1. Introduction: The Clinical Gap in Intermediate AMD Management
Age-related macular degeneration (AMD) remains the primary catalyst for irreversible vision loss in the population over age 50, presenting a persistent challenge to global ophthalmic health. Historically, the clinical paradigm has bifurcated around late-stage manifestations: exudative macular neovascularization (MNV) and geographic atrophy (GA). While the advent of anti-VEGF therapy transformed the prognosis for “wet” AMD, the atrophic spectrum long remained a therapeutic desert.
The recent FDA approvals of complement inhibitors, such as pegcetacoplan (Syfovre) and avacincaptad pegol (Izervay), represent a watershed moment in retina care. However, these therapies are indicated strictly for slowing the enlargement of established geographic atrophy. This creates a glaring clinical gap: our management remains reactive rather than proactive. Currently, there are no approved interventions to prevent the transition from intermediate AMD (iAMD) to structural atrophy. The ability to identify which iAMD eyes are at the highest risk for conversion within a meaningful clinical window is the “holy grail” of modern risk stratification.
Predicting this progression is technically demanding. Progression from iAMD to atrophy occurs slowly, with fewer than 1 in 21 patients converting to GA annually in natural history studies. To address this, a landmark retrospective study published in Ophthalmology evaluated the predictive utility of quantitative optical coherence tomography (OCT) biomarkers. By shifting the focus from the retinal pigment epithelium (RPE) to the health of the photoreceptors—specifically through the lens of ellipsoid zone (EZ) integrity—this research leverages machine learning-enhanced segmentation and deep-learning models to provide a reproducible, data-driven framework for predicting 2-year progression.
2. Methodology: A High-Resolution, Multi-Center Approach
This study utilized a robust, multi-center cohort of 502 eyes with iAMD, drawn from two premier tertiary sites: the Cleveland Clinic Cole Eye Institute and Retina Consultants of Texas. The methodology reflects a “real-world” high-resolution imaging environment, utilizing both Cirrus (41%) and Spectralis (59%) SD-OCT platforms.
Inclusion and Exclusion Criteria:
- Clinical Verification: All eyes required a verified diagnosis of iAMD based on the presence of large drusen (≥125 µm) on baseline SD-OCT.
- Baseline Status: Eyes with any evidence of total RPE loss (atrophy) or exudation (MNV) at baseline were excluded.
- Confounding Factors: Patients with a history of anti-VEGF injections, prior vitreoretinal surgery, or concurrent retinal disease were excluded to ensure the purity of the atrophic progression signal.
- Image Quality: Scans with inadequate signal-to-noise ratios or poor delineation of retinal layers were removed to maintain the integrity of the quantitative analysis.
- Follow-up: All participants required a 24-month (±1 month) follow-up OCT scan.
The “Converter” Outcome Measure A “Converter” was defined by the development of OCT-based total RPE loss with associated outer retinal atrophy. The study utilized a quantitative, area-based surrogate for complete retinal pigment epithelium and outer retinal atrophy (cRORA), requiring a minimum lesion area of ≥0.05 mm². This corresponds to a minimum diameter of approximately 250 microns, providing a reproducible threshold for early atrophic change.
Technical Nuance in Segmentation Data extraction was performed using a validated, machine learning-enhanced multilayer segmentation platform. This system delineated five critical interfaces: the internal limiting membrane (ILM), ellipsoid zone (EZ), retinal pigment epithelium (RPE), outer nuclear layer (ONL), and Bruch’s membrane (BM). To ensure the highest fidelity for imaging specialists, it is important to note that subretinal drusenoid deposits (SDDs) were accounted for by measuring the RPE as elevated in those areas, accurately reflecting the structural impact on the overlying photoreceptor outer segments.
