Machine learning-based predictive model for prevention of metabolic syndrome

Area of Science:

• Cardiology
• Endocrinology
• Public Health

Background:

• Metabolic syndrome (MetS) is a cluster of conditions including obesity, high blood pressure, high blood sugar, and dyslipidemia.
• Early detection and prevention of MetS are crucial to mitigate risks of cardiovascular disease and type 2 diabetes.
• Current MetS detection often requires invasive blood tests, limiting accessibility for daily monitoring.

Purpose of the Study:

• To develop a predictive model for MetS using only noninvasive health information.
• To create a practical, easily accessible tool for early MetS detection and risk assessment.
• To enhance MetS management through simplified, real-world health monitoring.

Main Methods:

• Utilized a large-scale Korean health examination dataset (n = 70,370).
• Developed a predictive model excluding blood test parameters (triglycerides, blood sugar, HDL cholesterol).
• Engineered novel synthetic features from waist circumference, blood pressure, and gender; employed decision tree algorithm.

Main Results:

• The decision tree model demonstrated strong performance with an AUC of 0.889, recall of 0.855, and specificity of 0.773.
• The model effectively predicts MetS using only four basic, noninvasive features, ensuring simplicity and interpretability.
• Calibrated prediction probabilities and introduced a MetS risk map for user-friendly risk assessment.

Conclusions:

• The proposed noninvasive model offers a practical and accessible method for MetS diagnosis and risk prediction.
• This simplified approach facilitates early intervention and management of metabolic syndrome.
• The MetS risk map empowers individuals to easily monitor their health status and understand their metabolic syndrome risk.