Health Data-Driven Machine Learning Algorithms Applied to Risk Indicators Assessment for Chronic Kidney Disease

Area of Science:

• Nephrology
• Public Health
• Data Science

Background:

• Global population aging increases the burden of chronic diseases.
• Taiwan faces a growing prevalence of chronic kidney disease (CKD), particularly among older adults.
• Effective health management strategies for CKD are crucial for public health initiatives.

Purpose of the Study:

• To identify significant risk factors for chronic kidney disease (CKD) stages G3a, G3b, and G4.
• To evaluate the performance of five different prediction models in identifying CKD risk.
• To investigate the influence of socioeconomic status on CKD risk.

Main Methods:

• Analysis of annual health screening data from 65,394 individuals in Taiwan (2010-2015).
• Utilized five prediction models: logistic regression (LR), C5.0 decision tree, stochastic gradient boosting (SGB), multivariate adaptive regression splines (MARS), and eXtreme gradient boosting (XGboost).
• Included 18 risk indicators and estimated glomerular filtration rate (e-GFR) data for risk factor determination.

Main Results:

• LR, SGB, and XGboost models demonstrated superior and comparable classification performance (AUCs ranging from 0.848 to 0.858).
• Blood urea nitrogen (BUN) and uric acid (UA) were consistently identified as the top two most significant CKD risk factors across all models.
• Education level emerged as the third most important indicator in the best-performing models, showing a significant negative correlation with CKD.

Conclusions:

• The five prediction models offer high and similar classification performance for CKD risk assessment.
• Education level, a socioeconomic factor, is a critical and significant predictor of CKD.
• Findings support integrating socioeconomic factors into CKD prevention and management strategies.