Machine Learning-Based Prediction of Gout Using Polygenic Risk Scores and Clinical Variables: A Korean Cohort Study.

Area of Science:

• Genetics and genomics
• Computational biology
• Epidemiology

Background:

• Gout prevalence is increasing globally.
• Polygenic risk scores (PRS) show potential for predicting gout outcomes.
• The clinical utility of PRS in disease prediction requires further investigation.

Purpose of the Study:

• To develop machine learning (ML) models for gout prediction using genetic and clinical data.
• To evaluate the performance of different ML algorithms in gout prediction.
• To assess the contribution of PRS and traditional risk factors to gout prediction.

Main Methods:

• Utilized data from the Korean Genome and Epidemiology Study.
• Developed and compared five supervised ML models: logistic regression, random forest (RF), decision tree, extreme gradient boosting, and light gradient boosting.
• Included polygenic risk scores (PRS), uric acid, lifestyle habits, and metabolic syndrome (MetS) profiles as predictors.

Main Results:

• The RF model integrating PRS, age, sex, MetS, and uric acid achieved the highest predictive performance (AUC = 0.7204).
• Uric acid levels were identified as the most significant predictor, followed by PRS and age.
• PRS demonstrated a modest but positive impact on the predictive power of ML models for gout.

Conclusions:

• Integrating genetic data (PRS) with clinical variables enhances gout prediction accuracy.
• Traditional risk factors, particularly uric acid levels, remain highly important for gout prediction.
• Further research is needed to optimize PRS utility across diverse populations for effective gout management.