基于机器学习构建ICU行CRRT患者尿路感染风险预测模型-现代医学

基于机器学习构建ICU行CRRT患者尿路感染风险预测模型

单位：六安市人民医院重症医学科, 安徽六安 237000

分类号：R691.4

出版年·卷·期（页码）：2025·53·第八期（1231-1239）

摘要：

目的：使用4种机器学习算法构建重症监护室(ICU)行连续肾脏替代治疗(CRRT)患者尿路感染风险预测模型，通过比较选择最佳模型，并对模型决策依据进行可解释性分析。方法: 收集2020年1月至2024年7月六安市某三甲医院ICU收治的272例行CRRT治疗的患者。通过合成少数过采样技术(SMOTE)方法对数据进行平衡化处理，使用Lasso回归进行变量筛选，支持向量机(SVM)、极限梯度提升(XGBoost)、Logistic回归、K近邻(KNN)算法构建模型并对模型效能进行比较，选择最佳模型。结果：通过Lasso回归共筛选出13个变量，分别是：年龄、高血压、糖尿病、慢性肾功能不全、输血次数、白蛋白、血红蛋白、血清肌酐、急性生理与慢性健康评分系统Ⅱ(APACHEⅡ)评分、治疗天数、抗菌药、免疫抑制剂、侵入性操作。基于上述变量构建SVM、XGBoost、Logistic回归及KNN模型。在训练集中，各模型的受试者工作特征曲线下面积(AUC)分别为：0.98(0.97～0.99)、0.91(0.88～0.95)、0.97(0.95～0.99)、0.98(0.96～0.99)；在测试集中的AUC分别为：0.85(0.74～0.96)、0.69(0.53～0.84)、0.82(0.68～0.96)、0.75(0.61～0.89)。决策曲线显示，各模型均可使临床净收益增加。结论：SVM算法构建的预测模型可准确识别ICU行CRRT治疗患者早期尿路感染的发生，可为早期防治提供参考。此外，通过可解释性分析能够更好地理解模型中各变量对模型预测的贡献度，及各变量间的交互作用，为进一步深入研究奠定基础。

Objective: To construct a model for predicting urinary tract infection risk in patients undergoing continuous renal replacement therapy(CRRT) in intensive care unit(ICU) using four machine learning algorithms, to select the best model by comparison and to conduct interpretability analysis of the model decision-making basis. Methods: 272 patients admitted to the ICU of a tertiary hospital in Lu'an City who underwent CRRT from January 2020 to July 2024 were collected. The data were balanced by the Synthetic Minority Over-sampling Technique(SMOTE) method and the variables were screened using Lasso regression. Support vector machine(SVM), eXtreme gradient boosting(XGBoost), Logistic regression and K-nearest neighbor(KNN) algorithms were used to construct the model and compare the model efficacy to select the best model. Results: A total of 13 variables were screened by Lasso regression, i.e., age, hypertension, diabetes mellitus, chronic renal insufficiency, number of blood transfusions, albumin, haemoglobin, serum creatinine, Acute Physiology and Chronic Health Evaluation(APACHE) Ⅱ score, days of treatment, antimicrobials, immunosuppressants and invasive operations. SVM, XGBoost, Logistic regression and KNN models were constructed based on the above variables, and the areas under curve(AUC) of each model in the training set were 0.98(0.97-0.99),0.91(0.88-0.95),0.97(0.95-0.99),0.98(0.96-0.99), respectively and in the test set the AUC were 0.85(0.74-0.96), 0.69(0.53-0.84), 0.82(0.68-0.96), and 0.75(0.61-0.89), respectively. Decision curves showed that each model resulted in an increase in net clinical benefit. Conclusion: SVM algorithm for constructing prediction models can accurately identify the occurrence of early urinary tract infection in patients undergoing CRRT treatment in ICU, which can provide a reference for early prevention and treatment. In addition, the contribution of each variable in the model to the model prediction and the interaction between the variables can be better understood by interpretability analysis, which lays the foundation for further in-depth research.

