Stratification of arrhythmic risk in patients admitted for acute coronary syndrome: the role of the machine learning-derived PRAISE score

Abstract

Abstract Funding Acknowledgements None. Background Patients with acute coronary syndrome (ACS) have a high arrhythmic burden, and either ventricular and supraventricular arrhythmias are very common during hospitalization[1]. Patients who develop arrhythmias show higher mortality and worse prognosis. The PRAISE (Prediction of Adverse Events Following an Acute Coronary Syndrome) score is a machine-learning based model for predicting 1-year all cause death, recurrent acute myocardial infarction and major bleedings in patients after ACS[2]. To date, its role to predict arrhythmic complications in ACS remains unknown. Purpose We hypothesized that patients with higher PRAISE score have a higher arrhythmic burden. Aim was to evaluate in a prospective protocol the PRAISE score capability for identifying patients with ACS at higher risk of arrhythmic complications during in-hospital stay. Methods A total of 365 consecutive patients admitted to our cardiac intensive care unit for ACS and undergoing percutaneous coronary intervention were enrolled. All patient was monitored by continuous electrocardiogram during hospitalization. The PRAISE score was obtained for each patient <24 hours from admission. Patients were divided into 2 groups based on the results of the PRAISE score for all-cause of death as patients without high PRAISE score (e.g. with low to intermediate risk, n=350) and patients with high PRAISE score (e.g. with high risk, n=15). The occurrence of atrial fibrillation (AF) and ventricular arrhythmias (VAs) up to discharge was recorded. AF was defined by detection of at least one episode of uncoordinated atrial electrical activation and irregular R-R intervals, in absence of distinct repeating P waves and with irregular atrial activations lasting at least 5 minutes. We considered as VAs all episodes of ventricular non-sustained tachycardia, sustained ventricular tachycardia and ventricular fibrillation. Results ROC curve analysis indicated a significant relationship between high PRAISE score and risk of both in-hospital AF (AUC 0.89, 95%CI 0.82-0.94, p=0.0001) and VAs (AUC 0.69, 95%CI 0.64-0.75, p=0.0001). A high PRAISE score had a specificity of 99% for AF and 92% for VAs. By Kaplan-Meier analysis, patients with high PRAISE score more frequently developed AF (30% vs 4% in those without high score; log rank p=0.00001) and VAs (77% vs 35%; log rank p=0.00001). Multivariate analysis showed that a high PRAISE score was an independent predictor of both AF (HR 4.08, 95%CI 1.14-14.58, p=0.030) and VAs (HR 2.36, 95%CI 1.08-5.14, p=0.030). Conclusions In patients hospitalized for ACS, the PRAISE score has a comprehensive capability to identify with high specificity those patients prone to develop arrhythmic events during hospitalization. This may be important in order to stratify patients with higher arrhythmic risk in whom individualized strategies may improve the clinical outcome.Baseline features of the populationKaplan-Meier anaysis