Elevated Retinal Neovascularization on Widefield Optical Coherence Tomography Angiography Predicts Complications in High-Risk Proliferative Diabetic Retinopathy.

PURPOSE: To determine whether retinal neovascularization (RNV) metrics derived from single-shot widefield swept-source OCT angiography (SS-OCTA) predict subsequent vision-threatening complications in eyes with high-risk proliferative diabetic retinopathy (PDR). DESIGN: Prospective case series. PARTICIPANTS: Eyes clinically graded as high-risk PDR at a tertiary care center, followed up for at least 6 months. METHODS: Eligible eyes underwent single-shot 26 × 21-mm SS-OCTA imaging (DREAM OCT, Intalight Inc.). A validated deep learning-based algorithm segmented the vitreous cavity slab to generate en face OCTA images for automated detection and quantification of RNV membrane and the vascular areas. Lesions were classified as elevated when they were separated from the internal limiting membrane (ILM) and attached when there was no space between the lesion and the ILM. We analyzed baseline OCTA-derived metrics for their predicting eyes that developed new or recurrent vitreous hemorrhage (VH) or tractional retinal detachment (TRD) during follow-up. MAIN OUTCOME MEASURES: Incidence of new or recurrent vitreous hemorrhage and traction retinal detachment. RESULTS: Over a median follow-up period of 291 days (range, 180-466), 8 of 18 eyes (44.4%) developed complications, with 7 (38.9%) developing VH and 1 (5.6%) developing TRD. Among the 115 identified RNV lesions, 87 (75.7%) were located outside the arcades. Compared to eyes without complications, eyes with complications had a larger median total RNV membrane area (25.72 mm² vs 1.33 mm²; P = .006) and a larger median total RNV vascular area (9.72 mm² vs 0.76 mm²; P = .010). Eyes with complications had a larger elevated RNV membrane area (5.13 mm² vs 0.10 mm²; P = .009) and vascular area (2.69 mm² vs 0.05 mm²; P = .007), whereas attached RNV metrics were not significantly different between groups. Total RNV membrane area demonstrated the highest predictive performance for identifying eyes at risk of complications (AUC = 0.888), with a sensitivity of 87.5% and a specificity of 80.0% at a cutoff value of 3.40 mm². CONCLUSIONS: Widefield SS-OCTA is useful for evaluating RNV burden and its axial relationship to the ILM in high-risk PDR. Elevated baseline RNV, incorporating spatial and anatomic features, predicts subsequent tractional complications such as VH and TRD. These imaging biomarkers may complement current clinical staging of PDR.