# Identification and validation of a risk assessment scoring tool for extended-spectrum beta-lactamase-producing Enterobacterales bacteremia at a tertiary teaching hospital

## TL;DR

This study developed a scoring tool to identify hospitalized patients likely to have ESBL-producing bloodstream infections based on risk factors like diabetes and antibiotic history.

## Contribution

The novel contribution is a validated risk assessment tool for ESBL bacteremia tailored to a specific hospital setting.

## Key findings

- Diabetes, recent invasive procedures, antibiotic use, and prior ESBL history were independent predictors of ESBL.
- The risk assessment tool showed strong predictive power with an ROC AUC of 0.88 in the validation cohort.
- The model demonstrated good calibration and discrimination in identifying ESBL cases.

## Abstract

To identify institution-specific risk factors for extended-spectrum beta-lactamase (ESBL) bloodstream infections (BSI) to develop and validate a risk assessment scoring tool that can be utilized for hospitalized patients.

Single-center, retrospective, case-control study.

Tertiary teaching hospital.

Hospitalized adult and pediatric patients with E. coli or Klebsiella spp. BSI were stratified based on ESBL production between August 2019 to July 2021. Exclusion criteria included patients < 28 days old, a positive blood culture resulting prior to admission/after discharge or a polymicrobial and/or carbapenem-resistant BSI.

Multivariable logistic regression assessed predictors of ESBL in a derivation cohort. Predictors were applied to a novel validation BSI cohort using area under the receiver-operator characteristics curve (ROC AUC) to assess the reliability of identifying patients likely to harbor ESBL at the time of organism identification.

A total of 238 patients in the derivation cohort met inclusion criteria stratified as ESBL (n = 68) or non-ESBL (n = 170). Multivariable logistic regression demonstrated diabetes, 30-day history of invasive procedure or antibiotic use, and/or history of ESBL as independent predictors of ESBL. After creation of an ESBL risk assessment tool, the results were applied to a validation cohort of 170 patients. This model displayed good calibration and discrimination with a strong predictive power (Hosmer-Lemeshow χ2= 4.66, p = 0.19; ROC AUC = 0.88, 95% CI = 0.7909 – 0.974).

A validated ESBL risk assessment tool reliably identified hospitalized patients likely to harbor ESBL E. coli or Klebsiella spp. BSI upon organism identification.