Related Experiment Videos
Early Prediction of Diabetic Macular Edema via Machine Learning Survival Analysis on Checkup Data
Yusuke Kashiwagi1,2, Katsuyuki Chida1,3, Ayaka Hananoe3,4
1Department of Artificial Intelligence Medicine, Graduate School of Medicine, Chiba University, Chiba, Japan.
Ophthalmology Science
|July 12, 2026
Summary
Predicting diabetic macular edema (DME) risk is possible using health checkup data. A random survival forest model identified key risk factors and subgroups, outperforming traditional methods for early detection.
Area of Science:
- Ophthalmology
- Diabetology
- Data Science
Background:
- Diabetic macular edema (DME) is a leading cause of vision loss in diabetic patients.
- Early prediction and risk stratification are crucial for effective management.
Purpose of the Study:
- To predict the risk of diabetic macular edema (DME) onset.
- To identify risk subgroups and their specific predictors.
Main Methods:
- A population-based case-control study using Japanese health checkup and claims data (2005-2020).
- Employed multivariate Cox models, regularized Cox models, and a random survival forest (RSF).
- Utilized 43 health checkup variables and 404 disease histories for risk assessment.
Main Results:
- RSF identified significant associations between checkup items/disease history and DME onset.
- RSF predicted 43.8% of DME cases over 5 years prior with 85.5% specificity.
- RSF demonstrated superior predictive performance (C-index 0.694, AUC 0.750) compared to Cox models.
- Identified three distinct DME risk subgroups with varying predictor importance.
Conclusions:
- Random survival forest (RSF) shows superior performance for DME risk prediction using health checkup data.
- The model can identify high-risk individuals and specific predictors for targeted interventions.
- External validation is necessary before clinical application.
