A data-driven machine learning algorithm to predict the effectiveness of inulin intervention against type II diabetes

Area of Science:

• Endocrinology and Metabolism
• Nutritional Science
• Computational Biology

Background:

• Rising incidence of type 2 diabetes mellitus (T2DM) necessitates advanced management strategies.
• Nutritional therapy, particularly inulin supplementation, is a key component in T2DM care.
• Identifying suitable T2DM patients for inulin intervention requires predictive tools.

Purpose of the Study:

• To develop and validate a machine learning model for predicting inulin treatment effectiveness in T2DM patients.
• To identify key patient characteristics influencing the response to inulin therapy.

Main Methods:

• Utilized data from a previous study on T2DM patients undergoing inulin intervention.
• Employed LASSO regression for feature selection and XGBoost for predictive model development.
• Evaluated model performance using accuracy, specificity, positive/negative predictive values, ROC, calibration, and decision curves.

Main Results:

• Inulin intervention successfully reduced glycated hemoglobin (HbA1c) in 62.93% of 758 T2DM patients.
• Key predictors identified by LASSO regression included HbA1c, glucose variability, fasting glucose, HDL, age, and BMI.
• The XGBoost model achieved high performance metrics, with training set accuracy of 0.819 and testing set accuracy of 0.709.

Conclusions:

• The XGBoost-SHAP framework effectively predicts inulin intervention outcomes in T2DM.
• This approach enables personalized treatment by assessing individual patient features and prediction abilities.
• Establishes a valuable link between machine learning and nutritional therapy for T2DM management.