Clinical Scorecard: Employing Unsupervised Machine Learning Techniques to Analyze Cardio-Metabolic Profiles in Middle-Aged and Older Chinese Individuals with General and Central Obesity

At a Glance

Key Highlights

• Obesity is a heterogeneous chronic disease linked to increased cardio-metabolic risks and mortality.
• BMI alone is insufficient to differentiate cardio-metabolic outcomes; central obesity indicators like WHtR provide additional risk stratification.
• Unsupervised machine learning can identify obesity subtypes using clinical and laboratory data to improve precision in diagnosis and management.

Guideline-Based Recommendations

Diagnosis

• Use both BMI and waist-to-height ratio (WHtR) to assess general and central obesity.
• Incorporate clinical indicators such as blood pressure, fasting plasma glucose, lipid profile, and uric acid levels for comprehensive cardio-metabolic risk evaluation.
• Apply unsupervised machine learning clustering techniques on clinical and laboratory data to identify obesity subtypes for precise risk stratification.

Management

• Tailor prevention and treatment strategies based on identified obesity subtypes and associated cardio-metabolic profiles.
• Focus on managing central obesity due to its stronger association with non-communicable diseases.

Monitoring & Follow-up

• Regularly monitor anthropometric measures and cardio-metabolic biomarkers to track disease progression and treatment response.
• Utilize longitudinal cohort data to validate and refine obesity subtype classifications.

Risks

• Recognize that reliance on BMI alone may underestimate cardio-metabolic risks in certain obesity phenotypes.
• Consider heterogeneity in obesity-related health outcomes influenced by genetic, metabolic, and fat distribution factors.

Patient & Prescribing Data

Middle-aged and older Chinese adults with general or central obesity

Data-driven obesity subtyping can inform personalized interventions targeting specific cardio-metabolic risk profiles.

Clinical Best Practices

• Incorporate multiple anthropometric and biochemical markers beyond BMI for obesity assessment.
• Leverage unsupervised machine learning methods to uncover hidden clinical phenotypes in heterogeneous populations.
• Use large, population-based cohort data to enhance generalizability and applicability of obesity subtyping models.

References

• China Health and Retirement Longitudinal Study (CHARLS)