Accuracy of a Neural Diagnostic Tree for the Identification of Acute Coronary Syndrome in Patients With Chest Pain and No ST-Segment Elevation

Abstract

Identifying acute coronary syndrome is a difficult task in the emergency department because symptoms may be atypical and the electrocardiogram has low sensitivity. In this prospective cohort study done in a tertiary community emergency hospital, we developed and tested a neural diagnostic tree in 566 consecutive patients with chest pain and no ST-segment elevation for the diagnosis of acute coronary syndrome. Multivariate regression and recursive partitioning analysis allowed the construction of decision rules and of a neural tree for the diagnosis of acute myocardial infarction and acute coronary syndrome. Predictive variables of acute coronary syndrome were: age > or =60 years (odds ratio [OR] = 2.3; P = 0.0016), previous history of coronary artery disease (OR = 2.9; P = 0.0008), diabetes (OR = 2.8; P = 0.0240), definite/probable angina-type chest pain (OR = 17.3; P = 0.0000) and ischemic electrocardiogram (ECG) changes on admission (OR = 3.5; P = 0.0002). The receiver operating characteristic curve of possible diagnostic decision rules of the regression model disclosed a C-index of 0.904 (95% confidence interval = 0.878 to 0.930) for acute coronary syndrome and 0.803 (95% confidence interval 0.757 to 0.849) for acute myocardial infarction. For both disorders, sensitivities of the neural tree were 99% and 93%, respectively, and negative predictive values were both 98%. Negative likelihood ratios were 0.02 and 0.1, respectively. It is concluded that this simple and easy-to-use neural diagnostic tree was very accurate in the identification of non-ST segment elevation chest pain patients without acute coronary syndrome. Patients identified as low probability of disease could receive immediate stress testing and be discharged if the test is negative.