Guo Xiaolan, Wu Dansen, Chen Xiaoping, Lin Jing, Chen Jialong, Wang Liming, Shi Songjing, Yang Huobao, Liu Ziyi, Hong Donghuang
Department of Critical Care Medicine, Shengli Clinical Medical College of Fujian Medical University, Fujian Provincial Hospital, Fuzhou University Affiliated Provincial Hospital, Fuzhou, Fujian, People's Republic of China.
Computer Science and Mathematics, Fujian University of Technology, Fuzhou, Fujian, People's Republic of China.
Infect Drug Resist. 2024 Oct 28;17:4717-4726. doi: 10.2147/IDR.S485915. eCollection 2024.
The objective of this study was to identify the risk factors associated with Carbapenem-resistant Enterobacteriaceae (CRE) colonization in intensive care unit (ICU) patients and to develop a predictive risk model for CRE colonization.
In this study, 121 ICU patients from Fujian Provincial Hospital were enrolled between January 2021 and July 2022. Based on bacterial culture results from rectal and throat swabs, patients were categorized into two groups: CRE-colonized (n = 18) and non-CRE-colonized (n = 103). To address class imbalance, Synthetic Minority Over-sampling Technique (SMOTE) was applied. Statistical analyses including T-tests, Chi-square tests, and Mann-Whitney -tests were employed to compare differences between the groups. Feature selection was performed using Lasso regression and Random Forest algorithms. A Logistic regression model was then developed to predict CRE colonization risk, and the results were presented in a nomogram.
After applying SMOTE, the dataset included 198 CRE-colonized patients and 180 non-CRE-colonized patients, ensuring balanced groups. The two groups were comparable in most clinical characteristics except for diabetes, previous emergency department admission, and abdominal infection. Eight independent risk factors for CRE colonization were identified through Random Forest, Lasso regression, and Logistic regression, including Acute Physiology and Chronic Health Evaluation (APACHE) II score > 16, length of hospital stay > 31 days, female gender, previous carbapenem antibiotic exposure, skin infection, multi-site infection, immunosuppressant exposure, and tracheal intubation. The risk prediction model for CRE colonization demonstrated high accuracy (87.83%), recall rate (89.9%), precision (85.6%), and an AUC value of 0.877. Patients were categorized into low-risk (0-90 points), medium-risk (91-160 points), and high-risk (161-381 points) groups, with corresponding CRE colonization rates of 1.82%, 7.14%, and 58.33%, respectively.
This study identified independent risk factors for CRE colonization and developed a predictive model for assessing the risk of CRE colonization.
本研究的目的是确定重症监护病房（ICU）患者中与耐碳青霉烯类肠杆菌科细菌（CRE）定植相关的风险因素，并建立CRE定植的预测风险模型。
在本研究中，纳入了2021年1月至2022年7月期间福建省立医院的121例ICU患者。根据直肠和咽喉拭子的细菌培养结果，将患者分为两组：CRE定植组（n = 18）和非CRE定植组（n = 103）。为了解决类别不平衡问题，应用了合成少数过采样技术（SMOTE）。采用T检验、卡方检验和曼-惠特尼检验等统计分析方法比较两组之间的差异。使用套索回归和随机森林算法进行特征选择。然后建立逻辑回归模型来预测CRE定植风险，并将结果以列线图的形式呈现。
应用SMOTE后，数据集包括198例CRE定植患者和180例非CRE定植患者，确保了组间平衡。除糖尿病、既往急诊科就诊和腹部感染外，两组在大多数临床特征上具有可比性。通过随机森林、套索回归和逻辑回归确定了8个CRE定植的独立风险因素，包括急性生理与慢性健康状况评估（APACHE）II评分> 16、住院时间> 31天、女性、既往碳青霉烯类抗生素暴露、皮肤感染、多部位感染、免疫抑制剂暴露和气管插管。CRE定植的风险预测模型显示出较高的准确性（87.83%）、召回率（89.9%）、精确率（85.6%）和AUC值0.877。患者被分为低风险（0 - 90分）、中风险（91 - 160分）和高风险（161 - 381分）组，相应的CRE定植率分别为1.82%、7.14%和58.33%。
本研究确定了CRE定植的独立风险因素，并建立了一个评估CRE定植风险的预测模型。
