Methods

We used a patient-level matched cohort design with uninfected patients as the control group.Reference Kaye, Engemann, Mozaffari and Carmeli³ Case patients were all veterans admitted to acute-care units at VHA hospitals in 48 continental states and the District of Columbia in the United States during January 2003–December 2013 and who had first positive blood cultures for E. coli or Klebsiella spp at least 48 hours after admission. If a patient had multiple positive blood cultures for the same species during the same hospital admission, we included only the first isolate in the analysis. We selected uninfected controls who were matched at the patient-level based on sex, month of admission, and facility. Also, length of stay of uninfected controls was matched to the length of stay before onset of bacteremia of case patients.Reference Harris, Karchmer, Carmeli and Samore⁴ For each case, up to 3 uninfected controls were selected.

We obtained data from the Corporate Data Warehouse (CDW) through the Veterans’ Affairs Informatics and Computing Infrastructure (VINCI), which includes data extracted from VHA’s integrated electronic medical record system. Patient characteristics were compared between cases and controls using the Fisher exact test for categorical variables and the Mann-Whitney U test for continuous variables. Susceptibility results in microbiology reports were recorded in a standardized manner, and isolates were classified as nonsusceptible if they were reported as intermediately resistant or resistant. Minimal inhibitory concentrations (MICs) and sizes of inhibition zones were not typically available.

Our primary outcome measure was all-cause 30-day mortality after the first positive blood culture. Dates of death were obtained from the VHA Vital Status File, which combines mortality information from multiple VHA and non-VHA sources and tracks patients even if they leave the VHA system.

Comorbidities were assessed using diagnosis codes based on the Charlson method with administrative data.Reference Deyo, Cherkin and Ciol⁵ Hospital-onset bacteremia was defined as the first positive blood culture taken after the patient had been in the hospital for 48 hours or longer. Fluoroquinolone resistance was defined as a nonsusceptible result to at least 1 of the fluoroquinolones: ciprofloxacin, levofloxacin, or moxifloxacin. ESCR was defined as a nonsusceptible result to at least 1 the following: ceftriaxone, ceftazidime, or cefepime. The attributable fraction of mortality was defined as the proportional reduction in mortality that would occur if risk factors were removed, and could be interpreted as relative impact on mortality.

To examine the association between each antibiotic resistance pattern and 30-day mortality, and to estimate their relative risks directly from the models, we used the modified Poisson regression analysis for binary outcomes proposed by Zou.Reference Zou⁶ Crude and adjusted attributable fractions of mortality were calculated with Levin’s formulaReference Levin⁷ for each antibiotic resistance group by comparing the group to uninfected controls, considering age, body mass index (BMI), and comorbidity index as patient-level confounders, based on separate models for each matched comparison group. Variables for risk-adjustment models were selected manually based on the logistic regression model to predict 30-day mortality for all groups combined, with comparisons of models for each group, and all models were adjusted for the same confounders. All statistical analyses were conducted using SAS version 9.4 software (SAS Institute, Cary, NC). The institutional review board of the University of Iowa and Research & Development Committee of Iowa City Veterans Affairs Health Care System approved this study. A waiver for informed consent was granted by the institutional review board for this retrospective cohort.