Uncovering Nonlinear Predictors of Serum Biomarker Uric Acid Using Interpretable Machine Learning in Healthy Men

Area of Science:

• Metabolic Health
• Biostatistics
• Machine Learning Applications

Background:

• Uric acid (UA) is associated with gout, kidney dysfunction, and cardiovascular disease.
• Previous research often assumes linear relationships, potentially missing complex physiological interactions.
• Understanding nonlinear associations is crucial for accurate risk assessment.

Purpose of the Study:

• To investigate nonlinear, threshold-dependent relationships between uric acid and various health indicators in healthy Taiwanese men.
• To compare the efficacy of traditional linear models with interpretable machine learning (MARS) for metabolic analysis.
• To identify specific thresholds for key variables impacting uric acid levels.

Main Methods:

• Analysis of data from 5200 healthy Taiwanese men.
• Application of Pearson correlation, multiple linear regression (MLR), and multivariate adaptive regression splines (MARS).
• MARS was used to detect nonlinear and threshold-based effects, offering enhanced interpretability.

Main Results:

• MARS identified significant nonlinear associations not found by linear models.
• Threshold effects were observed for waist-to-hip ratio (WHR < 0.969), creatinine (> 0.97 mg/dL), calcium (> 9.5 mg/dL), and hs-CRP (> 3.38 mg/L).
• Betel nut exposure showed a complex, nonlinear association with UA metabolism, missed by linear analysis.

Conclusions:

• Interpretable machine learning (MARS) effectively reveals critical nonlinear, threshold-dependent relationships impacting uric acid.
• Findings highlight the limitations of linear models in metabolic research.
• Identified thresholds provide valuable data for precise clinical risk stratification and management of uric acid levels.