External Validation of Difficult-to-Treat Resistance Prevalence and Mortality Risk in Gram-Negative Bloodstream Infection Using Electronic Health Record Data From 140 US Hospitals

Abstract

Difficult-to-treat resistance (DTR; ie, co-resistance to all first-line antibiotics) in gram-negative bloodstream infection (GNBSI) is associated with decreased survival in administrative data models. We externally validated DTR prevalence and associated mortality risk in GNBSI using detailed clinical data from electronic health records to adjust for baseline differences in acute illness severity.

Keywords: antimicrobial resistance; gram-negative bacteria; mortality.

Figure 1.

A–C, Adjusted relative risk of mortality in patients with gram-negative bloodstream infection in the Cerner Healthfacts Database by resistance phenotype. The risk of mortality was adjusted (aRR) for age, sex, Elixhauser comorbidity index, gram-negative bloodstream infection (GNBSI) source, taxon, hospital vs community onset, year, and hospital geographic region, bed capacity, and urban and teaching status, as well as (with and without) baseline Sequential Organ Failure Assessment (SOFA) score using a modified Poisson regression approach with robust error variance. The estimate of aRR of mortality is represented by dots, and corresponding 95% confidence intervals (CIs) are represented by vertical lines. aRR estimates and 95% CIs from the analysis with and without baseline SOFA score are displayed in red and blue, respectively. Subfigure (A) compares the aRR across difficult-to-treat resistance (DTR), carbapenems (CR), extended-spectrum cephalosporins (ECR), and fluoroquinolones (FQR) phenotypes in patients with GNBSI. Here, susceptible phenotype (ie, absence of DTR, CR, ECR, and FQR phenotypes) was used as the reference category. Similarly, subfigure (B) compares the aRR associated with DTR relative to the other 3 resistance phenotypes. Here, we individually alternated the reference category as CR, ECR, and FQR (where ≥1 first-line agent category was active) to determine the aRR relative to DTR. Subfigure (C) demonstrates the variation in the relationship between the aRR of mortality associated with DTR vs CR across bloodstream isolate taxa. Here, the reference category was CR where ≥1 first-line agent category was active.