Identification of prognostic factors for pediatric myocarditis with a random forests algorithm-assisted approach

Background Pediatric myocarditis is a rare disease with substantial mortality. Little is known regarding its prognostic factors. We hypothesize that certain comorbidities and procedural needs may increase risks of poor outcomes. This study aims to identify prognostic factors for pediatric myocarditis. Methods The national Kids' Inpatient Database was used in the study. A random forests algorithm was implemented for mortality prediction based on comorbidities and procedures. Linear regression analysis was then performed to quantify their associations with mortality and length of stay. Results The prevalence of pediatric myocarditis among all pediatric hospitalizations doubled from 2003 to 2016. The mortality rate peaked in 2006 (6.7%) and declined steadily thereafter, with a rate of 3.2% in 2016. Brain injury (including encephalopathy, cerebral edema, and intracranial hemorrhage), acute kidney injury, dysrhythmias, coagulopathy, sepsis, and ECMO use were all independent prognostic factors associated with increased mortality and prolonged hospital stay. Conclusion Prognostic factor identification may not be straightforward in rare diseases such as pediatric myocarditis due to small cohort size in each treating facility. Findings from this report provide insights into the prognostic factors for pediatric myocarditis, and may allow clinicians to be better prepared when informing patients and their families regarding disease outcomes. Impact The rate of hospitalization due to pediatric myocarditis was increasing but the mortality rate was declining over the past decade. End organ damage, including the brain and the kidney, was associated with mortality and prolonged hospital stay in pediatric myocarditis. Tachyarrhythmias and cardiac function compromise requiring ECMO were also associated with mortality and prolonged hospital stay. A data science approach combining machine learning algorithms and conventional regression modeling using a large dataset may facilitate risk factor identification and outcome correlation in rare diseases, as illustrated in this study.