AI-based qOCT analysis of phase 3 trials OAKS & DERBY data on the effect of pegcetacoplan on geographic atrophy

Abstract

Purpose : To quantify the effect of pegcetacoplan on geographic atrophy (GA) at the spectral domain optical coherence tomography (SD-OCT) level by querying the change in area and topographical location of each of the Classification of Atrophy Meeting (CAM)-defined endpoints (photoreceptor degeneration [PRD], retinal pigment epithelium [RPE] loss, hypertransmission (HTR), and RPE and Outer Retinal Atrophy [RORA]).

Methods : The SD-OCT data was collected for OAKS and DERBY phase 3 randomized controlled trials that assigned participants to receive intravitreal pegcetacoplan monthly (PM), every other month (PEOM), or corresponding sham treatment. Our deep-learning approach allowed automatic segmentation of all CAM-defined GA biomarkers. The primary endpoint was the least square (LS) mean change from baseline in area (mm2) of RORA in each treatment arm (PM, PEOM and sham-pooled) at 24 months. Secondary outcomes included temporally and spatially-resolved change in RORA and its features within the Early Treatment Diabetic Retinopathy Study [ETDRS] grid.

Results : Pegcetacoplan reduced the growth rate of SD-OCT GA biomarkers for the duration of the follow up. In comparison to the sham-pooled group, a statistically significant reduction in the LS mean change of RORA area was detectable from baseline to month 3 in study eyes receiving PEOM (0.27 SE 0.03 mm2 vs. 0.44 SE 0.03 in sham-pooled; P < 0.0001), as well as, at every time point up to and including month 24 (PEOM: 2.79 mm2 SE 0.09; sham-pooled: 3.48 mm2 SE 0.412; P < 0.0001). The effect was greater with increased dosing frequency from PEOM to PM (-0.17 mm2 [95% CI -0.43 to 0.08; P = 0.17]). The perifoveal and parafoveal areas, but not the foveal region, demonstrated significant differences from sham in LS mean area growth vs baseline of RORA for both the PM and the PEOM groups at 24 months.

Conclusions : Insights at the SD-OCT level reaffirmed the role of pegcetacoplan in slowing GA growth. Our deep-learning approach bypassed the rate-limiting step of imaging data analysis in assessing therapeutic efficacy in GA, made it feasible to instantly classify each case along the spectrum of GA subtypes according to the most widely established international consensus definitions, and allowed us to incorporate topographical information to yield greater insight into this spatially heterogeneous disease.

This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.