|
| Type of bias | Sensitivity [3, 16, 22, 26] | Specificity [3, 16, 22, 26] | RDOR from Rutjes et al. [16] | RDOR from Lijmer et al. [3] |
|
| Sampling bias (consecutive over nonconsecutive sampling) | Increases if complex cases are excluded | Increases if complex cases are excluded | 1.5, 95% CI (1.0–2.1) | 0.9, 95% CI (0.7–1.1) |
| Decreases if clear-cut cases are excluded | Decreases if clear-cut cases are excluded | | |
| Spectrum bias | Increases when severe cases are overrepresented in the patient sample (“the sickest of the sick”) | Increases when healthy controls are overrepresented in the patient sample (“the healthiest of the healthy”) | 4.9, 95% CI (0.6–37.3) | 3.0, 95% CI (2.0–4.5) |
| Information bias: lack of blinding | Variable | Variable | 1.1, 95% CI (0.8–1.6) | 1.3, 95% CI (1.0–1.9) |
| Information bias: post hoc definition of cutoff | Increases | Increases | 1.3 95% CI (0.8–1.9) | Not studied |
| Misclassification bias (imperfect gold standard) | Increases if errors in index and reference test are correlated | Increases if errors in index and reference test are correlated | Not studied | Not studied |
| Decreases if errors in index and reference test are independent | Decreases if errors in index and reference test are independent |
| Incorporation bias | Increases | Increases | 1.4, 95% CI (0.7–2.8) | Not studied |
| Verification bias: differential (i.e., different reference standards) | Increases if the gold standard is used for positive index results and a different reference test (e.g., noninvasive and less expensive) is used for negative index results | Increases if the gold standard is used for positive index results and a different reference test (e.g., noninvasive and less expensive) is used for negative index results | 1.6, 95% CI (0.9–2.9) | 2.2, 95% CI (1.5–3.3) |
| Verification bias: partial | Increases | Decreases | 1.1, 95% CI (0.7–1.7) | 1.0, 95% CI (0.8–1.3) |
|
