Abstract 12636: ECG-Based Risk Stratification of Long-Term Mortality in Suspected Acute Coronary Syndrome

Abstract

Introduction: Non-invasive risk stratification can play a significant role in the initial triage of patients with suspected acute coronary syndrome. Numerous ECG features have been previously linked with mortality in both the general and cardiac populations. We sought to build an ECG-based model that can predict time-to-death in high-risk chest pain patients transported by Emergency Medical Services. Methods: This was a prospective observational cohort study of patients evaluated for suspected acute coronary syndrome. The study recruited consecutive patients evaluated at the emergency department of three UPMC-affiliated tertiary care hospitals. We included consecutive patients in whom a prehospital echocardiogram was performed during indexed encounter. Baseline ECGs were preprocessed using manufacturer specific algorithms. We evaluated 554 temporal-spatial features of the 12-lead ECG in multivariate proportional hazards Cox regression model for predicting all-cause death during long term follow up. Death was ascertained using phenotyping UPMC electronic hospital records and through the CDC National Death Index database. Model fit was evaluated using C-statistic and AIC. We set alpha at 0.01 for two-tailed hypothesis testing. Results: Our sample included 2311 patients (aged 59 ± 16.6, 47% females, 41% Black). Over a median follow up period of 1868 days (IQR = 688; [Q1, Q3] = [1511, 2199]), there was 622 (26.9%) adjudicated deaths. During univariate analysis, 288 features were significantly associated with time-to-death. In the final multivariate model (c-statistic = 0.79, AIC = 9066.05), controlling for demographic and clinical data, 15 ECG features remained predictive of the study outcome, with Selvester score being the strongest predictor (Hazard Ratio = 1.09, 95% CI = 1.03 - 1.16). Other clinically important ECG features included horizontal-plane QRS axis, QRS-T angle, QTc interval, QTpeak, T peak-end, S wave morphology the inferior leads, QRS morphology in lead aVR, and T wave morphology in the anterior leads. Conclusions: The presenting 12-lead ECG can serve as a powerful risk stratification tool of long-term mortality in patients with chest pain. Selvester score can serve as a plausible and easy to use clinical decision tool.