Comparing Patient Risk Factor-, Sequence Type-, and Resistance Locus Identification-Based Approaches for Predicting Antibiotic Resistance in Escherichia coli Bloodstream Infections.

Derek R Macfadden,Nathalie Tijet,William P Hanage,Bryan Coburn,Roberto G Melano,Nick Daneman,Nathan A Ledeboer

doi:10.1128/jcm.01780-18

Abstract

Rapid diagnostic tests for antibiotic resistance that identify the presence or absence of antibiotic resistance genes/loci are increasingly being developed. However, these approaches usually neglect other sources of predictive information which could be identified over shorter time periods, including patient epidemiologic risk factors for antibiotic resistance and markers of lineage. Using a data set of 414 Escherichia coli isolates recovered from separate episodes of bacteremia at a single academic institution in Toronto, Ontario, Canada, between 2010 and 2015, we compared the potential predictive ability of three approaches (epidemiologic risk factor-, pathogen sequence type [ST]-, and resistance gene identification-based approaches) for classifying phenotypic resistance to three antibiotics representing classes of broad-spectrum antimicrobial therapy (ceftriaxone [a 3rd-generation cephalosporin], ciprofloxacin [a fluoroquinolone], and gentamicin [an aminoglycoside]). We used logistic regression models to generate model receiver operating characteristic (ROC) curves. Predictive discrimination was measured using apparent and corrected (bootstrapped) areas under the curves (AUCs). Epidemiologic risk factor-based models based on two simple risk factors (prior antibiotic exposure and recent prior susceptibility of Gram-negative bacteria) provided a modest predictive discrimination, with AUCs ranging from 0.65 to 0.74. Sequence type-based models demonstrated strong discrimination (AUCs, 0.83 to 0.94) across all three antibiotic classes. The addition of epidemiologic risk factors to sequence type significantly improved the ability to predict resistance for all antibiotics (P < 0.05). Resistance gene identification-based approaches provided the highest degree of discrimination (AUCs, 0.88 to 0.99), with no statistically significant benefit being achieved by adding the patient epidemiologic predictors. In summary, sequence type or other lineage-based approaches could produce an excellent discrimination of antibiotic resistance and may be improved by incorporating readily available patient epidemiologic predictors but are less discriminatory than identification of the presence of known resistance loci.

Highlights

Rapid diagnostic tests for antibiotic resistance that identify the presence or absence of antibiotic resistance genes/loci are increasingly being developed
Reducing the sequencing and bioinformatics time to such a short duration with minimal computational requirements means that affordable and portable sequencebased approaches could be employed to impact initial empirical therapy in the clinic or hospital. The success of this technique relies on the association between the lineage and the antibiotic susceptibility profile, which remains to be demonstrated across scale for many pathogens, including common Gram-negative bacteria, such as Escherichia coli
We found that the sequence type-based approaches provided an excellent ability to predict antibiotic resistance in E. coli

Summary

Introduction

Rapid diagnostic tests for antibiotic resistance that identify the presence or absence of antibiotic resistance genes/loci are increasingly being developed. Capable of early identification of infecting pathogens and their antibiotic resistance profiles, are increasingly being developed [7, 8] These tests offer the promise of reducing unnecessary antibiotic use, as well as improving the time to early adequate therapy. One option which combines both speed and genomic analysis is the lineage-calling method This method is somewhat analogous to multilocus sequence typing (MLST), where a pathogen can be matched (using k-mer-based approaches) to a database of known lineages and susceptibility patterns in order to predict antibiotic resistance/ susceptibility. This novel approach of identifying resistance-associated sequence elements (RASE) requires considerably less genomic processing/bioinformatics and can identify a probable phenotype within 1 to 5 min after the start of sequencing [14]. The success of this technique relies on the association between the lineage (e.g., the sequence type [ST] or phylogroup) and the antibiotic susceptibility profile, which remains to be demonstrated across scale for many pathogens, including common Gram-negative bacteria, such as Escherichia coli

Methods

Results

Conclusion