Abstract 324: Automated Classification of Proximal Left Anterior Descending and Proximal Right Coronary Artery Occlusion from More Distal Locations in ST-Segment Elevation Myocardial Infarction

Abstract

Background: Rapid classification of the occlusion location in the coronary artery tree during ST-elevation myocardial infarction (STEMI) can be used for further risk stratification. In this study, we introduce and characterize a complete automated algorithm for classifying occlusion location as left circumflex, or proximal or middle-to-distal in the left anterior descending and right coronary artery in STEMI. Methods: The new algorithm was developed and tested on two separate sets of ECGs using multiple decision trees optimized for high specificity. The development set of ECGs came from the emergency department of a large hospital. The test set of ECGs came from a prehospital emergency medical service using inclusion criteria of STEMI and angiogram confirmation of single culprit artery and lesion location (n=132). LBBB and paced ECGs were excluded. All left circumflex (LCx) locations (n=31) were lumped together. The split between proximal right coronary artery (RCA) lesion location (n=51) and middle-to-distal RCA location (n=50) was defined as above the acute marginal branch. Proximal lesion for the left anterior descending (LAD) coronary artery was defined as above the first major diagonal side branch. All patients met current age and gender based STEMI criteria. Accuracy of lesion location prediction was based on sensitivity and specificity calculated from a 2X2 table with the two groups as lesion location and other. Results: The automated algorithm classifies proximal LAD lesion location with a sensitivity and specificity of 75% and 94% respectively and sensitivity and specificity of 57% and 99% for proximal RCA lesion location. Conclusion: Since proximal lesion location indicates higher risk for both LAD and RCA, the classifier was designed for higher specificity and the specificity within STEMI cases was shown to be 94% and higher. High specificity is necessary for confidence in the classification before altering treatment. An automated algorithm for proximal lesion location allows for rapid risk assessment which should result in a valuable clinical decision support tool.