3. The Biomarker Deep-Dive: What Drives Progression?
The core of the analysis involved a granular comparison of baseline structural metrics between “Converters” (those progressing to atrophy by Year 2) and “Non-Converters.” The data reveals that converters possess a distinct structural signature long before overt RPE failure occurs.
| Biomarker Metric | Non-Converters (n=421) | Converters (n=81) | P-Value |
| Panmacular EZ-RPE Volume (mm³) | 1.255 ± 0.384 | 1.128 ± 0.131 | <0.001 |
| Panmacular Partial EZ Attenuation (≤20 µm) (%) | 1.71% ± 5.44% | 7.34% ± 9.27% | <0.001 |
| Panmacular Ten-Micron EZ Map Coverage (%) | 0.62% ± 3.02% | 3.43% ± 5.59% | <0.001 |
| Panmacular Total EZ Attenuation (0 µm) (%) | 0.32% ± 1.29% | 1.75% ± 3.34% | <0.001 |
| Hyperreflective Foci (HRF) Count | 2.15 ± 6.13 | 9.00 ± 17.70 | 0.001 |
| Central Subfield ONL-RPE Thickness (µm) | 153.4 ± 16.7 | 143.2 ± 22.1 | <0.001 |
| Panmacular Drusen Volume (mm³) | 0.121 ± 0.097 | 0.187 ± 0.166 | 0.001 |
Ellipsoid Zone (EZ) Integrity The status of the EZ, a proxy for photoreceptor health and mitochondrial activity, emerged as the most critical structural biomarker. The research identified a spectrum of degradation: partial EZ attenuation (≤20 µm), a “Ten-Micron” midpoint of significant disruption, and total EZ attenuation (0 µm).
Converters showed more than a four-fold increase in panmacular partial EZ attenuation compared to non-converters (7.34% vs. 1.71%). This suggests that pathological thinning of the outer segments is a massive precursor to eventual RPE death.
Hyperreflective Foci (HRF) HRF count was a standout clinical predictor. Converters averaged 9.00 foci at baseline, whereas non-converters averaged only 2.15. For the MSL communicating risk, the most powerful statistic here is the Odds Ratio of 9.18 for the presence of HRF in the converter group. These lesions, likely representing migrating RPE cells, serve as an urgent signal of impending structural collapse.
Outer Retinal Thinning and Drusen Volume The quantitative reduction in ONL-RPE volume serves as a structural manifestation of the progressive neurodegenerative component of iAMD, preceding overt RPE failure. While drusen volume was significantly higher in converters (0.187 mm³ vs 0.121 mm³), multivariate modeling demonstrated that while drusen burden is a contributory factor, it is a less dominant predictor of conversion than active EZ integrity loss.
4. Advanced Analytics: Deep Learning and “EZ At-Risk”
Beyond standard segmentation, two specific deep-learning (DL) models were employed to identify subtle, “invisible” decay signals.
Hypertransmission (HT)
The DL en face HT model identifies areas where RPE thinning allows increased OCT signal into the choroid, even in the absence of a defined atrophic “hole.” At baseline, converters exhibited a hypertransmission defect area of 7.48%, nearly double that of non-converters (3.73%). This underscores that RPE irregularity and “pre-atrophic” hypertransmission are vital indicators of imminent risk.
EZ At-Risk Model
This specialized model identifies focal EZ-RPE thinning (photoreceptor disruption) specifically in regions where the RPE is still intact. By excluding areas of existing RPE atrophy, the model focuses on identifying the earliest phase of structural failure. Converters had baseline EZ at-risk levels of 6.74%, compared to 2.55% in non-converters. This confirms that photoreceptor degradation is a primary event that often occurs independently of, and prior to, overt RPE loss.
5. Performance Analysis: The Machine Learning Random Forest Model
To synthesize these biomarkers into a predictive clinical tool, a random forest classifier was trained and validated. The study compared four iterations to determine the incremental value of advanced biomarkers.
- Model A (Benchmark): Utilized age, drusen volume, and HRF count. It achieved an AUC of 0.741.
- Model B (+HT): Adding deep-learning hypertransmission data improved the AUC to 0.783.
- Model C (+EZ): Incorporating quantitative EZ biomarkers resulted in a substantial jump in performance, reaching an AUC of 0.840.
