Development of machine learning predictive model for type 2 diabetic retinopathy using the triglyceride-glucose index...

Area of Science:

• Ophthalmology
• Endocrinology
• Data Science

Background:

• Diabetic retinopathy (DR) is a leading cause of vision loss.
• Accurate prediction models are crucial for early intervention.
• Assessing insulin resistance (IR) can be challenging in resource-limited settings.

Purpose of the Study:

• To develop and validate a machine learning model for diabetic retinopathy prediction.
• To incorporate the triglyceride-glucose index (TyG) as a novel predictor for DR.
• To interpret the model using SHapley Additive exPlanations (SHAP) for clinical insights.

Main Methods:

• Utilized real-world data from 1392 DR and 2358 non-DR patients.
• Employed Recursive Feature Elimination with Cross-Validation (RFECV) for variable selection.
• Evaluated Logistic Regression (LR), Decision Tree (DT), Random Forest (RF), Support Vector Machine (SVM), and XGBoost (XGB) models.

Main Results:

• Logistic Regression (LR) and Support Vector Machine (SVM) models showed superior performance (ROC AUC 0.85 and 0.82).
• Top predictors identified by SHAP analysis included TyG index, Insulin therapy, HbA1c, Diabetes Course, and HDL.
• HDL was a protective factor, while TyG, Insulin therapy, HbA1c, and Diabetes Course were associated with increased DR risk.

Conclusions:

• Developed the first machine learning DR prediction model integrating the TyG index.
• The TyG index offers a cost-effective alternative for IR assessment in primary care.
• This model facilitates practical DR risk stratification, especially in resource-limited environments.