Development of Chronic Disease Risk Prediction Models
Subproject One: Development of Chronic Disease Risk Prediction Models
Chronic diseases such as diabetes, chronic lung disease, cardiovascular disease, Parkinson’s disease, and liver diseases are significant challenges in modern healthcare. This project leverages health check-up data and artificial intelligence (AI) technologies to develop various risk prediction models, facilitating early screening and personalized health management. Below are the key research achievements and applications:
1. Diabetes Risk Prediction Model
We developed a 5-year diabetes risk prediction model targeting prediabetic and non-diabetic populations. Using machine learning algorithms such as XGBoost, the model achieved over 80% accuracy. The three most critical predictors identified are glycated hemoglobin (HbA1c), fasting blood glucose, and postprandial blood glucose. This model enables healthcare teams to identify high-risk individuals early, offering lifestyle recommendations and medical interventions.
2. Chronic Lung Disease Risk Prediction Model
Based on long-term lung function data from non-smokers, we developed a risk prediction model for chronic obstructive pulmonary disease (COPD). Key risk factors include total cholesterol, low-density lipoprotein (LDL), and lung function metrics such as FEV1/FVC. With an accuracy rate of 85%, the model provides early diagnosis and preventive strategies, enhancing patient quality of life.
3. Coronary Artery Disease Risk Prediction Model
By analyzing coronary artery calcium (CAC) scores, we created a model to predict cardiovascular hospitalization risk. Age, hypertension, and CAC scores emerged as key factors. The model, with an accuracy of 83%, aids healthcare professionals in formulating treatment plans for high-risk patients, preventing severe cardiovascular events.
4. Parkinson’s Disease Risk Prediction Model
Utilizing smartphone-recorded voice and facial expression data, we integrated biological features to perform early screening for Parkinson’s disease. Key indicators include speech characteristics (e.g., speed and pitch) and facial movement patterns (e.g., mouth and eye dynamics). The model achieved 90% accuracy and has been validated with international datasets, marking a breakthrough in the early detection of neurodegenerative diseases.
5. Liver Disease Risk Prediction Model
Focusing on high-risk factors for liver cancer, such as fatty liver and liver fibrosis, we developed risk screening models using health check-up data. The models achieved over 85% accuracy. Key predictors include BMI, triglycerides (TG), and liver function markers such as ALT. These models assist in early intervention for high-risk individuals, reducing the risk of disease progression.
6. HBV and HCC-Related Research
For chronic hepatits B (HBV) patients, we developed a risk assessment system based on HBcrAg to predict liver cancer (HCC) risk. The model achieved an AUC exceeding 0.90, significantly outperforming traditional risk scoring systems. This tool helps identify high-risk patients and prioritize antiviral treatments, effectively lowering HCC incidence rates.
Applications and Future Prospects
These risk prediction models provide scientific support for the early screening and management of chronic diseases. They not only assist medical professionals in diagnosis but also empower individuals to understand their health risks through check-up data and take preventive measures. Looking ahead, we aim to enhance the accuracy of these models and expand their adoption in more healthcare facilities, advancing preventive medicine and benefiting broader populations.
