Development of a 5-Year Risk Prediction Model for Transition From Prediabetes to Diabetes Using Machine Learning:...

Area of Science:

• Diabetes research
• Machine learning in healthcare
• Public health interventions

Background:

• Diabetes is a global health crisis, with prediabetes representing a key stage for intervention.
• A 5-year risk prediction model for diabetes progression in Chinese prediabetes patients is lacking.
• Early identification and intervention in prediabetes can mitigate diabetes incidence and healthcare burdens.

Purpose of the Study:

• To develop and validate a machine learning-based 5-year risk prediction model for prediabetes to diabetes progression in the Chinese population.
• To create an interactive web-based platform for identifying high-risk individuals and facilitating early interventions.
• To reduce overall diabetes incidence and associated healthcare costs.

Main Methods:

• A retrospective cohort study involving two Chinese prediabetes cohorts (n=6578 and n=2333) from 2019-2024.
• Utilized 42 demographic, physical, and hematologic variables, applying recursive feature elimination and seven machine learning algorithms, including CatBoost.
• Optimized models using grid search and cross-validation, assessing performance with ROC curves, precision-recall curves, accuracy, sensitivity, and specificity.

Main Results:

• The CatBoost model, using 14 selected features, demonstrated optimal performance with an AUC of 0.819 (test set) and 0.807 (external test set).
• The model exhibited superior discrimination and calibration, with key predictors including fasting blood glucose (FBG), HDL, ALT/AST, BMI, age, and MONO.
• Annual progression rates from prediabetes to diabetes were 8.34% and 7.04% in the primary and external cohorts, respectively.

Conclusions:

• A robust 5-year risk prediction model for prediabetes to diabetes progression was successfully developed for the Chinese population.
• The CatBoost model exhibited the highest predictive performance, effectively identifying individuals at high risk of developing diabetes.
• The developed model and identified predictors offer a pathway for targeted interventions to prevent diabetes.