Predicting diabetic retinopathy based on biomarkers: Classification and regression tree models

Area of Science:

• Biochemistry
• Ophthalmology
• Machine Learning

Background:

• Diabetic retinopathy (DR) is a significant complication of diabetes mellitus (DM).
• Identifying early clinical indicators for DR risk is crucial for timely intervention.
• Existing methods may not fully capture the complex interactions of risk factors.

Purpose of the Study:

• To identify key clinical indicators for diabetic retinopathy (DR) risk assessment.
• To determine the critical risk factors associated with DR using machine learning.
• To explore high-order interactive effects of biomarkers in DR prediction.

Main Methods:

• Classification and Regression Tree (CART) models were employed for analysis.
• Logistic regression was used to assess risk factors in 781 diabetic mellitus patients.
• Machine learning (CART) identified interactive effects and predicted DR.

Main Results:

• 11 critical clinical indicators were identified from 96 initial tests.
• Specific thresholds for urine creatinine (Ucr) and cortisol significantly correlated with DR prevalence.
• Elevated C-reactive protein (CRP) also increased DR likelihood, even with normal Ucr and cortisol levels.

Conclusions:

• Routine serum biochemistry markers (Ucr, cortisol, CRP) and their thresholds can help identify DR risk factors.
• These markers provide a useful reference for preliminary DR severity assessment in diabetic patients.
• Findings suggest kidney impairment, insulin resistance, and inflammation contribute to DR development.