Discovering the Type 2 Diabetes in Electronic Health Records Using the Sparse Balanced Support Vector Machine

Area of Science:

• Medical Informatics
• Machine Learning
• Computational Biology

Background:

• Early diagnosis of type 2 diabetes (T2D) is essential for effective patient management.
• The increasing volume of electronic health record (EHR) data presents challenges for traditional machine learning (ML) models, including overfitting, interpretability, and computational cost.

Purpose of the Study:

• To introduce a novel ML method, sparse balanced support vector machine (SB-SVM), for early T2D detection using EHR data.
• To evaluate the performance of SB-SVM against existing ML and deep learning techniques for T2D diagnosis.

Main Methods:

• Collected a novel EHR dataset (Federazione Italiana Medici di Medicina Generale) focusing on exemptions, examinations, and drug prescriptions prior to T2D diagnosis in a uniform age group.
• Developed and applied the SB-SVM algorithm, a sparse and balanced approach, to model the selected EHR features.
• Compared SB-SVM performance against state-of-the-art ML and deep learning methods.

Main Results:

• The SB-SVM method demonstrated superior performance compared to other state-of-the-art ML and deep learning approaches for T2D discovery.
• SB-SVM achieved the best balance between predictive accuracy and computational efficiency.
• The induced sparsity in SB-SVM enhanced model interpretability and effectively handled high-dimensional, imbalanced datasets.

Conclusions:

• SB-SVM offers a reliable and efficient method for early T2D diagnosis using EHR data.
• The approach addresses key challenges in ML modeling, including interpretability and computational cost.
• SB-SVM provides a promising tool for integrated T2D management systems.