基于随机森林算法的急性肾损伤患者CRRT撤机成功因素及预测模型效能分析-现代医学

基于随机森林算法的急性肾损伤患者CRRT撤机成功因素及预测模型效能分析

单位：国药葛洲坝中心医院肾内科, 湖北宜昌 443002

分类号：R542.22;R692

出版年·卷·期（页码）：2023·51·第一期（52-59）

摘要：

目的: 基于随机森林算法下探讨急性肾损伤(AKI)患者连续性肾脏替代治疗(CRRT)撤机成功的预测因素，构建预测模型并分析模型的效能。方法: 纳入2019年8月至2022年5月在我院接受CRRT治疗的并发AKI的患者200例，并按7:3的比例随机分割为训练集(140例)和验证集(60例)。根据是否撤机成功，将患者分为撤机成功组和撤机失败组。收集训练集患者的临床实验室资料，采用多因素Logistic回归分析和随机森林算法分别构建影响AKI患者CRRT撤机成功的预测模型，比较两个预测模型的预测效能。结果: 训练集140例患者中撤机成功82例，撤机失败58例；验证集中撤机成功37例，撤机失败23例。训练集中两组患者撤机时感染相关器官功能衰竭评分系统(SOFA)评分、撤机后尿量、撤机后Scr水平、CRRT持续时间、撤机后中性粒细胞明胶相关载脂蛋白(NGAL)水平、撤机时急性生理与慢性健康评分(APACHE Ⅱ评分)、撤机后尿肾损伤分子-1(Kim-1)水平比较差异具有统计学意义(P<0.05)。多因素Logistic回归分析显示，撤机时SOFA评分(OR=5.774)、APACHEⅡ评分(OR=1.065)、CRRT持续时间(OR=1.153)、撤机后NGAL(OR=1.015)、Kim-1水平(OR=1.071)为影响AKI患者CRRT撤机成功的相关因素(均P<0.05)；随机森林算法中各变量的重要程度排序依次为撤机后Kim-1水平、撤机时SOFA评分、CRRT持续时间、撤机后NGAL水平、撤机时APACHE Ⅱ评分、撤机后尿量、撤机后Scr水平。随机森林算法的准确率、敏感度、特异度、阳性及阴性预测值显著高于Logistic模型(P<0.05)；ROC曲线结果显示，随机森林算法模型的诊断效能(AUC=0.947)高于多因素Logistic回归模型的诊断效能(AUC=0.714)(Z=3.536，P<0.001)。结论: 基于随机森林算法构建的预测模型能更有效预测AKI患者CRRT撤机失败风险。撤机时SOFA评分、撤机后Scr水平、CRRT持续时间、撤机后NGAL及Kim-1水平为预测AKI患者CRRT撤机成功的相关因素。

Objective: To investigate the predictive factors of the success prediction model of continuous renal replacement therapy(CRRT) weaning in patients with acute kidney injury(AKI) based on random forest algorithm and analyze the performance of the prediction model. Methods: A total of 200 patients with AKI who underwent CRRT in our hospital from August 2019 to May 2022 were included and randomly divided into a training set(140 cases) and a validation set(60 cases) in a ratio of 7:3. According to whether the weaning was successful or not, the patients were divided into the weaning success group and the weaning failure group. The clinical laboratory data of the training set were collected, and the multivariate Logistic regression analysis and random forest algorithm were used to construct the predictive models affecting the success of CRRT weaning in AKI patients, respectively, and the predictive performance of the two predictive models was compared.Results: Among the 140 patients in the training set, 82 were successfully weaned and 58 failed to wean; 37 were successfully weaned and 23 failed in the validation set. In the training set, the Sequential Organ Failure Assessment(SOFA) score, urine volume after weaning, Scr level after weaning, duration of CRRT, neutrophil gelatinase-associated lipocalin(NGAL) level, Acute Physiology and Chronic Health Evaluation(APACHE Ⅱ score) during weaning, Kidney injury molecule 1 levels(Kim-1) after weaning were statistically different between the two groups(P<0.05). Multivariate logistic regression analysis showed that the SOFA score(OR=5.774), APACHEⅡ score(OR=1.065), CRRT duration(OR=1.153), NGAL after weaning(OR=1.015), Kim-1 level(OR=1.071) were relevant factors affecting the success of CRRT weaning in AKI patients(all P<0.05). The order of importance of each variable in the random forest model was Kim-1 level after weaning, SOFA score during weaning, duration of CRRT, NGAL level after weaning, APACHE Ⅱ score during weaning, urine output and Scr levels after weaning. The accuracy, sensitivity, specificity, positive and negative predictive values of the random forest model were significantly higher than those of the logistic model(P<0.05); the ROC curve results showed that the diagnostic performance of the random forest algorithm model(AUC=0.947) was higher than that of the logistic regression model(AUC=0.714)(Z=3.536, P<0.001).Conclusion: The prediction model based on random forest algorithm can effectively predict the risk of CRRT weaning failure in AKI patients. SOFA score at the time of weaning, Scr level after weaning, duration of CRRT, NGAL and Kim-1 levels after weaning are the relevant factors for predicting the success of CRRT weaning in AKI patients.

