Identification of Potential Type II Diabetes in a Large-Scale Chinese Population Using a Systematic Machine Learning...

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Public Health

Background:

• Diabetes mellitus affects over 425 million globally, incurring significant healthcare costs.
• The growing prevalence of diabetes poses a substantial burden, particularly in remote and underserved regions.
• Early detection and prevention are crucial to mitigate the impact of diabetes.

Purpose of the Study:

• To develop and compare machine learning models for identifying individuals at risk of type II diabetes mellitus (T2DM).
• To identify key risk factors and their importance in predicting T2DM.
• To propose a noninvasive, accurate classifier for early T2DM risk screening.

Main Methods:

• Logistic regression (LR) was used to identify initial risk factors from physical measurements and questionnaire data.
• Four machine learning classifiers (Decision Tree, Random Forest, AdaBoost, XGBoost) were trained and compared.
• The best-performing classifier was used to determine the importance of various risk factors for T2DM.

Main Results:

• Extreme Gradient Boosting (XGBoost) demonstrated superior performance with an accuracy of 0.906.
• Body Mass Index (BMI) was identified as the most significant predictor of T2DM.
• Other key predictors included age, waist circumference, systolic pressure, and lifestyle factors.

Conclusions:

• A novel LR-XGBoost classifier utilizing fourteen easily obtainable, noninvasive variables can accurately screen for T2DM risk.
• This classifier facilitates early detection of potential diabetes cases.
• The identified variable importance scores provide insights for targeted diabetes prevention strategies.