Risk prediction of clinical events in acute myocarditis patients: a frequentist and Bayesian analysis from the AMPHIBIA registry

Abstract Background To date, there is no validated risk stratification tools for patients presenting with acute myocarditis. The aim of this study was to develop tools based on clinical characteristics at admission to stratify the risk of outcomes in myocarditis patients using a frequentist and a Bayesian approach. Methods This is a retrospective cohort study based on the AMPHIBIA cohort, including patients with proven myocarditis hospitalized between 2008 and 2019. The composite outcome was defined as temporary circulatory support implantation, heart transplantation or death. A Cox model was used to identify variables associated with the risk of outcome and to create a score-based model from their coefficient of determination. An augmented Markov learning algorithm was used to modelize a Bayesian network stratifying the risk of outcomes. Results 359 patients were included with a median follow-up of 4.1 years. The composite outcome occurred in 67 (18.7%) patients. In multivariate analysis, 7 variables were associated with the primary outcome. The Bayesian network model selected 6 variables. After cross validation, both approaches demonstrated similar performance (area under the curve 0.92 and 0.91) and identify patient at low, intermediate and high risk of events. Conclusions Based on these frequentist and Bayesian approaches, this analysis allows to stratify with good performance the risk of severe clinical events among patients hospitalized for an acute myocarditis.Central figure