Methods

This retrospective study was conducted at a 4 hospital, 828 bed health system in the northern suburbs of Chicago. Data were collected through an electronic data warehouse and via chart review from patients age 18 and older who were hospitalized with an uncomplicated bloodstream infection from any source due to Escherichia coli, Klebsiella species, or Proteus species between 1/1/2018–12/31/2021. Uncomplicated GNB was defined as transient, monomicrobial and lacking an abscess or other undrained focus of infection. ^{4,Reference Heil, Bork and Abbo8} Patients with planned durations of therapy exceeding 21 days were excluded, as were patients who enrolled into hospice while on antibiotics for GNB. During the study period, there were two important changes related to patients with GNB. First, beginning 9/1/2019, a policy was enacted to have a mandatory ID consultation for any patient with GNB. Second, as of 6/1/2020, a consensus guideline for GNB management was developed collaboratively among the ID physicians. The consensus guideline included recommendations on antibiotic selection, duration, and transition to oral therapy. These two events separate the study into three periods, preID consultation, transitional period after mandatory ID consultation but before ID consensus, and postID consensus. The primary outcome was days of effective antibiotic therapy. Effective therapy was defined as demonstrating in vitro activity against the pathogen identified in the blood culture. ^{4,Reference Yahav, Franceschini and Koppel5,Reference Molina, Montero-Mateos and Praena-Segovia7,Reference Heil, Bork and Abbo8} For organisms producing extended spectrum beta-lactamase (ESBL) enzymes, piperacillin-tazobactam was not considered effective therapy. The presence of an ESBL was identified genotypically through molecular based detection of a CTX-M gene (cobas® eplex blood culture identification panels, Roche Diagnostics, Indianapolis, Indiana) and/or phenotypically using the Kirby-Bauer disk diffusion method recommended by the Clinical and Laboratory Standards Institute. ⁹

Antibiotic days of therapy were extracted manually by two pharmacists and an ID physician. Secondary outcomes included all-cause hospital mortality, length of hospital stay and 30-day readmission.

Continuous variables were described using median and interquartile range, and categorical data using frequency and prevalence. The Kruskal–Wallis rank sum test was used for continuous variables, and the χ2 test for categorical variables, to compare data between the preID consultation, transition, and postID consensus periods. To assess the impact of the ID consultation and consensus intervention on antibiotic duration, an interrupted time series analysis was conducted. The outcome was aggregated as the monthly median number of antibiotic days, selected to reduce the influence of right-skewness and extreme values. An ordinary least squares regression model was specified with terms for time (T), a postintervention indicator (D), and time since intervention (P). Time (T) was defined as a continuous monthly index across the study period. The postintervention indicator (D) was coded as 0 for all months before implementation of the intervention and 1 for all months thereafter. Time since intervention (P) was defined as the number of months elapsed after the intervention and coded as 0 for all preintervention months. This specification allowed estimation of both an immediate level change and a change in trend following the intervention. Model fit was assessed using the coefficient of determination (R ²) and the overall F-statistic, with statistical significance evaluated at a two-sided α level of 0.05. The analysis was performed using Python.