Prognostic value of OCT biomarkers and systemic factors for early and long term anatomical treatment response in diabetic macular edema

Abstract Background To evaluate the prognostic value of baseline optical coherence tomography (OCT) biomarkers and systemic factors for early and long-term anatomical response to anti-VEGF therapy in diabetic macular edema (DME). Methods This retrospective study included 164 eyes from 105 patients with type 2 diabetes and DME who received three monthly loading doses of intravitreal bevacizumab (January 2013 to April 2024) and were subsequently managed with a pro re nata (PRN) regimen using standard retreatment criteria. Completion of the loading phase (minimum follow-up, 3 months) was required for inclusion. Anatomical response was assessed after the loading phase (early response) and at final follow-up (long-term response); eyes were classified as full, partial, or non-responders based on CMT and OCT morphology. Baseline demographic, systemic, and OCT characteristics were analyzed using generalized estimating equation (GEE) logistic regression to account for inter-eye correlation, followed by one-eye-per-patient sensitivity analyses. Results After loading, 10 eyes (6.1%) were full responders, 72 (43.9%) partial responders, and 82 (50.0%) non-responders; 90% of early full responders maintained full response at final follow-up. In multivariable GEE analyses, higher baseline CMT, higher hyperreflective foci (HRF) count, and hyperreflective intraretinal material were independently associated with early anatomical response under the original percentage-reduction outcome definition; for long-term response, baseline CMT was only nominally associated and did not survive correction for multiple testing, and DRIL, EZ/ELM defect, and intraretinal cyst size were not significant. Both models showed moderate discrimination after bootstrap optimism correction (AUC 0.751 and 0.757, respectively). Sensitivity analyses across eye-selection strategies were consistent for baseline CMT, DRIL, EZ/ELM defect, and hyperreflective material, but intraretinal cyst size was unstable; all associations were sensitive to how anatomical response was defined. Conclusions Baseline OCT biomarkers showed candidate associations with anatomical response, but several, including the hyperreflective-material association, were not reproduced when the outcome was redefined using continuous final CMT or an absolute CMT threshold, despite robustness within the primary model. Discrimination was moderate (optimism-corrected AUC 0.75–0.76). These findings identify candidate prognostic OCT biomarkers requiring external validation before clinical application.