Table 2 Performance of CAD4TBv5 and v6 to identify definite TB at the radiologist’s sensitivity.

From: Computer-aided interpretation of chest radiography reveals the spectrum of tuberculosis in rural South Africa

 

Sensitivity % (CI)

Specificity % (CI)

PPV % (CI)

NPV % (CI)

% of participants required to test (absolute n)

NNT

Cases missed (%)

Radiologist: Any lung field abnormality

80.8 (71.7–88.0)

66.9 (65.6–68.2)

4.7 (3.8–5.8)

99.4 (99.1–99.7)

20.2 (2002)

25

19 (19.2)

CAD4TBv5: 39

82.8 (73.9–89.7)

55.4 (54.0–56.8)

3.6 (2.9–4.5)

99.4 (99–99.6)

28.4 (2820)

34

17 (17.2)

CAD4TBv5: 40

79.8 (70.5–87.2)

57.4 (56.0–58.8)

3.7 (2.9–4.5)

99.3 (98.9–99.6)

27.0 (2677)

34

20 (20.2)

CAD4TBv6: 47

82.8 (73.9–89.7)

62.6 (61.2–64)

4.3 (3.4–5.3)

99.4 (99.1–99.7)

23.7 (2348)

29

17 (17.2)

CAD4TBv6: 48

76.8 (67.2–84.7)

68.0 (66.7–69.3)

4.6 (3.7–5.8)

99.3 (99–99.6)

20.1 (1997)

26

23 (23.2)

  1. Definite TB was defined as microbiological evidence, either by positive XpertUltra or culture. Presented are the radiologist’s sensitivity to triage definite TB by classifying ‘any lung field abnormality’ in comparison to CAD4TBv5 and v6 scores that closely matched the radiologist’s sensitivity. Performance is given as sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV) with 95% confidence intervals (CI); percentage of participants required to be tested at this threshold, the number of needed to test (NNT) to identify one person with definite TB and the number of missed definite TB cases.
Back to article page