Macular Deep Capillary Plexus Ischemia as a Biomarker for Identifying Referable Diabetic Retinopathy

Purpose: To evaluate deep capillary plexus (DCP) ischemia measured on 3 × 3-mm macular optical coherence tomography angiography (OCTA) as a biomarker for referable diabetic retinopathy (refDR). Methods: In this cross-sectional study, we analyzed 290 eyes from 193 patients with diabetes and varying severity but without macular edema. Averaged macular OCTA images were processed to calculate a range of OCTA metrics, including geometric perfusion deficits (GPDs). Each OCTA metric was evaluated for its ability to discriminate refDR using (1) an OCTA-only generalized estimating equation logistic model and (2) a model combining the metric with clinical covariates (age, sex, diabetes duration, hemoglobin A1c, and hypertension). Accuracy was quantified by the area under the curve (AUC), and model-to-model differences (ΔAUC) were evaluated using 1000-iteration cluster-bootstrap resampling. Internal validity and calibration of the best-performing OCTA-only model were assessed using subject-wise fivefold cross-validation and the Brier score. Results: Among 10 OCTA metrics, GPDs in the DCP (GPD-DCP) achieved the highest standalone discrimination (AUC = 0.913; 95% confidence interval, 0.879-0.947) and significantly improved the model with clinical covariates (AUC increase from 0.865 to 0.934; ΔAUC = +0.070; P = 0.002). At clinically relevant thresholds, GPD-DCP achieved 95% sensitivity with 63.5% specificity, as well as 95% specificity with 63.2% sensitivity. Subject-wise cross-validation confirmed robust performance (mean [SD] AUC = 0.919 [0.034]), with excellent calibration (Brier score = 0.115). Conclusions: GPD-DCP offers excellent, well-calibrated discrimination of refDR and significantly improves risk stratification when combined with standard clinical factors, supporting its use for screening and risk assessment.