Modeling Likelihood of Coverage for Narrow Spectrum Antibiotics in Patients Hospitalized with Urinary Tract Infections

BACKGROUND: When prescribing empiric antibiotics, providers try to choose the narrowest spectrum antibiotic that will cover a patient’s infection. To do this they must assess the likelihood of coverage of different regimens. We developed a model for cefazolin (or cephalexin) coverage for patients admitted to the hospital with urinary tract infections (UTI), to identify agroup of patients with a high likelihood of coverage by this first-line, narrow spectrum antibiotic. We also compared cefazolin coverage to the coverage of patients’ actual empiric treatment regimens. METHODS: Patients admitted from 11/1/11 to 1/1/14 with a positive urine culture in the First 48 hours and a discharge diagnosis of UTI, were included in the dataset. Data extracted from our information warehouse included empiric antibiotic administration data, demographics, comorbidities, and past antibiotic use. Only the first eligible admission for each patient was included. A 20% random sample of patients was selected as the validation set. Logistic regression models estimated the predicted probability of cefazolin coverage. RESULTS: A total of 3,456 patients with an eligible UTI were included. Six hundred and Ninety-one (691) were held out for validation. Cefazolin covered 49% of the UTIs. The final model had an area under the receiver operating curve (AUC) of 69% (95% CI: 67%, 71%) in the test and 70%, (66%, 74%) in the validation set. Overall 49/65 (75%) in the highest estimated decile of cefazolin coverage had a UTI that would have been covered; only 13/66 (20%) in the lowest decile would have been covered. Of the patients in the highest decile of cefazolin coverage, 48/65 (74%) were covered by the actual empiric regimen given, however 35/65 (54%) of those regimens consisted of multiple antibiotics, and of those patients who would have been covered by cefazolin, 36/49 (73%) were empirically treated with broader spectrum antibiotics. CONCLUSION: Our findings suggest that the model can reasonably identify patients whose infections would be likely to be covered by cefazolin. Further, the majority of patients would have been covered by a narrower spectrum antibiotics than what they received. Research reported in this publication was supported by the National Institute of Allergy and Infectious Diseases of the NIH under Award Number R01AI116975. DISCLOSURES: All authors: No reported disclosures.