Analyzing multidrug resistance patterns across the food supply chain using association rule mining.

Abstract

We used the machine learning method association rule mining to analyze multidrug resistance (MDR) among cattle-associated Escherichia coli along the food supply chain in the USA. All datasets were stratified by year, source, and resistance indicator (genotypic/phenotypic). Pruned rulesets were compared by calculating the proportion of rules from a comparison ruleset that are captured in a reference ruleset. Rulesets were compared across years within each source and indicator type to quantify how MDR patterns change over time. At the class level, on average nearly 50 % or more of the MDR patterns remain the same year over year for genotypic and phenotypic indicators. Rulesets were compared between data sources to quantify how MDR patterns change across the food supply chain. These comparisons suggest that there is a greater diversity of MDR patterns present at slaughterhouse settings than at retail settings; and further, that there is a greater diversity of MDR patterns amongst sick cattle on farm settings than at either slaughterhouse or retail settings. Genetic evidence supports this being attributable to a greater genetic diversity associated with pathogenic bacteria vs commensals. Rulesets were compared between indicators to quantify the degree of correspondence between phenotypic and genotypic data. Genotypic rulesets were better able to capture phenotypic rulesets than the reverse. Adding another aminoglycoside (streptomycin) to the phenotypic analysis, improved ruleset correspondence. This asymmetry may be driven by drug specific aminoglycoside resistance genes, suggesting that more drugs need to be assessed to have a fuller understanding of the variation in MDR patterns.