An Interactive Online App for Predicting Diabetes via Machine Learning from Environment-Polluting Chemical Exposure...

Area of Science:

• Biostatistics
• Environmental Health
• Computational Biology

Background:

• Early identification of diabetes risk factors is crucial for prevention and delaying disease progression.
• Traditional risk factors alone may not fully capture an individual's susceptibility to diabetes.
• Environmental factors, such as exposure to environment-polluting chemicals (EPCs), may play a significant role.

Purpose of the Study:

• To develop an interactive online application for predicting prediabetes and diabetes risk.
• To utilize a Bayesian network (BN) classifier for interpretable machine learning in diabetes prediction.
• To assess the impact of serum biomarkers for EPC exposure on diabetes risk prediction.

Main Methods:

• A Bayesian network (BN) classifier was trained using data from the Korean Genome and Epidemiological Study (KoGES) Ansung cohort (2008-2012).
• Feature selection identified 11 key variables, including traditional risk factors and EPC biomarkers, for a tree-augmented BN model.
• An interactive online application was developed to provide personalized diabetes risk probabilities and simulate risk factor control effects.

Main Results:

• The developed BN model demonstrated good predictive performance, with accuracy and AUC metrics.
• Serum biomarkers quantifying exposure to environment-polluting chemicals (EPCs) were found to have interactive effects on diabetes progression.
• Incorporating EPC biomarkers significantly improved the prediction performance for diabetes risk.

Conclusions:

• The interactive online application provides a valuable tool for customized diabetes risk prediction.
• Environmental chemical exposure, as indicated by serum biomarkers, is an important factor in diabetes development.
• The integration of EPC biomarkers enhances the accuracy of diabetes risk prediction models.