OBJECTIVE  To develop a machine learning model for predicting hospital-acquired colonization by Carbapenem-resistant Enterobacteriaceae （CRE） in intensive care unit （ICU）, thereby providing a reference for the prevention and control of hospital-acquired CRE infections.

METHODS  We enrolled a total of 552 ICU patients screened for CRE at Wenzhou Integrated Traditional Chinese and Western Medicine Hospital from Jan. 2020 to Aug. 2025. Among them, patients with CRE screening transitioning from negative （≥1 time） to positive （≥1 time） were assigned to a positive group （n=71）, while patients with persistently negative CRE screening （≥3 times） were assigned to a control group （n=57）. The dataset was randomly divided into training and testing sets at a 3∶1 ratio. Recursive feature elimination was employed to screen 25 predictors from the training set, and six machine learning models were developed: logistic regression analysis, support vector machine （SVM）, artificial neural network （ANN）, extreme gradient boosting （XGBoost）, decision tree and random forest. Model performance was evaluated by metrics such as the area under the curve （AUC） to ultimately determine the optimal model.

RESULTS  After evaluation, the random forest model performed best （AUC of 0.950 for the training set and 0.826 for the testing set, respectively）. The identified predictors included age, length of stay in ICU, albumin level, length of combined antimicrobial use, mechanical ventilation, length of carbapenem antimicrobial use and total days of antimicrobial use. An online calculation tool was developed based on the random forest model.

CONCLUSIONS  The random forest model demonstrated superior performance. The optimal model and online tool can aid in early, timely and accurate identification of high-risk populations for hospital-acquired colonization by CRE, enabling prompt contact isolation.