Predicting Antimicrobial Resistance Prevalence and Incidence from Indicators of Antimicrobial Use: What Is the Most Accurate Indicator for Surveillance in Intensive Care Units?

Élise Fortin,Caroline Quach,David L Buckeridge,Patricia S Fontela,Robert W Platt,Patrick Butaye

doi:10.1371/journal.pone.0145088

Abstract

ObjectiveThe optimal way to measure antimicrobial use in hospital populations, as a complement to surveillance of resistance is still unclear. Using respiratory isolates and antimicrobial prescriptions of nine intensive care units (ICUs), this study aimed to identify the indicator of antimicrobial use that predicted prevalence and incidence rates of resistance with the best accuracy.MethodsRetrospective cohort study including all patients admitted to three neonatal (NICU), two pediatric (PICU) and four adult ICUs between April 2006 and March 2010. Ten different resistance / antimicrobial use combinations were studied. After adjustment for ICU type, indicators of antimicrobial use were successively tested in regression models, to predict resistance prevalence and incidence rates, per 4-week time period, per ICU. Binomial regression and Poisson regression were used to model prevalence and incidence rates, respectively. Multiplicative and additive models were tested, as well as no time lag and a one 4-week-period time lag. For each model, the mean absolute error (MAE) in prediction of resistance was computed. The most accurate indicator was compared to other indicators using t-tests.ResultsResults for all indicators were equivalent, except for 1/20 scenarios studied. In this scenario, where prevalence of carbapenem-resistant Pseudomonas sp. was predicted with carbapenem use, recommended daily doses per 100 admissions were less accurate than courses per 100 patient-days (p = 0.0006).ConclusionsA single best indicator to predict antimicrobial resistance might not exist. Feasibility considerations such as ease of computation or potential external comparisons could be decisive in the choice of an indicator for surveillance of healthcare antimicrobial use.

Highlights

The causal relationship between antimicrobial use and antimicrobial resistance is difficult to quantify due to the various settings and measures studied and to related biases, this relationship is generally accepted.[1,2,3] The European Surveillance of Antimicrobial Comsumption (ESAC) has shown that countries using antimicrobials more intensively tend to present higher levels of resistance.[4]
After adjustment for intensive care unit (ICU) type, indicators of antimicrobial use were successively tested in regression models, to predict resistance prevalence and incidence rates, per 4-week time period, per ICU
Carbapenem use, recommended daily doses per 100 admissions were less accurate than courses per 100 patient-days (p = 0.0006)

Summary

Introduction

The causal relationship between antimicrobial use and antimicrobial resistance is difficult to quantify due to the various settings and measures studied and to related biases, this relationship is generally accepted.[1,2,3] The European Surveillance of Antimicrobial Comsumption (ESAC) has shown that countries using antimicrobials more intensively tend to present higher levels of resistance.[4] Considering that antimicrobial use is modifiable, surveillance of antimicrobial use is often recommended as a complement to surveillance of antimicrobial resistance in hospitals.[5,6,7,8]. Many authors have exposed either the limitations of different indicators, their own choice of indicator, or the necessity for more research to identify the most appropriate indicator(s) for surveillance of antimicrobial use, very few published studies have compared these indicators’ ability to predict resistance levels in hospital settings.[1, 11, 15,16,17,18]

Methods

Results

Conclusion