Skip to main content
. 2017 May 22;70(6):500–507. doi: 10.1136/jclinpath-2016-204089

Table 3.

Performance of the assay

Indeterminate (Bethesda III, IV,V)* Indeterminate (Bethesda III,IV)* Bethesda II and Bethesda VI*
Training† #Malignant 115 59 40
#Benign 147 137 26
Sensitivity 86 [78–92] 78 [65–88] 96 [85–100]
Specificity 75 [67–81] 76 [68–83] 82 [62–94]
Validation‡, entire set #Malignant 61 31 0
#Benign 128 119 0
Sensitivity 85 [74–93] 74 [55–88] NA
Specificity 72 [63–79] 74 [65–82] NA
NPV 91 [84–96] 92 [84–96] NA
PPV 59 [48–69] 43 [29–57] NA
Validation‡, agreement set #Malignant 40 14 0
#Benign 110 102 0
Sensitivity 98 [87–100] 100 [77–100] NA
Specificity 78 [69–85] 80 [71–88] NA
NPV 99 [94–100] 100 [96–100] NA
PPV 62 [49–74] 41 [25–59] NA
Evaluation study on Bethesda II/VI samples§ #Malignant NA NA 9
#Benign NA NA 32
Sensitivity NA NA 89 [52–100]
Specificity NA NA 63 [44–79]

*95% CIs are in square brackets.

†For training, estimates are based on the mean of 10 10-fold cross-validation runs. Samples with very low expression in any of the classification steps, as well as medullary samples, are not included.

‡Samples that failed quality control are not included.

§Additional blinded study.

NA, not applicable.