Machine learning for predicting diabetes risk in western China adults

Area of Science:

• Endocrinology and Metabolism
• Public Health
• Data Science in Healthcare

Background:

• Diabetes mellitus is a global epidemic associated with chronic tissue damage due to prolonged hyperglycemia.
• Early diagnosis and screening are critical for managing diabetes and improving population health outcomes.
• Large-scale data analysis is essential for developing effective diabetes risk prediction models.

Purpose of the Study:

• To establish a robust risk prediction model for type-2 diabetes mellitus using comprehensive national physical examination data.
• To identify key indicators contributing to type-2 diabetes risk in a diverse population.
• To develop a user-friendly diabetes risk score card for population-level screening.

Main Methods:

• Analysis of over 4 million national physical examination records from Xinjiang, China (2020).
• Utilized questionnaire data, routine physical examinations, and laboratory values.
• Employed integrated learning, deep learning (XGBoost), and logistic regression to build the risk model.

Main Results:

• The XGBoost-based risk prediction model achieved a high AUC of 0.9122, outperforming other algorithms.
• Key predictors identified include hypertension, fasting blood glucose, age, coronary heart disease, ethnicity, parental diabetes, triglycerides, waist circumference, total cholesterol, and BMI.
• A logistic regression-based risk score card was developed for practical population risk assessment.

Conclusions:

• A novel, multi-ethnic diabetes risk assessment model was developed using a large dataset and diverse indices.
• The model provides a valuable tool for screening type-2 diabetes patients and informing prevention strategies.
• This research offers a new forecasting method to classify population diabetes risk and guide public health interventions.