A clinically interpretable machine learning model for early detection of diabetic retinopathy in multiple community health centers
By
Juncheng Tong
Aifa Tang
Lifang Liu
Luyuan Zhang
Hainan Wang
Mengyuan Qu
Bing Liu
May 4, 2026
Clinical Scorecard: A machine learning approach for the early identification of diabetic retinopathy in various community health settings
At a Glance
Category Detail
Condition Diabetic Retinopathy (DR) Key Mechanisms Machine learning models utilizing health data to predict DR risk. Target Population Diabetic individuals aged ≥18 years from community health facilities. Care Setting Primary care settings in low- and middle-income countries.
Key Highlights
Prevalence of DR was found to be 13.5% in the study cohort. GLMNET model achieved an AUROC of 0.770 and AUPRC of 0.452. Urine glucose identified as the most significant predictor of DR. Decision curve analysis indicated net benefit for threshold probabilities of 10% to 40%. Study emphasizes the need for external validation before clinical application. Guideline-Based Recommendations
Diagnosis
Utilize routine fundus examinations for DR identification. Employ machine learning models for risk stratification in primary care. Management
Incorporate clinical variables such as diabetes type, duration, and HbA1c levels. Monitoring & Follow-up
Regularly assess patients for DR risk using predictive models. Risks
Low compliance with ophthalmology referrals in community settings. Patient & Prescribing Data
Diabetic individuals from community health facilities in China.
Focus on early identification and risk stratification to prevent visual impairment.
Clinical Best Practices
Integrate machine learning tools into routine clinical workflows. Ensure model interpretability to enhance provider confidence in decision-making. Conduct prospective studies for validation of predictive models. References
