A computational framework of routine test data for the cost-effective chronic disease prediction

Area of Science:

• Biomedical Informatics
• Machine Learning in Healthcare
• Public Health

Background:

• Chronic diseases pose a significant global health burden due to insidious onset and long latency.
• Current diagnostic methods relying on genetic markers or imaging are costly and lack universal accessibility.

Purpose of the Study:

• To develop a cost-effective, high-accuracy framework for chronic disease prediction using routine blood test data.
• To create predictive models for various chronic diseases, including cancers, cardiovascular diseases, and mental illnesses.
• To build an explainable AI-driven clinical decision support system for chronic disease management.

Main Methods:

• Analysis of massive routine blood and biochemical test data from 32,448 patients.
• Development of 20 classification models using the XGBoost algorithm for 17 chronic disease types.
• Model interpretation using the SHAP algorithm to identify key predictive biomarkers.

Main Results:

• Achieved high prediction accuracy (AUC 87.32%) for chronic diseases.
• Highest accuracy of 90.13% for cardia cancer and lowest of 76.38% for rectal cancer.
• Identified key indices like CREA, R-CV, GLU, NEUT%, and PDW for disease identification and classification.
• Revealed specific biomarker associations (e.g., R-CV for cancer, ALP for cardiovascular disease, GLU for mental illness).

Conclusions:

• A novel, cost-effective machine learning framework enables accurate chronic disease prediction from simple blood tests.
• The developed system (DisPioneer) can assist clinicians in predicting, classifying, and treating chronic diseases.
• This approach facilitates wider clinical implementation for chronic disease prevention and surveillance programs.