AI-Based OCT Assessment of Anatomic Response During the Loading Phase of Anti-VEGF Therapy in Neovascular Age-Related Macular Degeneration

Purpose: To compare functional and anatomic outcomes after the loading phase of intravitreal bevacizumab, aflibercept 2 mg, or faricimab in treatment-naive neovascular age-related macular degeneration (nAMD), using artificial intelligence (AI)-based optical coherence tomography (OCT) analysis. Methods: This multicenter retrospective cohort study included 224 eyes. Patients received three monthly injections of bevacizumab (n = 80), aflibercept 2 mg (n = 78), or faricimab (n = 66). Best-corrected visual acuity (BCVA) and spectral-domain OCT were assessed at baseline and week 12. AI-based segmentation quantified volumetric measurements of intraretinal fluid (IRFv), subretinal fluid (SRFv), and pigment epithelial detachment. The primary outcome was qualitative dry macula (absence of intraretinal fluid and subretinal fluid). An exploratory composite OCT biomarker (ΔCOMBO), defined as the lesser percentage reduction between IRFv and SRFv, was evaluated as a hypothesis-generating metric using receiver operating characteristic analysis. Baseline-adjusted analyses and multivariable logistic regression were performed. Results: Descriptive within-group analyses showed improvement in BCVA and reduction in OCT parameters at week 12 in treatment groups. BCVA improvement was statistically significant in the bevacizumab and faricimab groups. In baseline-adjusted analyses, between-group differences emerged for OCT outcomes, whereas no significant between-group differences were observed for BCVA. Higher rates of dry macula were observed with faricimab (80.3%) and aflibercept 2 mg (67.9%) compared with bevacizumab (41.3%) (P < 0.001). The exploratory composite metric ΔCOMBO showed high discriminative performance (area under the curve = 0.959; 95% confidence interval, 0.907-0.999). Conclusions: All agents improved functional and anatomic outcomes after the loading phase in treatment-naive nAMD. Baseline-adjusted analyses showed differences in anatomic response patterns across treatments. Exploratory AI-based OCT biomarkers may help characterize short-term anatomic response; however, these findings require external validation.