Predictive value of extra-rectal colonization and the Giannella risk score for subsequent carbapenem-resistant Enterobacterales infections in ICU patients: an Argentinean retrospective cohort study in a mixed KPC and NDM setting

Background The increasing incidence of carbapenem-resistant Enterobacterales (CRE) infections, driven by carbapenemase-producing Klebsiella pneumoniae (KPC) and New Delhi metallo-β-lactamase (NDM) dissemination, demands reliable predictive tools to optimize empiric therapy. The Giannella risk score (GRS), initially developed in a CRE-KPC-dominant setting, may not perform equally in heterogeneous contexts. This study aimed to evaluate the predictive value of extra-rectal CRE colonization and the GRS for overall CRE infections, specific CRE-KPC and CRE-NDM infections, and bacteraemia in patients rectally colonized with CRE.Methods We conducted a retrospective cohort study of adults with rectal CRE-KPC and/or CRE-NDM colonization admitted to an intensive care unit in Argentina during a Providencia stuartii-NDM outbreak in a setting with endemic CRE-KPC circulation. Patients were followed up for up to 90 days. Associations were analysed using bivariate statistics, and logistic regression and predictive performance were evaluated through receiver operating characteristic (ROC) curves.Results Among 327 CRE rectal-colonized patients, 49.2% carried CRE-KPC, 30.3% CRE-NDM, and 20.5% were co-colonized. Overall, 10.7% developed CRE infections. Extra-rectal colonization was independently associated with infection (adjusted odds ratio: 3.5 [1.53-7.94], P ≤ 0.01), with high specificity (93.9%) but low sensitivity (20.0%). The GRS showed moderate predictive performance for overall infection (area under the ROC curve, 0.66) with a threshold of ≥3 points. However, when applied specifically to CRE-KPC- or CRE-NDM-colonized patients, neither extra-rectal cultures nor GRS calculations reliably predicted infection or bacteraemia.Conclusions In this cohort, extra-rectal CRE colonization was the only independent predictor of subsequent infection. These findings highlight the need for carbapenemase-independent dynamic risk models better suited to mixed resistance environments.