参考文献：

[1] ZHANG X, YUAN Y.Effect of replacement therapy(CRRT) and hemodialysis(IHD) on severe acute renal failure[J].Front Pharmacol, 2023, 14:1122778.
[2] ZHANG Y, WANG L, MENG L, et al.Expression changes of autophagy-related proteins in AKI patients treated with CRRT and their effects on prognosis of adult and elderly patients[J].Immun Ageing, 2018, 15:1-6.
[3] 杨柳, 王改, 侯媛.影响慢性肾衰竭CRRT患者细菌感染发生的独立危险因素分析[J].中华保健医学杂志, 2024, 26(3):299-302.
[4] STANSKI N L, GIST K M, HASSON D, et.al.Characteristics and outcomes of children and young adults with sepsis requiring continuous renal replacement therapy:a comparative analysis from the worldwide exploration of renal replacement outcomes collaborative in kidney disease(WE-ROCK)[J].Crit Care Med.2024, 52(11):1686-1699.
[5] 姜艳华, 邢唯杰, 周兴梅, 等.连续性肾脏替代治疗患者中心静脉导管维护的最佳证据总结[J].护士进修杂志, 2021, 36(6):533-539.
[6] 肖茹.ICU行连续性肾脏替代治疗患者医院感染风险预测模型的构建[D].济南：山东大学, 2023.
[7] 李文静, 杨杰, 陈冬梅.WGCNA与机器学习算法识别结直肠癌发病机制中潜在PANoptosis关键基因[J].东南大学学报(医学版), 2023, 42(5):688-696.
[8] 肖桂林.基于机器学习算法预测行连续性肾脏替代治疗的急性肾损伤患者死亡率[D].南昌：南昌大学, 2023.
[9] 刘长生, 徐志杰.《EAU泌尿系统感染指南(2023)》要点解读[J].中华全科医学, 2024, 22(6):907-914.
[10] KRANZ J, BARTOLETTI R, BRUYÈRE F, et al.European Association of Urology Guidelines on Urological Infections:Summary of the 2024 Guidelines[J].Eur Urol, 2024, 86(1):27-41.
[11] 文细毛，任南，吴安华，等.全国医院感染监测网2012年综合ICU医院感染现患率调查监测报告[J].中国感染控制杂志，2014，13(8)：458-462.
[12] SANTIAGO M J, LÓPEZ-HERCE J, VIERGE E, et al.Infection in critically ill pediatric patients on continuous renal replacement therapy[J].Int J Artif Organs, 2017, 40(5):224-229.
[13] 杨丽萍, 程立军, 李潇, 等.某三甲中医医院ICU感染发生率时间序列分析及趋势预测[J].西部中医药, 2024, 37(9):78-82.
[14] UDDIN S, KHAN A, HOSSAIN M E, et al.Comparing different supervised machine learning algorithms for disease prediction[J].BMC Med Inform Decis Mak, 2019, 19(1):1-16.
[15] 黄若谷.基于XGBoost算法的糖尿病视网膜病变识别研究[D].温州:温州医科大学, 2023.
[16] 陈慧玲, 张媛, 徐敏, 等.医养结合机构老年坠积性肺炎患者多重耐药菌分布及感染的危险因素[J].中国老年学杂志, 2020, 40(16):3425-3427.
[17] COLCERIU M C, ALDEA P L, RǍCHIŞAN A L, et al.Vesicoureteral reflux and innate immune system:physiology, physiopathology, and clinical aspects[J].J Clin Med, 2023, 12(6):2380.
[18] SHENG S, ZHANG Y H, MA H K, et al.Albumin levels predict mortality in sepsis patients with acute kidney injury undergoing continuous renal replacement therapy:a secondary analysis based on a retrospective cohort study[J].BMC Nephrol, 2022, 23(1):52.

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