Predictive model and risk analysis for diabetic retinopathy using machine learning: a retrospective cohort study in...

Area of Science:

• Ophthalmology
• Endocrinology
• Data Science

Background:

• Diabetic retinopathy (DR) is a leading cause of blindness in type-2 diabetes mellitus (T2DM) patients.
• Accurate risk prediction models are crucial for timely intervention and management of DR.
• Large-scale data analysis offers potential for identifying novel DR risk factors.

Purpose of the Study:

• To investigate diabetic retinopathy (DR) risk factors using machine learning.
• To develop and evaluate predictive models for DR in T2DM patients.
• To identify key predictors and their clinical significance for DR risk.

Main Methods:

• Retrospective study of 32,452 T2DM inpatients.
• Utilized 60 variables, selecting 17 optimal features via recursive feature elimination.
• Developed prediction models using XGBoost, logistic regression, random forest, and support vector machine, validated with SHAP for interpretability.

Main Results:

• DR prevalence was 6.28% (2038 patients).
• The XGBoost model achieved the highest AUC (0.90).
• Significant risk factors identified: HbA1c >8%, nephropathy, serum creatinine >100 µmol/L, insulin treatment, and diabetic lower extremity arterial disease. Age >65 was protective.

Conclusions:

• The XGBoost model demonstrates high reliability for assessing DR risk indicators.
• SHAP analysis provides clinical interpretability, identifying critical risk factors and their thresholds.
• This approach facilitates early detection and personalized management strategies for DR.