参考文献：

[1] BELLOMO R,RONCO C,MEHTA R L,et al.Acute kidney injury in the ICU:from injury to recovery:reports from the 5th Paris International Conference[J].Ann Intensive Care,2017,7(1):49.
[2] REWA O G,VILLENEUVE P,LACHANCE P,et al.Quality indicators of continuous renal replacement therapy(CRRT) care in critically ill patients:a systematic review[J].Intensive Care Med,2017,43(6):750-763.
[3] CHO A Y,YOON H J,LEE K Y,et al.Clinical characteristics of sepsis-induced acute kidney injury in patients undergoing continuous renal replacement therapy[J].Ren Fail,2018,40(1):403-409.
[4] 李承圣，包绮晗，郝晓燕，等.基于随机森林算法的胰腺癌术后预测模型构建[J].吉林大学学报(医学版)，2022，48(2)：426-435.
[5] 郭锦洲，译.谢红浪，校.改善全球肾脏病预后组织(KDIGO)临床实践指南：急性肾损伤[J].肾脏病与透析肾移植杂志，2013(1)：57-60.
[6] 血液净化急诊临床应用专家共识组.血液净化急诊临床应用专家共识[J].中华急诊医学杂志，2017，26(1)：24-36.
[7] 文丹，李志辉.连续肾脏替代疗法治疗急性肾损伤的时机选择[J].儿科药学杂志，2021，27(7)：50-53.
[8] 谢龙昇，余春澜，谢鑫，等.并发急性肾损伤重症病人CRRT撤机失败的预测模型构建[J].护理研究，2022，36(2)：230-234.
[9] 张琪，费雅楠，姜利.ICU脓毒症合并急性肾损伤患者CRRT后死亡?危险因素：一项多中心观察研究数据的二次分析[J].中华危重病急救医学，2019，31(2)：155-159.
[10] LIN K,HU Y,KONG G.Predicting in-hospital mortality of patients with acute kidney injury in the ICU using random forest model[J].Int J Med Inform,2019,125(5):55-61.
[11] CONNOR M J,KARAKALA N.Continuous renal replacement therapy:reviewing current best practice to provide high-quality extracorporeal therapy to critically ill patients[J].Adv Chronic Kidney Dis,2017,24(4):213-218.
[12] DA HORA PASSOS R,RAMOS J G R,MENDONÇA E J B,et al.A clinical score to predict mortality in septic acute kidney injury patients requiring continuous renal replacement therapy:the HELENICC score[J].BMC Anesthesiology,2017,17(1):21.
[13] HANSRIVIJIT P,YARLAGADDA K,PUTHENPURA M M,et al.A meta-analysis of clinical predictors for renal recovery and overall mortality in acute kidney injury requiring continuous renal replacement therapy[J].J Crit Care,2020,60(7):13-22.
[14] STADS S,KANT K M,DE JONG M F C,et al.Predictors of short-term successful discontinuation of continuous renal replacement therapy:results from a prospective multicentre study[J].BMC Nephrol,2019,20(1):129.
[15] SOARES D B,MAMBRINI J V D M,BOTELHO G R,et al.Drug therapy and other factors associated with the development of acute kidney injury in critically ill patients:a cross-sectional study[J].Peer J,2018,6(8):e5405.
[16] WESTHOFF J H,SEIBERT F S,WALDHERR S,et al.Urinary calprotectin,kidney injury molecule-1,and neutrophil gelatinase associated lipocalin for the prediction of adverse outcome in pediatric acute kidney injury[J].Eur J Pediatr,2017,176(6):745-755.
[17] 王杰，林思浩，丁杰，等.急性肾损伤早期诊断标志物的研究进展[J].东南大学学报(医学版)，2022，41(3)：444-447.
[18] CHEN X H,CHEN Z W,WEI T T,et al.The effect of serum neutrophil gelatinase-associated lipocalin on the discontinuation of continuous renal replacement therapy in critically ill patients with acute kidney injury[J].Blood Purif,2019,48(1):10-17.
[19] 史小琴，钱瑶.连续性肾脏替代治疗联合中药灌肠对急性肾损伤患者尿KIM-1、NGAL水平的影响[J].陕西中医，2020，41(2)：210-212，216.
[20] KATAYAMA S,UCHINO S,UJI M,et al.Factors predicting successful discontinuation of continuous renal replaceement therapy[J].Anaesth Intensive Care,2016,44(4):453-457.

服务与反馈：

【文章下载】【发表评论】【查看评论】【加入收藏】

提示：您还未登录，请登录！点此登录