Diagnostic accuracy of a bayesian network model in acute coronary syndromes

Abstract

Introduction: The characterization and diagnosis of chest pain, with emphasis on acute coronary syndromes (ACS), is a fundamental requirement for the doctors at the emergency service. Objective: The aim of the present study is to design and evaluate the performance of Bayesian networks to back up the diagnosis of ACS. Methodology: A diagnostic tests study in which two models of Bayesian networks  were designed and trained in the framework OpenMarkov, using the variables of the Braunwald angina probability scale in a group of 159 patients, which was validated afterwards in a cohort of 108 adult patients hospitalized with suspicion of ACS in a third level hospital. Results: Low sensitivity was obtained, with adequate specificity and positive predictive values, though (62, 86, and 87% respectively). Performance was better in the cases that had electrocardiogram and negative biomarkers. Conclusions: A model of Bayesian networks trained from the variables of the Braunwald unstable angina probability scale, exhibits an acceptable performance for the diagnosis of ACS.