Skip to main content

Table 3 Accuracy Measure (Sensitivity, Specificity, PPV and NPV) Estimates, with 95% Confidence Intervals, Adjusted for Verification Bias

From: Adjusting for verification bias in diagnostic accuracy measures when comparing multiple screening tests - an application to the IP1-PROSTAGRAM study

  

Statistical Methods for Verification Bias Adjustment

 

Complete Case Analysis (Unadjusted Approach)

(N = 166)

Begg and Greenes using Three Screening Tests

(N = 403)

Multiple Imputation using Three Screening Tests (using standard logistic regression)

(N = 403)

Multiple Imputation using Three Screening Tests, using penalised logistic regression (Firth’s method)

(N = 403)

Prevalence of clinically significant prostate cancer

9.6%

(5.6 to 15.2%)

4.5%

(2.6 to 6.8%)

4.7%

(3.0 to 7.3%)

4.0%

(2.5 to 6.4%)

Screening Tests

MRI

Prevalence of positive MRI

53.0%

(45.4 to 60.5%)

24.1%

(20.1 to 28.5%)

24.1%

(20.1 to 28.5%)

24.1%

(20.1 to 28.5%)

Accuracy measures

Sens = 87.5%

(61.7 to 98.4%)

Spec = 50.7%

(42.4 to 58.9%)

PPV = 15.9%

(9.0 to 25.2%)

NPV = 97.4%

(91.0 to 99.7%)

Sens = 87.9%

(68.8 to 100.0%)

Spec = 78.9%

(74.8 to 82.9%)

PPV = 16.4%

(9.8 to 24.8%)

NPV = 99.3%

(98.0 to 100.0%)

Sens = 82.2%

(58.9 to 88.0%)

Spec = 78.8%

(74.4 to 82.6%)

PPV = 16.0%

(10.0 to 24.9%)

NPV = 98.9%

(97.0 to 99.6%)

Sens = 87.5%

(64.0 to 96.5%)

Spec = 78.6%

(74.2 to 82.4%)

PPV = 14.4%

(8.8 to 22.8%)

NPV = 99.3%

(97.6 to 99.8%)

US

Prevalence of positive ultrasound

51.2%

(43.6 to 58.8%)

23.6%

(19.7 to 28.0%)

23.6%

(19.7 to 28.0%)

23.6%

(19.7 to 28.0%)

Accuracy measures

Sens = 56.3%

(29.9 to 80.2%)

Spec = 49.3%

(41.1 to 57.6%)

PPV = 10.6%

(5.0 to 19.2%)

NPV = 91.4%

(83.0 to 96.5%)

Sens = 56.3%

(31.7 to 78.3%)

Spec = 78.0%

(73.8 to 81.6%)

PPV = 10.7%

(5.2 to 18.1%)

NPV = 97.4%

(95.3 to 99.0%)

Sens = 53.7%

(31.6 to 74.3%)

Spec = 77.9%

(73.5 to 81.7%)

PPV = 10.6%

(5.9 to 18.7%)

NPV = 97.2%

(94.7 to 98.5%)

Sens = 56.3%

(33.2 to 76.9%)

Spec = 77.8%

(73.4 to 81.6%)

PPV = 9.5%

(5.1 to 17.0%)

NPV = 97.7%

(95.4 to 98.9%)

PSA

Prevalence of positive PSA

21.1%

(15.5 to 28.0%)

9.7%

(7.1 to 13.0%)

9.7%

(7.1 to 13.0%)

9.7%

(7.1 to 13.0%)

Accuracy measures

Sens = 37.5%

(15.2 to 64.6%)

Spec = 80.7%

(73.4 to 86.7%)

PPV = 17.1%

(6.6 to 33.6%)

NPV = 92.4%

(86.4 to 96.3%)

Sens = 36.4%

(14.2 to 63.8%)

Spec = 91.6%

(88.5 to 94.0%)

PPV = 16.9%

(5.4 to 29.3%)

NPV = 96.8%

(94.5 to 98.4%)

Sens = 35.6%

(17.1 to 58.0%)

Spec = 91.5%

(88.4 to 93.9%)

PPV = 16.6%

(8.1 to 31.4%)

NPV = 96.6%

(94.2 to 98.0%)

Sens = 37.5%

(18.5 to 61.4%)

Spec = 91.5%

(88.3 to 93.9%)

PPV = 15.4%

(7.2 to 29.7%)

NPV = 97.3%

(95.0 to 98.5%)

  1. Accuracy measure (sensitivity, specificity, PPV and NPV) estimates adjusted for verification bias, using complete cases analysis (unadjusted approach), Begg and Greenes using three screening tests, multiple imputation using three screening tests, and multiple imputation using three screening tests and penalised logistic regression (Firth’s correction) (IP1-PROSTAGRAM trial data [12])