OCT Photoreceptor Biomarkers in AMD Need a Common Language
Advanced age-related macular degeneration has long been defined by the damage it leaves behind, but the deeper clinical problem is more granular: photoreceptor integrity is the final shared pathway to irreversible vision loss, and current imaging conventions still do not describe that pathway in a uniform way. A scoping review and nested meta-analysis in Frontiers in Medicine examined how optical coherence tomography biomarkers of the ellipsoid zone, outer nuclear layer, and external limiting membrane are defined, segmented, validated, and linked to visual function in AMD.
The review is clinically important because OCT has moved beyond static anatomic description. It is now being used to quantify tissue preservation, disease progression, and potential treatment response. Without consistent definitions of what constitutes photoreceptor loss, however, two studies can measure seemingly similar endpoints while actually capturing different biological structures.
Where the Variability Begins
MEDLINE, Embase, and Scopus were searched from January 2015 through December 2025 using PRISMA-ScR methods. Ninety-four studies met inclusion criteria, covering spectral-domain and swept-source OCT across a range of AMD phenotypes. The breadth of that literature is reassuring, but it also exposes a persistent problem: investigators have not used a single shared framework for boundary definitions, analytic domains, or spatial sampling.
Boundary selection alone created major inconsistency. Some studies defined the ellipsoid zone relative to the retinal pigment epithelium, while others used the Bruch’s membrane reference. Measurement strategies also diverged, with studies reporting thickness, area, volume, or reflectivity rather than a common endpoint format. Spatial analysis varied just as widely, ranging from the central 1 mm to ETDRS subfields and lesion-centered masks.
A biomarker cannot become a trial endpoint if every group measures a different version of the same lesion.
Reliability Was Strong When the Rules Were Clear
When segmentation boundaries and analytic conventions were explicitly stated and consistently applied, reliability was high. Ellipsoid zone and outer nuclear layer metrics reached ICC > 0.90, a level that supports serious consideration for longitudinal research and interreader consistency. Automated and deep-learning approaches performed comparably to expert manual grading, suggesting that computational methods can support scale without sacrificing technical validity when the input definitions are disciplined.
This matters because reproducibility is the threshold issue for biomarker adoption. A metric that cannot be repeated with confidence across readers, platforms, and sites will stall before it reaches regulatory or trial use. The review suggests that the imaging technology itself is not the limiting factor; the limiting factor is the lack of standardization around what exactly is being measured.
Structure and Function Tracked in the Expected Direction
The relationship between OCT photoreceptor loss and functional decline was clinically coherent. Structure-function analyses showed moderate-to-strong correlations between photoreceptor damage and microperimetry sensitivity, with r values ranging from 0.50 to 0.80. Best-corrected visual acuity also tracked with photoreceptor integrity, with correlations between 0.40 and 0.70, especially in advanced disease stages.
Those associations are biologically plausible. As the ellipsoid zone disappears and the outer nuclear layer thins, the retina loses the cellular architecture required for phototransduction and signal transmission. Visual acuity may remain deceptively preserved early in disease, but once photoreceptor loss becomes widespread or topographically central, functional collapse becomes more measurable. Microperimetry appears particularly well aligned with these structural changes because it samples localized retinal sensitivity rather than relying only on global acuity.
Geographic Atrophy Detection Was Consistent, But Early Disease Was Not
Agreement between OCT and fundus autofluorescence for geographic atrophy detection was high, with κ > 0.80. In advanced atrophic disease, the anatomy is easier to classify because the transition from preserved to lost tissue is more obvious across modalities. Early atrophic features were harder to define consistently, and that ambiguity showed up in the reporting.
Phenotypic stratification and topographic reporting were inconsistently documented throughout the literature. Some studies described lesion context in enough detail to support meaningful comparison, while others did not specify the exact retinal zone, stage, or analytic mask used. Without those details, cross-study synthesis becomes fragile, and apparent differences in effect size may simply reflect different conventions rather than different biology.
CLEAR Sets the Agenda for Endpoint Harmonization
The authors frame these findings as the conceptual basis for the CLEAR initiative, the Consensus Layer Evaluation for AI Algorithm Reporting. That direction is sensible because biomarker development in AMD now depends on more than image acquisition. It requires minimum reporting standards, harmonized boundary definitions, and validation frameworks that allow a metric to travel across cohorts, platforms, and trial designs.
For prevention and early-intervention trials, the stakes are practical. A biomarker that identifies subtle photoreceptor change before overt atrophy could help enrich cohorts, monitor progression, and measure response before functional vision is irreversibly lost. Yet that promise depends on whether the field can agree on the anatomy beneath the label. Right now, the literature supports reliability and clinical relevance, but not uniformity.
The review leaves one central challenge in plain view: the photoreceptor metrics are ready for broader use, but the reporting system around them is not yet mature enough to make every study directly comparable.