Adapting the REVEAL Lite 2 score to the pediatric population through machine learning: insights from the REHIPED registry for pediatric pulmonary hypertension

INTRODUCTION AND OBJECTIVES: Risk scores for pulmonary hypertension (PH) have been proven useful in adults. However, no risk score has been validated in pediatric populations. Our aim was to adapt the REVEAL Lite 2 score to pediatrics, including groups 1 (pulmonary arterial hypertension) and 3 (PH associated with lung diseases). METHODS: Data from the REHIPED was used, which is the PH registry of the Spanish Society of Pediatric Cardiology and Congenital Heart Disease. The registry also incorporates some Colombian centers. Patients in group 1 and group 3 were included. The score was derived using machine learning. The contribution of each variable and its cutoffs were identified through gradient boosting. The REVEAL Lite 2 variables were normalized and entered into the model, along with each patient's weight, sex, and age. Risk stratification into 3 categories (low, intermediate, and high risk) was performed. RESULTS: A total of 420 children were analyzed. The final model included functional class, age at diagnosis, heart rate, weight-for-height percentile, systolic blood pressure, natriuretic peptides, weight-for-age percentile, and 6-minute walk test. The area under the curve was 0.82, with an estimated area under the curve of 0.73 on unseen data. The log-rank P value for differences in transplant-free survival among the 3 risk strata (low, intermediate, and high) was<.001. HR for intermediate risk vs low risk was 3.10 (95%CI, 1.02-9.40; P=.046), and HR for high risk vs low risk was 12.30 (95%CI, 4.42-34.25; P <.001). CONCLUSIONS: Our score adequately stratifies the risk of PH in the pediatric population, including infants, and is based on noninvasive variables.