Abstract
PURPOSE
To estimate from available literature the specificity (true-negative rate) of MR angiography for detecting severe carotid artery stenoses when applied as a confirmatory test after screening with duplex Doppler sonography.
METHODS
We reviewed the pertinent MR angiographic literature published between 1990 and 1994 and recalculated the specificity of MR angiography after deleting from the database results for normal vessels and for vessels with mild and moderate stenoses, since the study of these vessels is not germane to an exploration of the utility of MR angiography as a confirmatory test.
RESULTS
Seventeen articles provided data for our analysis. We divided vessels into four categories on the basis of data supplied within each article. Seven of the articles provided data that could be configured to match the categories used in the North American Symptomatic Carotid Endarterectomy Trial (NASCET). In one study, the criterion of severe stenosis was more than 70% construction, but the moderate category was limited to stenoses of 50% to 69%. The remaining series defined severe stenoses as more than 80% (four series), more than 75% (two series), or more than 60% (three series) constriction. The stated specificity of MR angiography ranged from 64% to 100%. Before revision, 15 of 17 articles had stated specificity values above 75%. Our recalculated values ranged from 18% to 100%. Only seven of 17 studies would have had MR angiographic specificity of greater than 75%. Nine of 17 articles would have had specificities of less than 60%. For all articles specifically identifying vessels with false-positive findings at sonography, the specificity of MR angiography was 16%.
CONCLUSION
To base specificity values for MR angiography as a confirmatory test of carotid artery stenosis on studies that include nondiseased vessels incurs spectrum bias. The actual specificity for MR angiography as a confirmatory test remains unknown, but it is lower than that reported in the literature.
Full Text
The Full Text of this article is available as a PDF (147.7 KB).