A Clinical Prediction Rule for Fluoroquinolone Resistance in Healthcare-acquired Urinary Tract Infections (UTIs)

Background: The prevalence of fluoroquinolone (FQ) resistance among gram-negative uropathogens has been increasing. Development of a clinical prediction score to identify patients most likely to be infected with a FQ resistant gram-negative uropathogen would be useful for optimizing appropriate empiric antibiotic therapy.

Objective: To derive and validate a clinical prediction score for predicting FQ resistance in healthcare-acquired gram-negative bacilli urinary tract infections (GNB-UTIs). 

Methods: To derive the prediction rule, we used a case-control study of risk factors for FQ resistance in healthcare-acquired GNB-UTIs among patients hospitalized at 2 medical centers in Philadelphia. A clinical prediction score was developed by simplifying coefficients of each independent significant risk factor associated with FQ resistance. To validate the prediction rule, a hypothetical validation cohort was derived from the case-control population by random sampling of cases and controls based on a frequency weighting scheme. We subsequently evaluated sensitivity, specificity and area under the receive operating characteristic (ROC) curve in the validation cohort. Finally, we performed “overoptimisim” adjustment for the prediction error.  

Results:  The clinical prediction score is a summation of scores from the presence of the following predictor variables: 1) score = 1 point (male gender, inpatient metronidazole exposure in the preceding 30 days); 2) score = 2 points (residence for a long term care facility, medicine service, chronic renal insufficiency, chronic respiratory disease, hydronephrosis, and inpatient cotrimoxazole exposure in the preceding 30 days); 3) score = 5 points (inpatient FQ exposure in the preceding 30 days). At the cutoff score of ≥ 2, the clinical prediction score demonstrated 75% sensitivity and 73% specificity to identify FQ-resistant UTI.  The area under ROC curve was 0.816 and the over-optimism adjusted area under ROC curve was 0.815. The ROC curve, sensitivity and specificity for each cut-off value are shown in figure 1.

Conclusions: This clinical prediction score has an acceptable sensitivity and specificity to predict FQ resistance in patients with healthcare-acquired GNB-UTI.  External validation of this score is necessary before testing the implementation of this score on optimizing empiric antibiotic use in clinical practice.  

Figure 1. (a) ROC curve and (b) Sensitivity and specificity for each cut-off value of clinical prediction score