TABLE 3.
Summary of dog training on 24 Getxent tubes used in gas chromatography-mass spectrometry (GC-MS) analysis
| Dog 1 | Dog 2 | Dog 3 | Total (%) | Mean (Q3–Q1) | |
| Total runs on specified positive gargle samples (n=14) | 99 | 94 | 71 | 264 (100) | 20 (23) |
| Total runs with false positives | 4 | 1 | 1 | 6 (2.3) | NA |
| Total runs on specified negative gargle samples (n=10) | 55 | 60 | 47 | 162 (100) | 16 (16) |
| Total runs with false negatives | 6 | 2 | 4 | 12 (7.4) | NA |
Using the discrepancy matrix, samples with a failure rate of dog alert above 15% were considered negative for the purpose of this analysis. Altogether, the model correctly classified 68% of the samples with F1 and Matthew's correlation coefficient scores of 0.69 and 0.37, respectively, considered to be a good concordance between the two methods (table 4). Individually, the Kappa agreement between the dogs and GC-MS was moderate: 0.503 for dog A, 0.510 for dog B and 0.412 for dog C.