Prediction of Glucose Metabolism Disorder Risk Using a Machine Learning Algorithm: Pilot Study.

Area of Science:

• Endocrinology and Metabolism
• Computational Biology and Bioinformatics
• Medical Diagnostics

Background:

• The 75-g oral glucose tolerance test (OGTT) is valuable for assessing glucose metabolism but is invasive and costly.
• Complete OGTT data, including 1- and 2-hour postload glucose and insulin levels, can predict future diabetes and glucose metabolism disorders (GMD).

Purpose of the Study:

• To develop and compare machine learning models, particularly XGBoost, against logistic regression (LR) for predicting diabetes and GMD risk.
• To evaluate the impact of including comprehensive OGTT data on prediction accuracy.

Main Methods:

• Trained multiple classification models using XGBoost and logistic regression (LR) on a large dataset of OGTTs from medical check-ups.
• Utilized data from 13,581 OGTTs for diabetes risk prediction (Study 1) and 6760 for GMD risk prediction (Study 2).
• Compared model performance using receiver operating characteristic (ROC) curves and area under the curve (AUC) values.

Main Results:

• XGBoost models demonstrated superior performance over traditional LR methods in both studies (AUCs ranging from 0.63 to 0.93).
• Prediction accuracy significantly improved when all OGTT variables were incorporated into the models.
• XGBoost analysis highlighted the importance of OGTT variables over fasting plasma glucose or glycated hemoglobin for risk prediction.

Conclusions:

• Advanced machine learning techniques like XGBoost offer enhanced accuracy for early detection of diabetes and GMD.
• Complete OGTT data is crucial for precise prediction of future diabetes and GMD risk.
• XGBoost represents a promising tool for improving diabetes risk assessment.