Exploration of Machine Learning for Hyperuricemia Prediction Models Based on Basic Health Checkup Tests

Area of Science:

• Medical Informatics
• Computational Biology
• Health Data Science

Background:

• Machine learning (ML) shows potential for healthcare predictions but lacks clinical data validation.
• Hyperuricemia, a marker for chronic diseases, requires effective prediction methods.
• This study explores ML for predicting uric acid status from routine health checkup data.

Purpose of the Study:

• To predict hyperuricemia using various ML algorithms.
• To evaluate and compare the performance of ML models against conventional methods.
• To assess the utility of ML in clinical data analysis for disease biomarkers.

Main Methods:

• A prediction model for hyperuricemia was developed using a health checkup database.
• ML algorithms including discrimination analysis, K-nearest neighbor, Naïve Bayes (NBC), support vector machine, decision tree, and Random Forest Classification (RFC) were employed.
• Performance was evaluated and compared to conventional logistic regression (CLR) using receiver operating characteristic curve analysis.

Main Results:

• Naïve Bayes (NBC) achieved the highest sensitivity (0.73), and RFC showed the highest balanced classification rate (BCR) (0.68).
• Both NBC (AUC=0.669) and RFC (AUC=0.775) demonstrated significantly superior performance compared to CLR (AUC=0.568) (p < 0.001).
• RFC and NBC outperformed other ML algorithms in predicting uric acid status.

Conclusions:

• ML models, particularly RFC and NBC, are superior to conventional logistic regression for hyperuricemia prediction.
• Further research is needed to identify optimal ML algorithms based on specific dataset characteristics.
• This study provides valuable insights for applying ML tools in clinical research.