Prediction of traumatic aortic rupture from plain chest film findings using stepwise logistic regression.

We reviewed initial chest radiographs of 21 patients with, and 26 without, aortic rupture, and examined the presence or absence of individual signs previously cited to be associated with aortic rupture. Using stepwise logistic regression, the three most significant signs associated with rupture were loss of aortic contour, tracheal deviation, and mediastinal-to-chest ratio. A formula calculating the probability of aortic rupture (P) using these three variables was derived. Using a low cutoff point, this equation would approach 100% sensitivity in detecting aortic rupture while reducing the number of negative aortographs. The reliability of this equation and the optimal cutoff point must be determined in a prospective study before being used to make clinical decisions.