A two-stage diagnostic model for discriminating and assessing risk of meropenem heteroresistance in Pseudomonas aeruginosa

BACKGROUND: Heteroresistance (HR) in Pseudomonas aeruginosa causes misclassification as 'susceptible (S)' on routine antibiotic susceptibility techniques, potentially contributing to subsequent treatment failure. This study aimed to explore clinically relevant risk factors for HR compared with S (S-HR), and resistant (R) compared with HR (HR-R) phenotypes. Additionally, we explored whether integrating medical history and laboratory data can enable rapid and accurate identification of HR and R phenotypes in this pathogen. METHODS: This retrospective study included 420 P. aeruginosa strains collected from 2011 to 2024 in China. The strains were categorized into three groups according to their sensitivity to meropenem: non-heteroresistant susceptible, heteroresistant and non-heteroresistant resistant. Logistic regression, random forest and XGBoost models were constructed using variables identified through LASSO (least absolute shrinkage and selection operator) regression. The models' performance was evaluated via 10-fold cross-validation comparing area under the receiver operating characteristic curve (AUROC), sensitivity and specificity. RESULTS: Multivariate analyses identified central venous catheters as an independent risk factor for S-HR, and malignant solid tumours, pulmonary infections, mechanical ventilation and carbapenem use for HR-R. A discriminating diagnostic model, combining clinical and laboratory data, showed an AUROC of 0.919 for HR-R and 0.856 for S-HR. The calibration plots indicated good alignment between the estimated and observed probabilities. CONCLUSIONS: This study presents a validated two-stage risk assessment model to discriminate the two phases of meropenem heteroresistance in P. aeruginosa. By identifying novel stage-specific risk factors and delivering a practical tool compatible with clinical workflows, this model facilitates the early identification and targeted intervention of HR, offering novel insights into the mechanistic dissection of HR.