A machine learning-based framework for predicting type 2 diabetes mellitus using hematological indices

Area of Science:

• Biomedical Informatics
• Clinical Prediction Modeling
• Hematology

Background:

• Type 2 diabetes mellitus (T2DM) is a growing global health concern.
• Early identification of individuals at high risk for T2DM is crucial for timely intervention.
• Conventional risk factors may not fully capture the complexity of T2DM development.

Purpose of the Study:

• To develop and validate a machine learning (ML) framework for predicting incident T2DM.
• To evaluate the predictive value of routine hematological and renal biomarkers.
• To compare ML model performance against traditional clinical risk factors.

Main Methods:

• Analysis of data from 6093 diabetes-free participants in the MASHAD cohort.
• Inclusion of hematological parameters (RBC, WBC, RDW) and renal biomarkers.
• Application of logistic regression and ML models (Random Forest, XGBoost, LightGBM) with cross-validation.

Main Results:

• RBC, WBC, and RDW were identified as independent predictors of T2DM.
• The Random Forest model demonstrated the highest predictive performance (ROC-AUC 0.73).
• Metabolic syndrome, BMI, uric acid, and age were key predictors in the ML models.

Conclusions:

• Routine hematological indices are significantly associated with incident T2DM.
• ML offers a valuable tool for early T2DM risk stratification.
• Further research is needed to validate these ML models for clinical use.