An ensemble-based machine learning model for predicting type 2 diabetes and its effect on bone health

Area of Science:

• Biomedical Informatics
• Endocrinology
• Medical Imaging

Background:

• Diabetes mellitus is a chronic condition associated with significant long-term physiological, metabolic, and neurological complications.
• Early detection of diabetes is crucial for timely diagnosis and effective treatment planning.

Purpose of the Study:

• To develop and validate machine learning models for early diabetes prediction.
• To identify key clinical and bone health indicators for diabetes detection.
• To enhance the explainability of predictive models using SHAP analysis.

Main Methods:

• A case-control study utilizing a diabetic cohort of 1000 participants from Qatar Biobank.
• Machine learning models (XGBoost, RF) were employed to predict diabetes status.
• Recursive Feature Elimination (RFE) was used to optimize feature selection, and SHAP analysis for model interpretability.

Main Results:

• Ensemble ML models achieved over 84% accuracy in diabetes detection, improving to 87.2% with 20 selected features via RFE.
• Key predictors included HDL-Cholesterol, Neutrophil, Vitamin B12, and Testosterone levels.
• While bone density was slightly lower in diabetics, significant differences were only noted in T12, troch, and trunk areas; no long-term negative impact on bone health was observed.

Conclusions:

• Machine learning models integrating Dual Energy X-ray Absorptiometry (DXA) and clinical data offer a promising approach for early diabetes diagnosis.
• The study recommends incorporating such ML models into clinical practice for improved patient outcomes.