Utility of a clinical risk factor scoring model in predicting infection with extended-spectrum β-lactamase-producing enterobacteriaceae on hospital admission.

OBJECTIVE: To validate the utility of a previously published scoring model (Italian) to identify patients infected with community-onset extended-spectrum β-lactamase-producing Enterobacteriaceae (ESBL-EKP) and develop a new model (Duke) based on local epidemiology. METHODS: This case-control study included patients 18 years of age or more admitted to Duke University Hospital between January 1, 2008, and December 31, 2010, with culture-confirmed infection due to an ESBL-EKP (cases). Uninfected controls were matched to cases (3:1). The Italian model was applied to our patient population for validation. The Duke model was developed through logistic-regression-based prediction scores calculated on variables independently associated with ESBL-EKP isolation. Sensitivities and specificities at various point cutoffs were determined, and determination of the area under the receiver operating characteristic curve (ROC AUC) was performed. RESULTS: A total of 123 cases and 375 controls were identified. Adjusted odds ratios and 95% confidence intervals for variables previously identified in the Italian model were as follows: hospitalization (3.20 [1.62-6.55]), transfer (4.31 [2.15-8.78]), urinary catheterization (5.92 [3.09-11.60]), β-lactam and/or fluoroquinolone therapy (3.76 [2.06-6.95]), age 70 years or more (1.55 [0.79-3.01]), and Charlson Comorbidity Score of 4 or above (1.06 [0.55-2.01]). Sensitivity and specificity were, respectively, more than or equal to 95% and less than or equal to 47% for scores 3 or below and were less than or equal to 50% and more than or equal to 96% for scores 8 or above. The ROC AUC was 0.88. Variables identified in the Duke model were as follows: hospitalization (2.63 [1.32-5.41]), transfer (5.30 [2.67-10.71]), urinary catheterization (6.89 [3.62-13.38]), β-lactam and/or fluoroquinolone therapy (3.47 [1.91-6.41]), and immunosuppression (2.34 [1.14-4.80]). Sensitivity and specificity were, respectively, more than or equal to 94% and less than or equal to 65% for scores 3 or below and were less than or equal to 58% and more than or equal to 95% for scores 8 or above. The ROC AUC was 0.89. CONCLUSION: While the previously reported model was an excellent predictor of community-onset ESBL-EKP infection, models utilizing factors based on local epidemiology may be associated with improved performance.

Duke Scholars

Author Deverick John Anderson Medicine, Infectious Diseases

Author Richard Howard Drew Medicine, Infectious Diseases