Background: Exercise treadmill testing (ETT) is a testing modality that has shown to be a useful chest pain observation unit (CPU). One limitation of this tool is the high rate of nondiagnostic tests. We aim to create a predictive model to discriminate a patient's risk for a nondiagnostic test.
Methods: This is a retrospective analysis of consecutive subjects admitted to our CPU and undergoing an ETT from January 2001 to December 2006. To account for any variation in physician practice, the training set was those patients admitted January 2004 to December 2006 and the testing set comprised those evaluated January 2001 to December 2003. Recursive partitioning with 10-fold cross validation was used to identify significant variables associated with the outcome measure of a nondiagnostic treadmill test. The beta coefficient from the regression model was used to create a risk score. This risk score was then used stratify patients.
Results: A total of 1708 subjects underwent ETT during the study period. The training set comprised 408 subjects with 62 having a nondiagnostic test. Logistic regression identified age, prior history of coronary artery disease, smoking, and diabetes variables used to create a scoring system. The testing set identified 387 (29.7) subjects meeting our criteria as low risk (9.0%) nondiagnostic test and identified 298 (22.9%) at high risk for a nondiagnostic test (32.8%).
Conclusion: Using a simple scoring system to stratify patients undergoing ETT into 3 risk groups, we were able to identify a low-risk group <10% and a high-risk group >30% for having a nondiagnostic ETT.