Modeling Carbapenem Resistance Among Gram-Negative Pathogens Using Commonly Available Clinical Data

Abstract

Newer technologies such as matrix-assisted laser desorption ionization-time of flight mass spectrometry (MALDI-TOF) are providing clinicians rapid identification of pathogen species. However, this technology cannot provide antibiotic susceptibility. Infection with carbapenem resistance (CR) among Gram-negative rods (GNRs) species is now extremely common in hospitalized patients and associated with higher mortality. We sought to create a model to predict high-risk CR among GNR pathogens. We performed an analysis of all patients admitted from 2006 to 2016 to two-medical centers, one academic and one community, that shared a common clinical microbiology laboratory. Routine susceptibility testing was conducted with gold standard broth micro-dilution. We examined all admissions with positive cultures of Klebsiella spp. and Pseudomonas aeruginosa and correlated CR with commonly collected data from the medical records such as demographics, diagnoses, and clinical data. We created bivariate and multivaritate logistic models to predict CR. Data were analyzed for complete cases only and were analyzed once per hospital admission. CR was seen in 5.7% (232/4,043) of admissions where Klebsiella was identified and 29.7% (1,102/3,709) for Pseudomonas. Independent predictors for CR among Klebsiella included male gender, younger age, skilled nursing residency, and more co-morbidities. Independent predictors for CR among Pseudomonas included male gender, younger age, longer duration of hospitalization, skilled nursing residency, more co-morbidities, renal failure, cystic fibrosis, and lack of solid organ transplant. Both models had very good predictive ability (AUC of 0.76 and 0.73, respectively) We developed a simple algorithm to identify select GNR pathogens associated with a high-risk of CR in our medical system that had good predictive ability. Among Klebsiella CR is more associated with factors present on admission, where as CR among Pseudomonas is associated with longer hospitalization and hospital associated factors. Findings from this model can be tested in real-time situations and validated in other medical care systems. L. G. Miller, Sage Products: Receipt of contributed product, Conducting studies in healthcare facilities that are receiving contributed product. Xttrium: Receipt of contributed product, Conducting studies in healthcare facilities that are receiving contributed product. Clorox: Receipt of contributed product, Conducting studies in healthcare facilities that are receiving contributed product. 3M: Receipt of contributed product, Conducting studies in healthcare facilities that are receiving contributed product.