Risk prediction of death and heart transplantation in adult patients with myocarditis

Aims: Identifying early risk predictors in myocarditis is clinically relevant, as patients’ outcomes may be very diverse. We aimed to explore predictors of death and heart transplant (HTx) in a large single-centre cohort of adult patients with myocarditis using a machine learning (ML) technique. Methods and results: We retrospectively enrolled consecutive adult patients with biopsy-proven or clinically suspected myocarditis, collecting clinical, laboratory, and imaging data, both at diagnosis and during follow-up. A predictive model of death/HTx was developed using random forest (RF), ranking covariates according to their predictive accuracy. We included 938 patients (median age 36 years, 69% male) with clinically suspected (n = 549) or biopsy-proven (n = 389) myocarditis. During follow-up, 35 patients died, and 26 underwent HTx. The most important variables in predicting survival were NYHA class (variable importance, VIMP, 10%) LVEF (3.6%) and clinical presentation (2.5%) at diagnosis, histological type of myocarditis on endomyocardial biopsy (EMB)(2.9%), anti-endothelial cell antibodies (0.6%), and anti-nuclear antibodies (0.4%) positivity. Overall, the predictive accuracy of our RF model was good (89.2%, 95% C.I. 86.1–92.3). Conclusion: Based on a ML approach, we found, with good predictive accuracy, that advanced NYHA class, reduced LVEF and heart failure at diagnosis, and giant cell myocarditis on EMB are predictors of worse prognosis in adult patients with myocarditis.