A recent study in Diabetes Research and Clinical Practice determined the association of machine learning (ML) with diabetic retinopathy (DR) and dementia. The study aimed to develop an ML model to predict DR and to determine its significance as a biomarker for dementia.

A total of 27,929 patients diagnosed with type 2 diabetes mellitus (T2DM) without dementia were included in the study. Using three ML algorithms, a prediction model was developed. Across eight institutions, the best-performing model was validated. To assess dementia incidence, patients were followed for a period of 3 years. Using Kaplan-Meier and Cox regression, the dementia risk between ML-predicted DR and non-DR groups was compared.

The best performance was shown by XGBoost (AUROC: 0.746), with external validation AUROCs ranging from 0.555 to 0.620. Increased all-cause dementia risk (HR: 1.32), vascular dementia (HR: 1.38), and Alzheimer's disease (HR: 1.30) were associated with predicted DR.