ABSTRACT Background: This study aimed to identify optical coherence tomography (OCT) biomarkers for predicting response to anti-VEGF treatment in diabetic macular edema (DME) Methods: Bilateral DME patients with asymmetric response to a loading dose of anti-VEGF (ranibizumab/aflibercept) treatment were retrospectively studied. The morphologic response criterion was central subfield thickness (CST) ≤300 µm; asymmetric response was defined as ≥10% difference in CST reduction between the eyes. The functional response criterion was an increase in logMAR acuity of ≥3 lines, with an increase below this threshold in the fellow eye considered asymmetric response. Relationships between final morphologic and functional responses to anti-VEGF therapy and baseline values of the following OCT-derived biomarkers were evaluated: DME subtype, CST, vitreoretinal interface anomalies, disorganization of the inner retinal layers (DRIL), external limiting membrane (ELM) disruption, ellipsoid zone (EZ) disruption, and subretinal fluid (SRF). Results: After a loading dose of anti-VEGF, 31 eyes that met both morphologic and functional response criteria were classified as responders (RR) and 27 eyes that did not respond morphologically or functionally based on the defined criteria were classified as resistant (RT). Eyes that showed only functional (n = 5) or morphological response (n = 1) were excluded due to their small number. The presence of SRF or simple epiretinal membrane (ERM) was not associated with any difference in treatment responses (p > .05), while tractional ERM, extensive DRIL (≥500 µm), and ELM and EZ disruptions in the fovea-centered 1000-µm zone were important OCT biomarkers in predicting resistance (p < .001). A multilayer perceptron model ranked predictive power as 100% for ELM disruption, 51.7% for tractional ERM, 25.4% for DRIL, and 24.5% for EZ disruption. Conclusion: Extensive ELM disruption was the strongest OCT biomarker to predict anti-VEGF resistance, followed by tractional ERM. EZ disruption and DRIL had relatively lower predictive value.
Read full abstract