Exploratory Machine Learning-Based Classification of Type 2 Diabetes Using Routine Clinical Parameters: A...

Area of Science:

• Endocrinology
• Medical Informatics
• Machine Learning

Background:

• Type 2 diabetes mellitus (T2DM) is a widespread metabolic disorder with significant long-term health consequences.
• Routine clinical data, including anthropometric, biochemical, and hematological variables, may offer valuable discriminatory information for T2DM classification.
• Data-driven approaches using machine learning can potentially enhance diagnostic capabilities.

Purpose of the Study:

• To compare the classification performance of various machine learning algorithms for distinguishing individuals with and without T2DM.
• To evaluate the utility of routinely collected clinical parameters in T2DM classification.
• To assess the impact of HbA1c inclusion on model performance.

Main Methods:

• An observational, single-center study involving 160 adult patients from an Endocrinology Outpatient Clinic.
• Utilized a dataset comprising anthropometric measurements, biochemical markers, and complete blood count parameters.
• Applied stratified 5-fold cross-validation and SHAP analysis to evaluate eight supervised machine learning models (with and without HbA1c).

Main Results:

• Random Forest achieved the highest classification performance, with 92.2% accuracy without HbA1c and 98.0% accuracy when HbA1c was included.
• Tree-based models generally outperformed linear classifiers.
• Including HbA1c significantly improved model accuracy, sensitivity, specificity, and F1 scores.

Conclusions:

• Machine learning models utilizing routine clinical and anthropometric data show potential for T2DM classification.
• Tree-based algorithms, especially Random Forest, demonstrated the most promising results.
• While HbA1c inclusion enhanced classification, potential target leakage necessitates cautious interpretation and external validation before clinical application.