Identifying Glucose Metabolism Status in Nondiabetic Japanese Adults Using Machine Learning Model with Simple...

Area of Science:

• Metabolic Health
• Machine Learning in Healthcare
• Preventive Medicine

Background:

• Glucose metabolism status is crucial for identifying individuals at risk of developing diabetes.
• Current methods for assessing glucose metabolism often require clinical procedures like the oral glucose tolerance test (OGTT).
• Developing non-invasive, accessible methods for assessing glucose metabolism is essential for widespread screening.

Purpose of the Study:

• To develop and validate a machine learning model for identifying glucose metabolism categories in nondiabetic Japanese adults using questionnaire data.
• To determine the key lifestyle and physical characteristics predictive of different glucose metabolism statuses.
• To assess the model's performance in classifying individuals into distinct glycometabolic categories.

Main Methods:

• A cross-sectional study involving 977 nondiabetic Japanese adults (aged 20-64) who completed an OGTT and a lifestyle questionnaire.
• Classification of participants into four glycometabolic categories: best glucose metabolism, low insulin sensitivity, low insulin secretion, and combined characteristics.
• Development of machine learning models (random forest, XGBoost, etc.) using questionnaire data, followed by feature selection and external validation on a separate dataset.

Main Results:

• The study identified four glycometabolic categories with 46% in the best category, 21% in low insulin sensitivity, 14% in low insulin secretion, and 19% in combined categories.
• A random forest model utilizing the top 10 most important variables achieved areas under the receiver operating characteristic curve (AUCs) ranging from 0.61 to 0.70 for classifying different categories.
• External validation in a separate cohort demonstrated the model's generalizability, with AUCs ranging from 0.57 to 0.66.

Conclusions:

• A machine learning model effectively identifies glucose metabolism status in nondiabetic adults using a limited set of 10 lifestyle and physical factors.
• This predictive model offers a practical tool for the general population to understand their metabolic health without invasive testing.
• The findings support the potential of this model to promote early awareness and encourage lifestyle modifications for diabetes prevention.