汕头大学医学院学报2026，Vol.39Issue(2)：89-94,6.DOI:10.13401/j.cnki.jsumc.2026.02.005
重症监护病房医院感染相关因素分析及机器学习风险预测模型构建
Analysis of associated factors for nosocomial infection in intensive care unit and construction of machine learning risk prediction model
摘要
Abstract
Objective:To construct a risk prediction model for nosocomial infection in intensive care unit(ICU)based on machine learning algorithms,and to identify key associated factors.Methods:A total of 13 716 patients aged≥18 years with ICU hospitalization time≥3 days in Shantou Central Hospital from May 2018 to December 2024 were retrospectively enrolled.Recursive feature elimination with cross-validation(RFECV)was used to screen key features.Three risk prediction models including Random Forest,XGBoost and Logistic regression were constructed respectively.The model performance was evaluated by area under the receiver operating characteristic curve(AUC),sensitivity,specificity and F2 score.SHAP(Shapley Additive exPlanations)method was used to analyze feature importance.Results:A total of 13 716 patients were enrolled,including 405 cases of nosocomial infection,with an incidence rate of 2.95％.Hospitalization time≥10 days,multidrug-resistant organism infection,fever,history of surgery during hospitalization,combined use of antimicrobial agents,mechanical ventilation,central venous catheterization and urinary catheterization were significantly associated to the occurrence of nosocomial infection(P<0.05).Among the three models,XGBoost showed the optimal performance,with AUC of 0.807(95％CI:0.774-0.832),sensitivity of 72.8％,specificity of 73.4％,and F2 score of 0.272.The Brier score was 0.027 after Isotonic regression calibration.SHAP analysis showed that hospitalization time≥10 days,combined use of antimicrobial agents,continuous fever time≥3 days,multidrug-resistant organism infection and history of surgery during hospitalization were the top five risk enhancement factors.Conclusion:The ICU nosocomial infection risk prediction model constructed based on XGBoost algorithm has moderate discrimination and good calibration,which can provide a reference tool for early identification of high-risk patients in clinical practice.关键词
重症监护病房/医院感染/风险预测模型/机器学习/XGBoostKey words
intensive care unit/nosocomial infection/risk prediction model/machine learning/XGBoost分类
医药卫生引用本文复制引用
杜沛玲,温熙麟,周霓..重症监护病房医院感染相关因素分析及机器学习风险预测模型构建[J].汕头大学医学院学报,2026,39(2):89-94,6.基金项目
汕头市医疗卫生科技计划(211114106491754) （211114106491754）
