Objective: Hyperuricemia (HUA) is a significant comorbidity in pediatric primary hypertension, yet tools for its early prediction are lacking. This study aimed to develop and validate an explainable machine learning (ML) model to predict concurrent HUA in this population, thereby providing actionable insights for the prevention and management of childhood hypertension. Design and method: This two-center retrospective study included 1564 children and adolescents aged 6–17 years with new-onset primary hypertension, who were allocated to a training set (n = 975), an internal validation set (n = 417), and an external validation set (n = 172). Core predictors were identified from 109 variables via a multi-step feature selection process including the least absolute shrinkage and selection operator (LASSO) regression, and SHapley Additive exPlanations (SHAP)-guided feature reduction. Nine ML algorithms were trained and compared, with model performance evaluated using variable metrics, including the area under the receiver operating characteristic curve (AUROC). The generalizability of the optimal model was assessed via external validation. To enhance interpretability, the model was explained using SHAP analysis, and a user-friendly, web-based calculator was developed and deployed. Results: A five-feature Support Vector Machine (SVM) model (incorporating body mass index (BMI) Z-score, sex, creatinine, prealbumin, and age) achieved AUROCs of 0.840 in internal validation set and 0.776 in external validation set. To improve clinical accessibility, a simplified model that intentionally excluded laboratory variables was developed. This three-feature Multilayer Perceptron (MLP) model (BMI Z score, age, and sex) attained AUROCs of 0.801 and 0.775 in the internal and external validation sets, respectively. Both models demonstrated good robustness across subgroups. SHAP analysis confirmed BMI Z-score as the most influential predictor of HUA risk. To facilitate clinical application, web-based risk calculators were deployed. Conclusions: This study developed and validated two explainable ML models for predicting HUA in children with hypertension. We further constructed a convenient and practical online tool that enables early identification of HUA in this population by inputting three routinely available variables: age, sex, and BMI.
Lin et al. (Fri,) studied this question.