Diagnostic Accuracy of Computed Tomographic Angiography for Blunt Cerebrovascular Injury Detection in Trauma Patients

Authors:
Publisher website
Google Scholar
Abstract
To compare the diagnostic accuracy of computed tomographic angiography (CTA) with digital subtraction angiography (DSA) for blunt cerebrovascular injury (BCVI) detection in trauma patients. Controversy exists as to whether the diagnostic performance of CTA compares favorably with the reference-standard, DSA. We searched electronic databases (1950 to May 22, 2012), article bibliographies, conference proceedings (2008-2011), and clinical trial registries for studies comparing the accuracy of CTA with DSA for BCVI detection in trauma patients. Pooled estimates of sensitivity, specificity, and positive and negative likelihood ratios were calculated using bivariate random effects models. Eight studies that examined 5704 carotid or vertebral arteries in 1426 trauma patients met inclusion criteria. The pooled sensitivity and specificity for BCVI detection with CTA versus DSA was 66% (95% CI, 49%-79%; I = 80.4%) and 97% (95% CI, 91%-99%; I = 94.6%), respectively. Corresponding pooled positive and negative likelihood ratios were 20.0 (95% CI, 6.9-58.4; I = 87.7%) and 0.35 (95% CI, 0.22-0.56; I = 74.9%), respectively. Although pooled sensitivity varied with the number of available CT slices, the training of interpreting radiologists, and in a pattern suggestive of differences in diagnostic threshold for judging CTA positivity, it remained 80% or less among studies that used scanners with 16 or more slices per rotation and where the CTA was read by neuroradiologists. Existing evidence suggests that the diagnostic performance of CTA varies considerably across studies, likely due to an implicit variation in diagnostic threshold across trauma centers. Moreover, although CTA appears to lack sensitivity to adequately rule out BCVI, it may be useful to rule in BCVI among trauma patients with a high pretest probability of injury.

This publication has 42 references indexed in Scilit: