Development and application of statistical methodology for analysis of the phenomenon of multi-drug resistance in the EU: demonstration of analytical approaches using antimicrobial resistance isolate-based data

Abstract

Since antimicrobial resistance (AMR) has been one of the major public health burdens over the last decade, it is of great importance to appropriately monitor and analyse AMR data. Isolate-based data within the EU have been routinely collected since 2010 and reported to EFSA on a yearly basis. AMR data are collected for several bacterial species, tested for susceptibility against different antimicrobials and minimum inhibitory concentration (MIC) is reported. For analysis purposes, a dichotomised version of the MIC values based on the epidemiological cut-off is used to represent different resistance patterns. This report describes various methods to analyse multi-drug resistance data, including the identification of structure to construct groups of isolates with similar resistance patterns or with similar MIC values. Multivariate classification trees and hierarchical cluster analysis after application of principal components and multiple correspondence analyses are applied aiming at group discovering. Latent class analysis is presented as an alternative model-based approach. The generalised estimating equations method is presented handling univariate and multivariate binary outcomes. Bayesian network analysis provides the user with a graphical representation of the underlying associations in the data to identify new co-resistance patterns. Models that deal with spatial distribution of resistant isolates, in combination with their evolution over time, are constructed for univariate and bivariate outcomes. Finally, pattern and source attributions tools are presented, providing, in addition to exploratory analyses, a logistic model to assess variables influencing certain resistance patterns. Source attribution is used to attribute resistance cases in humans to resistance observed in animal, human food consumption patterns and antimicrobial usage data. For illustration purposes, these methods are applied to a subset of the AMR data using an application developed with the R package “shiny”.