Can we rely on mediastinal widening on chest radiography to identify subjects with aortic injury?

Widening of the mediastinum on chest radiography is widely promoted as a useful criterion for detecting aortic injury. We sought to determine the reliability, sensitivity, and specificity of this sign. The initial chest radiographs from 30 subjects with aortic injury and 47 controls were independently reviewed by six radiologists, who were blinded to diagnosis. The radiologists were asked to decide whether the mediastinum was normal or not normal, as well as whether the mediastinum was widened. Agreement, sensitivity, and specificity were assessed. Agreement for overall assessment of the mediastinum was substantial (kappa = 0.64). Individual radiologists had sensitivity varying from 0.77 to 0.97 and specificity varying from 0.62 to 0.89. For "widening" of the mediastinum, agreement was moderate (kappa = 0.49). "Widening" was less sensitive than the radiologists' overall impression (P = 0.01), varying from 0.50 to 0.83, but no difference was detected in specificity (P = 0.36), varying from 0.81 to 0.94. Mediastinal width has unacceptable sensitivity for predicting aortic injury, with substantial inter-reader variability. Medical education has ingrained the widely promoted concept of mediastinum widening, which may be misleading.