Summary estimates of the accuracy of each test for diagnosis of COPD in ever smokers
| Index test* | Studies | Cases/participants | Sensitivity (95% CI) | Specificity (95% CI) | PPV (95% CI) | NPV (95% CI) | NNS (95% CI) | NND (95% CI) |
|---|---|---|---|---|---|---|---|---|
| CDQ (score ≥19.5) | 3 | 495/1703 | 64.5 (59.9 to 68.8) | 65.2 (52.9 to 75.8) | 9.7 (6.9 to 14.2) | 96.9 (95.8 to 97.7) | 29 (26 to 31) | 11 (7 to 15) |
| CDQ (score ≥16.5) | 4 | 580/2322 | 87.5 (83.1 to 90.9) | 38.8 (27.7 to 51.3) | 7.7 (6.3 to 9.8) | 98.2 (96.6 to 99.0) | 21 (20 to 22) | 13 (11 to 16) |
| Handheld flow meters | 3 | 224/1133 | 79.9 (74.2 to 84.7) | 84.4 (68.9 to 93.0) | 23.0 (12.2 to 41.3) | 98.6 (97.9 to 99.1) | 23 (22 to 24) | 5 (3 to 9) |
| CDQ and handheld flow meter | 1 | 90/624 | 74.4 (64.2 to 83.1) | 97.0 (95.2 to 98.3) | 59.1 (43.8 to 74.0) | 98.5 (97.9 to 99.0) | 25 (22 to 29) | 2 (2 to 3) |
The PPV, NPV, NNS and NND to identify one individual with COPD have been calculated assuming a prevalence of undiagnosed COPD of 5.5% in a theoretical population of 1000 people.
*Owing to the complexity of the bivariate model and the limited number of studies, only the four CDQ studies that used a score threshold ≥16.5 were pooled using a bivariate model. We carefully examined the parameter estimates of the model, especially the variances of the random effects, to check whether the model was reliable. There were only three studies of the CDQ that used a score threshold ≥19.5 and three studies of handheld flow meters. These were pooled using univariate random effects logistic regression models.
COPD, chronic obstructive pulmonary disease; CDQ, COPD Diagnostic Questionnaire; NND, number needing a diagnostic assessment to identify one with COPD; NNS, number-needed-to-screen to identify one with COPD; NPV, negative predictive value; PPV, positive predictive value.