Building Risk Prediction Models for Type 2 Diabetes Using Machine Learning Techniques

Area of Science:

• Medical Informatics
• Public Health
• Machine Learning

Background:

• Type 2 diabetes is a prevalent chronic disease in the US, causing significant health and financial burdens.
• Early diagnosis and intervention are crucial for managing type 2 diabetes and reducing associated costs.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting type 2 diabetes risk.
• To identify key risk factors for type 2 diabetes, including novel indicators.

Main Methods:

• Analysis of cross-sectional data from 138,146 participants in the 2014 Behavioral Risk Factor Surveillance System.
• Development and comparison of six machine learning models: support vector machine, decision tree, logistic regression, random forest, neural network, and Gaussian Naive Bayes.
• Investigation of risk factor associations using univariable and multivariable weighted logistic regression.

Main Results:

• All predictive models demonstrated high predictive power, with Area Under the Curve (AUC) values ranging from 0.7182 to 0.7949.
• The neural network model achieved the highest accuracy (82.4%) and AUC (0.7949), while the decision tree model offered the highest sensitivity (51.6%).
• Sleeping 9 or more hours daily (aOR=1.13) and checkup frequency of less than 1 year (aOR=2.31) were identified as significant risk factors.

Conclusions:

• Machine learning models, particularly neural networks, show strong performance in predicting type 2 diabetes.
• The decision tree model is recommended for initial screening due to its high sensitivity.
• Sleep duration and checkup frequency are identified as novel risk factors for type 2 diabetes, complementing established ones.