Research of Prediction Diabetes Risk Using Logistic Regression Models

Abstract

Diabetes has become a major global health concern, with its prevalence steadily rising. Early prediction and prevention are crucial to reducing the disease burden. This study explored the relationship between HbA1c level, age, smoking history, BMI, and diabetes risk by analyzing electronic health records (EHR) of more than 10,000 individuals and logistic regression models. By establishing a prediction model, the impact of these variables on the likelihood of an individual developing diabetes was evaluated. The results showed that the overall accuracy of the logistic regression model reached 89%, and the ROC-AUC score was as high as 0.9624, showing excellent discrimination between diabetic and non-diabetic cases. Among them, HbA1c level (coefficient 2.49), blood glucose concentration (coefficient 1.3), and age (coefficient 1.16) were confirmed to be key predictors for diabetes diagnosis, especially HbA1c level (coefficient 2.49) was the most influential factor. The study also discussed the potential limitations of the model performance and future improvement directions.