数据中未发生脑卒中的患者定义为0,发生脑卒中的患者定义为1,收集受试者临床数据,采用软件包中随机森林等数据包筛选急性持续性眩晕患者脑卒中发生的影响因素,并构建基于随机森林算法的预测模型,通过绘制列线图对预测模型进行可视化处理,使用一致性指数(C-index)、决策曲线分析预测结果;通过验证集数据验证该模型预测效能。结果随机森林算法共筛选出中枢性眩晕史、高血压、房颤史、高脂血症4个变量为急性持续性眩晕患者脑卒中发生的影响因素,构建基于随机森林算法的预测模型,绘制列线图显示,预测急性持续性眩晕患者脑卒中的C-index为0.846(95%CI 0.760~0.910),校正曲线显示绝对误差为0.043;验证集C-index为0.827(95%CI 0.738~0.895),绝对误差为0.048。受试者工作特征(ROC)曲线分析显示,预测模型预测急性持续性眩晕患者脑卒中的ROC曲线下面积(AUC)为0.820,敏感度为81.00%,特异度为78.00%。决策曲线分析训练集的阈值概率范围为22%~100%,在该范围内根据模型的预测概率进行干预的临床净收益高于对所有人不进行(无)和对所有人进行干预(所有)。结论中枢性眩晕史、高血压、房颤史、高脂血症是持续性眩晕患者脑卒中风险预测因子,基于此建立的随机森林算法预测模型可用于急性持续性眩晕患者脑卒中发生风险的预测。

Objective To construct a risk prediction model for stroke in the patients with acute persistent vertigo based on random forest algorithm.Methods 400 patients with acute persistent vertigo in Zhengzhou Seventh People's Hospital from January 2019 to December 2022 were selected retrospectively as research subjects,and were randomly divided into training set(240 cases)and validation set(160 cases)with a ratio of 6∶4.The patients who had not suffered from stroke in the final training set data were defined as 0 and those who had suffered from stroke were defined as 1,and the clinical data of the subjects were collected by using R 4.1.3 software.The random forest and other data packets in R 4.1.3 software were used to screen the influencing factors of stroke in the patients with acute persistent vertigo,and a prediction model based on random forest algorithm was constructed.The prediction model was visualized by drawing a nomograph,and the consistency index(C-index)and decision curve analysis were used to analyze the prediction results.Then the prediction efficiency of the model was verified by the validation set data.Results The random forest algorithm screened out the influencing factors of stroke in the patients with acute persistent vertigo,including four variables of central vertigo history,hypertension,atrial fibrillation history and hyperlipidemia.A prediction model based on the random forest algorithm was constructed,and the nomograph showed that the C-index of the prediction model in predicting stroke in the patients with acute persistent vertigo was 0.846(95%CI 0.760-0.910),and the absolute error of the calibration curve was 0.043.The C-index of validation set was 0.827(95%CI 0.738-0.895),and the absolute error was 0.048.The receiver operating characteristic(ROC)curve analysis showed that the area under the ROC curve(AUC)of prediction model for predicting the occurrence of stroke in the patients with acute persistent vertigo was 0.820,the sensitivity was 81.00%,and the specificity was 78.00%.When the threshold probabilities of the training set were 22%-100%by decision curve analysis,the clinical benefits after clinical intervention based on the predicted probability of the model within the range were higher than those of the patients with and without interventions.Conclusions The central vertigo history,hypertension,atrial fibrillation history and hyperlipidemia are the predictive factors for stroke in the patients with persistent vertigo.The prediction model based on random forest algorithm can be used for the prediction of stroke risks in the patients with acute persistent vertigo.