Methods

At NorthShore University HealthSystem, a 4-hospital, 832-bed system based in the Chicago suburbs, hospitalized patients aged ≥18 years receiving empiric antibiotics for suspected infection were identified through the electronic medical record (EMR) between May 2, 2016 and June 30, 2018. Patients were excluded if they had an unclear focus of infection, multiple sources of infection, were receiving pathogen-directed therapy (rather than empiric therapy), or had an infectious disease (ID) consultation. 7 days per week, an ID physician reviewed the patient’s chart within 24 hours of antibiotic initiation. For those patients who were deemed unlikely to have a bacterial infection, a note was left in the EMR recommending antibiotic discontinuation, and the treating team was paged. Most often these patients were believed to have asymptomatic bacteriuria or a viral illness. Patient cases were considered “adherent” to ID recommendations if antibiotics were discontinued within 48 hours of the note and “non-adherent” to recommendations if antibiotics were continued beyond this.

Two investigators (P.M. and N.I.) performed a retrospective chart review, collecting data on demographics, duration of antibiotics, and potential antibiotic-associated ADEs within 30 days following the recommendation to discontinue antibiotics. ADE definitions (Supplement) were derived from the study by Tamma et al., ^{Reference Tamma, Avdic, Li, Dzintars and Cosgrove1} and the adverse event must be known to occur with at least 1 antibiotic used for that patient. Evidence of the ADE had to be documented clearly in nursing notes, progress notes, and/or office visit notes or had to be documented in the results tab for groups with strict lab value changes such as the hematologic, hepatobiliary, renal, cardiac, or musculoskeletal ADEs. Attribution of the adverse event to the antibiotic was decided upon by the investigators based on chart review. Disagreement between investigators on the presence of an ADE prompted a tie-breaker review (J.G. or M.A.). The investigators performing chart review were not blinded to group assignment (adherent vs non-adherent) nor were the tie-breaker reviewers. We compared the incidence of ADEs by individual categories and composite score as well as duration of total antibiotic therapy between the adherent and non-adherent groups. For patients with more than one ADE, only one ADE contributed to the composite score. Continuous variables were compared using Wilcoxon rank sum test and represented as median (interquartile range) if there was a skewed distribution or compared using t test and represented as mean (standard deviation). Indicator variables were compared using χ² test or Fisher's exact test where applicable. Statistical significance of the composite ADE score and duration of antibiotic therapy was calculated using Bonferroni adjusted alpha level of 0.0055 (0.05/9) to account for multiple comparisons.