An empirical comparison of isolate-based and sample-based definitions of antimicrobial resistance and their effect on estimates of prevalence.

Antimicrobial resistance is primarily a problem in human medicine but there are unquantified links of transmission in both directions between animal and human populations. Quantitative assessment of the costs and benefits of reduced antimicrobial usage in livestock requires robust quantification of transmission of resistance between animals, the environment and the human population. This in turn requires appropriate measurement of resistance. To tackle this we selected two different methods for determining whether a sample is resistant - one based on screening a sample, the other on testing individual isolates. Our overall objective was to explore the differences arising from choice of measurement. A literature search demonstrated the widespread use of testing of individual isolates. The first aim of this study was to compare, quantitatively, sample level and isolate level screening. Cattle or sheep faecal samples (n=41) submitted for routine parasitology were tested for antimicrobial resistance in two ways: (1) "streak" direct culture onto plates containing the antimicrobial of interest; (2) determination of minimum inhibitory concentration (MIC) of 8-10 isolates per sample compared to published MIC thresholds. Two antibiotics (ampicillin and nalidixic acid) were tested. With ampicillin, direct culture resulted in more than double the number of resistant samples than the MIC method based on eight individual isolates. The second aim of this study was to demonstrate the utility of the observed relationship between these two measures of antimicrobial resistance to re-estimate the prevalence of antimicrobial resistance from a previous study, in which we had used "streak" cultures. Boot-strap methods were used to estimate the proportion of samples that would have tested resistant in the historic study, had we used the isolate-based MIC method instead. Our boot-strap results indicate that our estimates of prevalence of antimicrobial resistance would have been considerably lower in the historic study had the MIC method been used. Finally we conclude that there is no single way of defining a sample as resistant to an antimicrobial agent. The method used greatly affects the estimated prevalence of antimicrobial resistance in a sampled population of animals, thus potentially resulting in misleading results. Comparing methods on the same samples allows us to re-estimate the prevalence from other studies, had other methods for determining resistance been used. The results of this study highlight the importance of establishing what the most appropriate measure of antimicrobial resistance is, for the proposed purpose of the results.