Incremental value of the exercise test for diagnosing the presence or absence of coronary artery disease.

Abstract

To determine the incremental value of the exercise test (ETT) for diagnosing coronary artery disease (CAD), we derived a multivariate logistic regression model for the pre-ETT prediction of CAD using data from 3840 patients at Duke University. We then applied the model to 324 patients at the Brigham and Women's Hospital. Using seven clinical factors, the multivariate model had an 84% overall predictive accuracy on both the training (Duke) and the validation (Brigham) sets of patients. Three ETT factors (ST-segment change in patients not taking digitalis, absence of ST-segment change in patients taking digitalis, ETT stopped because of ECG or blood pressure changes) had incremental, significant predictive power, but overall predictive accuracy based on both clinical and ETT factors improved only to 87%. When the ETT result was important enough to move the probability of CAD across a potential therapeutic threshold, the direction of the change in probability was correct only two-thirds of the time. Thus, the ETT was of limited value in predicting the presence or absence of CAD after other easily obtainable clinical data were taken into account.