Skip to main content
. 2017 May 12;64(11):1494–1501. doi: 10.1093/cid/cix169

Table 3.

Sensitivity and specificity for predicting phenotypic resistance using the Coll et al. and Cohen et al. lists of mutations

Drug Mutation set True Positivesa False Negatives False Positives True Negatives Ambiguousa Sensitivity Specificity
Isoniazid Coll 24 9 9 155 3 72.7% 94.5%
Cohen 25 9 5 159 2 73.5% 97.0%
Coll+Cohen 26 8 9 155 3 76.5% 94.5%
Rifampicin Coll 8 3 2 186 1 72.7% 98.9%
Cohen 8 3 1 187 1 72.7% 99.5%
Coll+Cohen 8 3 2 186 1 72.7% 98.9%
Streptomycin Coll 4 11 3 180 1 26.7% 98.4%
Cohen 9 6 7 177 1 60.0% 96.2%
Coll+Cohen 9 6 7 176 1 60.0% 96.2%
Ethambutol Coll 4 1 137b 57 2 80.0% 29.4%
Cohen 2 2 1 194 2 50.0% 99.5%
Coll+Cohen 4 1 137 57 3 80.0% 29.4%

aAmbiguous single-nucleotide polymorphism calls, with reads supporting a drug sensitive and a drug resistant genotype. These positions were excluded from the calculations for sensitivity and specificity.

bThe large number of false positives for streptomycin using the Coll dataset is due to the presence of the embB E378A mutation, which other publications suggest does not cause ethambutol resistance. This mutation has been previously reported as a phylogenetic marker of lineage 1 of the ancestral MTBC, and this mutation is indeed present in most lineage 1 strains in this collection.

cAnalysis performed for 201 strains, after removing highly similar strains found within the same patient.