The use of randomization tests to assess the degree of similarity in PFGE patterns of E. coli O157 isolates from known outbreaks and statistical space–time clusters

Abstract

Using isolates from reported cases of Escherichia coli O157 from Alberta, Canada in 2002, we applied randomization tests to determine if cases associated with an outbreak or statistical space-time cluster had more similar pulsed-field gel electrophoresis patterns, based on Dice coefficients, than expected by chance alone. Within each outbreak and space-time cluster, we assessed the mean, median, 25th percentile, 75th percentile, standard deviation, coefficient of variation, and interquartile range of the Dice coefficients of each pairwise comparison among the isolates. To assess the statistical significance of measures of location (e.g. mean) and variation (e.g. standard deviation) we created randomization distributions using all isolates or only isolates from sporadic cases. We determined that randomization tests are an appropriate tool for evaluating the similarity among isolates from cases that have been linked epidemiologically or statistically. We found little difference between using all cases or only sporadic cases when creating our randomization distributions.

Combining analytical epidemiology and genomic surveillance to identify risk factors associated with the spread of antimicrobial resistance in Salmonella enterica subsp. enterica serovar Heidelberg

“…The SID was computed for each clustering threshold by changing the number of different cgMLST loci ( k ) permitted in the same cluster, starting with no allowable differences (i.e., k =0) considered for inclusion into the same cluster up to a maximum of k =35, where all genomes in the sample formed a single cluster. To evaluate whether cgMLST clusters were generating genetically distinct groups, we employed a Monte Carlo sampling approach seen previously [39], where the mean genomic distance within and between each of the defined cgMLST clusters (measured by the average percent difference of core SNVs) was compared to equal-sized, randomly-selected clusters of genomes from the sample population. The size of the randomly selected groups was kept consistent with the number of genomes for each comparison cluster, and 9999 iterations of random sampling were used to create a distribution of mean similarities for hypothesis testing.…”