A cohort study on the predictive capability of body composition for diabetes mellitus using machine learning

Area of Science:

• Biomedical Informatics
• Metabolic Disease Research
• Data Science in Healthcare

Background:

• Diabetes mellitus is a global health concern with complex etiological factors.
• Regional body fat distribution, beyond overall obesity, is increasingly recognized as a significant contributor to metabolic dysfunction.
• Understanding these regional fat patterns can improve diabetes risk prediction and management.

Purpose of the Study:

• To investigate the predictive capability of regional body fat distribution for diabetes mellitus using machine learning.
• To identify specific body fat metrics and their associations with diabetes status in a community adult population.
• To compare the performance of various machine learning classifiers in predicting diabetes.

Main Methods:

• Retrospective analysis of the Fasa cohort study data.
• Utilized segmental body composition measurements (e.g., fat-free mass, fat percentage, limb and trunk fat) as input features.
• Employed diverse machine learning algorithms, including individual classifiers (SVM, Decision Tree) and ensemble methods (Random Forest, XGBoost).

Main Results:

• Positive associations were found between diabetes and fat mass/percentages in legs, arms, and trunk.
• Negative associations were observed between diabetes and fat-free mass in the same regions.
• The XGBoost model achieved the highest performance with accuracy, precision, recall, and F1-score exceeding 89%.

Conclusions:

• Regional body fat composition is a significant predictor of diabetes status.
• Machine learning models can effectively leverage body fat distribution data for diabetes risk assessment.
• These findings support the integration of body composition analysis into diabetes screening and prevention strategies.