Figure 2.

Example of the effect of verification bias in a hypothetical true distribution of degree restenosis in patients treated with a carotid artery stent. If a DUS cut-off value of for example 50% stenosis is used to select patients for a diagnostic study, only the patients in part B and C are included and will undergo the reference test. Patients in part A are excluded. The number of patients below a certain threshold, for example 70% is too low (proportion B instead of proportion A and B) and therefore leads to incorrect estimates of the diagnostic value, the specificity in particular. Therefore, DUS criteria obtained from a diagnostic study with comparison with the reference test in part B and C, can not be applied to routine clinical practice for all stented patients (part A, B, and C).

Nederkoorn and Brown BMC Neurology 2009 9:36   doi:10.1186/1471-2377-9-36
Download authors' original image