Development and internal validation of a scoring model for early identification of ventricular arrhythmia risk in children with acute myocarditis.

BACKGROUND

Early identification of ventricular arrhythmia (VA) risk in children with acute myocarditis (AMC) is challenging, as existing tools lack pediatric targeting. AMC’s non-specific early symptoms in children lead to underdiagnosis. VA, a major complication of AMC, is linked to poor prognosis, but not all cases present at admission, highlighting the need for simple, accessible predictive indicators.

METHODS

This retrospective single-center study included 312 children (1 month − 17 years) with AMC (2021–2024), divided into a training set (n = 208, 2021–2023) and validation set (n = 104, 2024). Eligibility required first-time AMC diagnosis without admission VA. Univariate analysis (P < 0.05) and binary logistic regression identified independent VA risk factors. Continuous variables were dichotomized via receiver operating characteristic curves, and a scoring model was constructed, with internal validation of discriminative performance and calibration.

RESULTS

Forty-four children (14.1%) developed VA, mostly within the first week. Independent risk factors were cardiac troponin I (cTnI ≥ 0.1945 ng/mL, OR = 9.114), blood urea nitrogen (BUN ≥ 4.55 mmol/L, OR = 9.796), and left ventricular fraction shortening (LVFS ≤ 0.33, OR = 6.005). The model assigned 3 points to cTnI/BUN and 2 to LVFS (total 0–8 points). At cutoff ≥ 4 points: training set (sensitivity = 70.0%, specificity = 90.4%, AUC = 0.871); validation set (sensitivity = 78.6%, specificity = 94.4%, accuracy = 92.3%, AUC = 0.837). Calibration was acceptable (Hosmer-Lemeshow P = 0.288).

CONCLUSION

The cTnI-, BUN-, and LVFS-based scoring model offers a simple, effective tool for early VA risk stratification in children with AMC. It aids targeted monitoring and intervention, improving clinical decision-making. Limitations include retrospective single-center design and small sample size; prospective multicenter validation is needed.