Latent class comparison of test accuracy when evaluating antimicrobial susceptibility using disk diffusion and broth microdilution to test Escherichia coli and Mannheimia haemolytica isolates recovered from beef feedlot cattle.

The study objective was to use Bayesian latent class analysis to evaluate the accuracy of susceptibility test results obtained from disk diffusion and broth microdilution using bacteria recovered from beef feedlot cattle. Isolates of Escherichia coli and Mannheimia haemolytica were tested for susceptibility to ampicillin, ceftiofur, streptomycin, sulfisoxazole, tetracycline, and trimethoprim-sulfamethoxazole. Results showed that neither testing method was always or even generally superior to the other. Specificity (ability to correctly classify non-resistant isolates) was extremely high for both testing methods, but sensitivity (ability to correctly classify resistant isolates) was lower, variable in the drugs evaluated, and variable between the two bacterial species. Predictive values estimated using Bayesian Markov chain Monte Carlo models showed that the ability to predict true susceptibility status was equivalent for test results obtained with the two testing methods for some drugs, but for others there were marked differences between results obtained from disk diffusion and broth microdilution tests.