Predictive models for diabetes mellitus using machine learning techniques.

Area of Science:

• Medical Informatics
• Biostatistics
• Public Health

Background:

• Diabetes Mellitus is a growing global health concern due to impaired glucose metabolism.
• Early identification of at-risk individuals is crucial for effective management and prevention.
• Canadian healthcare data provides a unique population for diabetes risk prediction research.

Purpose of the Study:

• To develop and validate predictive models for Diabetes Mellitus risk in Canadian patients.
• To achieve high sensitivity and selectivity in identifying individuals prone to developing diabetes.
• To leverage demographic and laboratory data for enhanced diabetes risk assessment.

Main Methods:

• Utilized data from 13,309 Canadian patients (aged 18-90) including demographic and laboratory results.
• Developed predictive models using Logistic Regression and Gradient Boosting Machine (GBM) techniques.
• Evaluated model performance using Area Under the Receiver Operating Characteristic Curve (AROC) and sensitivity, comparing with Decision Tree and Random Forest.

Main Results:

• The GBM model achieved an AROC of 84.7% with 71.6% sensitivity; Logistic Regression achieved 84.0% AROC with 73.4% sensitivity.
• Both GBM and Logistic Regression models outperformed Random Forest and Decision Tree models in predictive accuracy.
• Key predictors identified include fasting blood glucose, body mass index, high-density lipoprotein, and triglycerides.

Conclusions:

• The developed models demonstrate high predictive ability and satisfactory sensitivity for identifying patients with Diabetes Mellitus.
• These models can be integrated into clinical decision support tools for early intervention and prevention strategies.
• The models are specifically validated for the Canadian population, offering greater relevance than models from other demographics.