Machine Learning Prediction and SHAP Interpretability Analysis of Heart Failure Risk in Patients with Hyperuricemia

Tian-ming GanShi-rong WangGuan-lian MoShu-hu LiYong-qi LuJin-yi Li^*

• Affiliated Hospital of Guilin Medical University, Guilin, China

Aims: Cardiovascular disorders, particularly heart failure (HF), are a critical global health challenge. Hyperuricemia, a key cardiovascular risk factor, significantly increases HF susceptibility. Existing HF risk prediction tools are often cumbersome, relying on extensive clinical parameters and tests, limiting their practical use. Therefore, there is an urgent need for a simple, interpretable model to assess HF risk in hyperuricemia patients. Methods and results: Using 2005–March 2020 NHANES data (85,750 participants), 1,603 adults (≥18 years) with confirmed hyperuricemia were included. Accessible multidimensional indicators were selected for routine clinical use. Multiple machine learning models (Random Forest, Logistic Regression, XGBoost, SVM, etc.) were applied, with performance evaluated via accuracy, sensitivity, F1-score, and ROC AUC, etc. SHAP values analyzed feature importance for the best model. The SVM model showed the best overall performance, with chronic kidney disease, coronary heart disease, hypertension, serum potassium, serum osmolality, and sedentary time emerging as the top predictors. These six indicators demonstrated strong predictive power for HF in hyperuricemia patients, highlighting their clinical relevance. Conclusion: Integration of these six readily available indicators provides a simple, interpretable tool for HF risk stratification in hyperuricemia patients. While further longitudinal and multicenter validation is needed, the model shows promise for early identification and targeted intervention in clinical practice.