Skip to main content

Table 2 Performance of the readers with the score-based and with the standard approach

From: Improving diagnostic accuracy in assessing pulmonary edema on bedside chest radiographs using a standardized scoring approach

 

Score-based approach

Standard approach

 
 

Pulmonary edema

No pulmonary edema

Pulmonary edema

No pulmonary edema

Net reclassification index

Reader 1 (3)

8

10

5

10

0.3 (0.04; 95% CI: 0.02 – 0.58)

Reader 2 (3)

6

10

5

10

0.1 (0.56; 95% CI: -0.23 – 0.43)

Reader 3 (3)

8

10

6

6

0.6 (0.003; 95% CI: 0.21 – 0.99)

Reader 4 (4)

8

10

5

9

0.4 (0.02; 95% CI: 0.06 – 0.74)

Reader 5 (4)

7

10

5

9

0.3 (0.06; 95% CI: -0.01 – 0.61)

Reader 6 (6)

9

10

8

10

0.1 (0.29; 95% CI: -0.09 – 0.29)

Reader 7 (7)

8

10

6

9

0.3 (0.06; 95% CI: -0.01 – 0.61)

All readers

54/70

70/70

40/70

63/70

0.3 (0.000002; CI: 0.18 – 0.42)

Mean sensitivity [%]

77.14 (95% CI: 65.28 – 85.99)

57.14 (95% CI: 44.78 – 68.72)

 

Mean specificity [%]

100 (95% CI: 93.52 -100)

90.00 (95% CI: 79.90 – 95.55)

 

Positive predictive value [%]

100 (95% CI: 91.73 – 100)

85.11 (95% CI: 71.08 – 93.31)

 

Negative predictive value [%]

81.40 (95% CI: 71.25 – 88.67)

67.74 (95% CI: 57.15 – 76.85)

 

Free-marginal multirater kappa value

0.68

0.34

 

Percentage of overall agreement [%]

67

84

 

Duration [seconds]

23 ± 12

7 ± 3

 
  1. The first seven rows show the numbers of correct diagnoses (10 bedside CXRs total) during the assessment of ten bedside chest radiographs with and without pulmonary edema each, practicing the score-based and the standard approach (extravascular lung water (EVLW) measurements determined by a pulse contour cardiac output (PiCCO) system served as reference standard). Work experience of the participating radiologists in years is shown in brackets. The net reclassification index demonstrates the improvement in prediction performance gained by the score-based approach (p-values and confidence intervals are shown in brackets). Sensitivities, specificities, positive and negative predictive values, free-marginal multirater kappa values, the percentage of overall agreement and duration are shown for both approaches. The confidence level is 95%. CI = Confidence interval.