基于随机森林算法构建良性前列腺增生术后并发尿路感染的风险预测模型及防治策略研究  

Study on the risk prediction model and prevention strategy of urinary tract infection after benign prostatic hyperplasia surgery based on random forest algorithm

作　　者：许婷婷 陈海莺 魏俞博 许清清 黄亚云 Xu Tingting;Chen Haiying;Wei Yubo;Xu Qingqing;Huang Yayun(Department of Urology,901 Hospital,Joint Logistic Support Force,Quanzhou 362000,Fujian,China;General Surgery,901 Hospital,Joint Logistic Support Force,Quanzhou 362000,Fujian,China)

机构地区：[1]联勤保障部队第九一〇医院泌尿外科,福建泉州362000 [2]联勤保障部队第九一〇医院普通外科,福建泉州362000

出　　处：《中国男科学杂志》2025年第1期39-45,共7页Chinese Journal of Andrology

摘　　要：目的探讨良性前列腺增生(BPH)术后并发尿路感染的危险因素,并建立、验证随机森林模型。方法回顾性分析282例BPH手术患者的临床资料,采用Logistic回归筛选BPH术后并发尿路感染的危险因素,通过R语言构建预测随机森林模型并对模型进行内部验证。结果BPH患者术后尿路感染率为30.14%;年龄>60岁、术前尿潴留导尿术、术前残余尿量≥160 mL、手术时间≥60 min、留置尿管时间≥5 d、MCP-1升高、IL-6升高是BPH术后并发尿路感染的危险因素,而术前抗菌药物使用是尿路感染的保护因素(P<0.05);随机森林模型的受试者操作特征(ROC)曲线值高于多因素Logistic回归模型(Z=2.615,P=0.008);5折交叉验证的模型正确率为74.2%。结论本研究构建的随机森林模型具有临床预测BPH术后并发尿路感染的潜力。Objective To investigate the risk factors of urinary tract infection after benign prostatic hyperplasia(BPH)surgery and establish a predictive random forest model.Methods The clinical data of 282 patients who underwent BPH surgery were retrospectively analyzed,and the risk factors of urinary tract infection after BPH surgery were screened by Logistic regression.The predictive random forest model was constructed by R language and verified internally.Results The postoperative urinary tract infection rate was 30.14%in BPH patients.Age>60 years old,preoperative urinary retention catheterization,preoperative residual urine volume≥160 mL,operative time≥60 min,catheter retention time≥5 d,elevated the serum level of MCP-1 and IL-6 were the risk factors for postoperative urinary tract infection after BPH surgery,while preoperative antibiotic use was the protective factor for urinary tract infection(P<0.05).The receiver operating characteristic(ROC)curve value of random forest model was higher than that of multivariate Logistic regression model(Z=2.615,P=0.008).The accuracy rate of the model was 74.2%.Conclusion The random forest algorithm constructed in this study has the potential to help clinical prediction of urinary tract infection after BPH surgery.

关 键 词：良性前列腺增生 术后尿路感染 LOGISTIC回归模型 随机森林模型 

分 类 号：R699.8[医药卫生—泌尿科学]