Abstract
Background
Risk stratification for death and heart transplantation (HTx) in myocarditis is complex. A random forest (RF) is a tree-based machine learning technique (MLT) which is being increasingly used for clinical data analysis; it allows the detection of complex relationships between the outcome of interest and the covariates, overcoming the limits of traditional statistical analysis (i.e. regression approaches). Purpose To assess the potential role of clinical and diagnostic features at presentation as predictors of death and HTx in biopsy (Bx)-proven myocarditis using RF. Methods From January 1993 to August 2019, we consecutively enrolled 357 patients with Bx-proven myocarditis (65% male, median age 39 years, interquartile range (IQR) 26–51). An RF approach for survival data was used. Variables included in the analysis were: histology type by Bx, NYHA, type of presentation (infarct-like, arrhythmia, heart failure), viral genome detection on Bx, serum antiheart (AHA), antiintercalated disk (AIDA), anticardiac endothelial cells (AECA), antinuclear (ANA) autoantibodies, immunosuppressive therapy, cardiac catheterisation (left ventricular enddiastolic volume (LVEDV), mean capillary wedge pressure, right and left ventricular enddiastolic pressure) and 2-D echocardiographic measures (LVEDV, left ventricular ejection fraction (LVEF) at presentation and at follow-up, right ventricular fractional area change (FAC%), right ventricular diastolic area). Results The median follow-up time was of 1352 days (IQR 423.25–2535.75). At the end of follow-up, 42 patients were dead or transplanted. The 1-year, 5-year, and 10-year survival probabilities were of 0.928, 0.854, and 0.817, respectively. The most relevant predictors of death or HTx identified by the RF algorithm (according to the variable importance measure) were histological type, NYHA, clinical presentation, LVEF, and FAC%. Among the circulating auto-antibodies AECA were found to be the most important. Histological type was the strongest predictor of death/HT (100% relative importance, (RI)), giant cell myocarditis having a lower survival probability compared to other types. The next stronger predictors were advanced (III-IV) NYHA and heart failure presentation with lower survival probabilities (90% and 84% RI respectively). AECA-positive patients had lower survival probability compared to AECA negative ones (20% RI). The RF algorithm revealed an excellent predictive performance in the correct identification of all alive patients, with only 5 dead patients being misclassified (balanced accuracy 94%).
Conclusions
Autoimmune features, i.e Giant cell myocarditis and AECA, as well as severity of heart failure and of left ventricular disfunction at presentation were the strongest predictors of dismal prognosis. Our RF approach provides a new automated powerful tool for accurate risk stratification for death/HTx in Bx-proven myocarditis.
Funding Acknowledgement
Type of funding source: Public Institution(s). Main funding source(s): Budget Integrato per la Ricerca dei Dipartimenti (BIRD, year 2019), Padova University, Padova, Italy (project Title: Myocarditis: genetic background, predictors of dismal prognosis and of response to immunosuppressive therapy.)
Giant Cell Myocarditis Following COVID-19 Successfully Treated by Immunosuppressive Therapy
