Diabetic nephropathy nomogram construction based on optical coherence tomography angiography variables

To develop a prediction model and corresponding nomogram for diabetic nephropathy (DN) using optical coherence tomography angiography (OCTA) variables.

Patients with type 2 diabetes mellitus (T2DM) were retrospectively enrolled during diabetic retinopathy screening and randomly assigned to training and validation sets in a 7:3 ratio. Predictive OCTA variables were selected using the least absolute shrinkage and selection operator (LASSO) method and used to establish the model. Model performance was evaluated using receiver operating characteristic (ROC) curves, calibration curves, and decision curve analysis. A nomogram was then constructed based on the final model.

A total of 324 eyes were included in the training set and 140 in the validation set. Deep capillary plexus (DCP) parafoveal density, foveal capillary density in the 300 µm-wide area surrounding the foveal avascular zone (FD-300), age, sex, and axial length were incorporated into the model. In the training set, the model achieved a C-index of 0.728 with single sampling and 0.747 with repeated sampling. In the validation set, the C-index was 0.678 with single sampling and 0.681 with repeated sampling. Calibration curves demonstrated good agreement between predicted and observed outcomes in both sets. Decision curve analysis supported the clinical utility and applicability of the nomogram.

We developed a prediction model for DN with relatively good performance using OCTA-derived variables. DCP density and the FD-300 area were identified as key predictors. The resulting nomogram may serve as a useful diagnostic tool for DN and support future advances in OCTA-based artificial intelligence diagnostic systems. However, as external validation datasets are still missing, the results of this study should still be considered somewhat preliminary.