Hospital-onset bacteremia and fungemia: An evaluation of predictors and feasibility of benchmarking comparing two risk-adjusted models among 267 hospitals

• Type of Material: Article
• Authors: Kalvin C Yu, Becton, Dickinson and Company, Franklin Lakes, New Jersey; Gang Ye, Becton, Dickinson and Company, Franklin Lakes, New Jersey; Jonathan R Edwards, Centers for Disease Control and Prevention, Atlanta; Vikas Gupta, Emory University; Andrea Benin, Emory University; ChinEn Ai, Becton, Dickinson and Company, Franklin Lakes, New Jersey; Raymund Dantes, Emory University
• Language: English
• Date: 2022-09-09
• Publisher: CAMBRIDGE UNIV PRESS
• Publication Version: Final Published Version
• Copyright Statement: © The Author(s) 2022
• License: Creative Commons Attribution 4.0 International
• Final Published Version (URL): http://dx.doi.org/10.1017/ice.2022.211
• Title of Journal or Parent Work: INFECTION CONTROL AND HOSPITAL EPIDEMIOLOGY
• Volume: 43
• Issue: 10
• Start Page: 1317
• End Page: 1325
• Grant/Funding Information: Financial support for the manuscript was provided by Becton, Dickinson & Company. Support for the data analysis was provided by the CDC. The findings and conclusions in this report are those of the authors and do not necessarily represent the views of the CDC.
• Keywords: Life Sciences & Biomedicine; Infectious Diseases; BLOOD-STREAM INFECTIONS; Public, Environmental & Occupational Health; Science & Technology

Abstract

Objectives: To evaluate the prevalence of hospital-onset bacteremia and fungemia (HOB), identify hospital-level predictors, and to evaluate the feasibility of an HOB metric. Methods: We analyzed 9,202,650 admissions from 267 hospitals during 2015-2020. An HOB event was defined as the first positive blood-culture pathogen on day 3 of admission or later. We used the generalized linear model method via negative binomial regression to identify variables and risk markers for HOB. Standardized infection ratios (SIRs) were calculated based on 2 risk-adjusted models: a simple model using descriptive variables and a complex model using descriptive variables plus additional measures of blood-culture testing practices. Performance of each model was compared against the unadjusted rate of HOB. Results: Overall median rate of HOB per 100 admissions was 0.124 (interquartile range, 0.00-0.22). Facility-level predictors included bed size, sex, ICU admissions, community-onset (CO) blood culture testing intensity, and hospital-onset (HO) testing intensity, and prevalence (all P <.001). In the complex model, CO bacteremia prevalence, HO testing intensity, and HO testing prevalence were the predictors most associated with HOB. The complex model demonstrated better model performance; 55% of hospitals that ranked in the highest quartile based on their raw rate shifted to a lower quartile when the SIR from the complex model was applied. Conclusions: Hospital descriptors, aggregate patient characteristics, community bacteremia and/or fungemia burden, and clinical blood-culture testing practices influence rates of HOB. Benchmarking an HOB metric is feasible and should endeavor to include both facility and clinical variables.