Abstract
<title>Abstract</title> <p>Objective This study aimed to develop and compare five machine learning models utilizing the MIMIC-IV database to identify key risk factors for multidrug-resistant organism (MDRO) infection in patients with invasive procedures. Methods A cohort of 23 281 patients was extracted from MIMIC-IV. Demographic, physiological, laboratory, and clinical variables were collected. The weighted calibration method was applied to address class imbalance. The dataset was randomly divided into training and testing subsets at an 7:3 ratio. Feature selection was performed using LASSO regression and feature importance scoring. Predictive models were constructed with Logistic Regression (LR), Random Forest (RF), Gradient Boosting Machine (GBM), Extreme Gradient Boosting (XGB), and Light Gradient Boosting Machine (LGB). Model performance was assessed via receiver operating characteristic (ROC) analysis. Calibration was evaluated using calibration curves and Brier scores, while clinical utility was examined through decision curve analysis (DCA). Model interpretability was enhanced using Shapley Additive Explanation (SHAP) values. Results Of the 23 281 patients, 1956 had invasive procedures associated MDRO infection and 21 325 had no invasive procedures associated MDRO infection. Significant differences were observed across multiple clinical parameters. XGB demonstrated the highest predictive performance, incorporating 15 features and achieving an AUC of 0.751 (95% CI: 0.728–0.773) on the test set. Calibration curves indicated good fit, supported by DCA confirming clinical usefulness. SHAP analysis highlighted the top influential features: the length of ICU stay, total antibiotic using hours, have CVC inserted, APS Ⅲ, age at ICU. Conclusion XGB emerged as the optimal predictor, which exhibited the strongest predictive performance for MDRO infections among patients undergoing invasive procedures, thereby equipping clinicians with a tool for early risk stratification and personalized treatment strategies</p>