Chronic Disease Prediction Using the Common Data Model: Development Study.

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Preventive Medicine

Background:

• Chronic disease management is a global health challenge.
• Preventive medicine emphasizes early disease prediction.
• Machine learning aids in accurate medical judgment for chronic conditions.

Purpose of the Study:

• Develop high-performance machine learning models for predicting four major chronic diseases.
• Utilize the common data model (CDM) for model development.
• Assess the potential for extending these predictive models.

Main Methods:

• Selected diabetes, hypertension, hyperlipidemia, and cardiovascular disease for prediction within a 10-year timeframe.
• Employed the Atlas analysis tool for data extraction from the CDM.
• Built and compared prediction models using logistic regression, gradient boosting, random forest, and extreme gradient boosting algorithms with grid search optimization.

Main Results:

• All four machine learning algorithms achieved over 80% accuracy in predicting the selected chronic diseases.
• Extreme gradient boosting demonstrated the highest predictive performance across all four diseases.
• Area under the curve (AUC) values ranged from 0.84 to 0.93, indicating strong predictive power.

Conclusions:

• Machine learning models using the CDM can effectively predict chronic disease occurrence.
• These models facilitate preemptive management of chronic diseases.
• Risk stratification for major chronic diseases is possible by identifying health risk factors using real-world data.