- Model D (Full Model): The comprehensive model, including Age, HRF, Drusen, HT, and all EZ metrics, reached the highest performance with an AUC of 0.853.
Feature Importance Ranking In the “Feature Importance” analysis, EZ integrity metrics and HRF count consistently ranked as the strongest predictors, significantly outperforming traditional clinical measures like age. The data validates that EZ loss is an independent and critical predictor of progression, providing the superior predictive value required for modern risk stratification.
6. Discussion: The Evolving Role of Photoreceptor Health
For the Senior MSL and the practicing Retina Specialist, these findings necessitate a shift in how we conceptualize iAMD. We are moving from an “RPE-centric” view of the disease toward one that prioritizes the health of the photoreceptor. Quantitative EZ integrity metrics offer an objective, reproducible measurement that can be integrated into routine monitoring.
Clinical Trial Connectivity and Strategic Impact These findings are particularly relevant when viewed alongside recent clinical trials. The ReCLAIM-2 Phase 2 trial (elamipretide) specifically utilized EZ total attenuation as a secondary endpoint, demonstrating that targeting mitochondrial health could slow the rate of EZ loss. Similarly, the GATHER 1/2 (avacincaptad pegol) and ARCHER (ANX007) trials have shown that preserving the EZ is a measurable and achievable therapeutic goal.
Importantly, EZ loss has been identified as an “approvable endpoint” by the FDA in dry AMD. This study’s 2-year prediction window is highly relevant because most clinical trials for GA (such as GATHER or ARCHER) operate on a similar 12-to-24-month timeframe. By achieving an AUC of 0.853, this model provides a framework for “enriching” future clinical study populations—selecting eyes that are most likely to progress—thereby increasing the statistical power and efficiency of prevention-prevention trials.
Limitations As with any retrospective analysis, limitations exist. The study relied solely on SD-OCT, and while this is the central modality for clinical practice and CAM-based definitions, multi-modal imaging could provide additional context. Furthermore, the “cRORA surrogate” (0.05 mm² area) is a conservative threshold that may exclude the very smallest, non-circular lesions defined by CAM.
7. Conclusion: Strategic Implications for Patient Care
The ability to identify high-risk “converters” using quantitative OCT biomarkers is a prerequisite for the next era of AMD therapy: intervention before the onset of irreversible atrophy.
Key Clinical Takeaways:
- EZ Integrity is the Primary Signal: Partial EZ attenuation (thinning to ≤20 µm) and the ten-micron midpoint are potent heralds of imminent atrophy.
- HRF are High-Priority Markers: A baseline HRF presence carries an Odds Ratio of 9.18 for 2-year conversion; these are “red flags” for the clinician.
- Automation Drives Accuracy: Deep-learning models for “EZ At-Risk” and Hypertransmission enhance predictive AUC from a standard 0.741 to a sophisticated 0.853.
- Photoreceptor-First Paradigm: Structural failure in the EZ and ONL often precedes the total loss of the RPE band, making them the most sensitive indicators for early intervention.
In conclusion, quantitative OCT biomarkers, led by high-resolution EZ integrity measures, provide the necessary framework for identifying high-risk iAMD eyes. This data allows us to move beyond reactive management toward a strategy of proactive risk stratification and targeted clinical trial design.
8. References (Abridged for Weblog)
- Matar K, Ehlers JP, et al. Assessment of Ellipsoid Zone Integrity and Other Quantitative OCT Biomarkers for Intermediate AMD Progression to Atrophy. Ophthalmology. 2026. doi:10.1016/j.ophtha.2026.08.005.
- Sadda SR, et al. Consensus Definition for Atrophy Associated with Age-Related Macular Degeneration on OCT: Classification of Atrophy Report 3. Ophthalmology. 2018;125(4):537-548.
- Ehlers JP, et al. ReCLAIM-2: A Randomized Phase II Clinical Trial Evaluating Elamipretide in Age-related Macular Degeneration. Ophthalmology Science. 2025;5(1).
- Jaffe GJ, et al. Imaging Features Associated with Progression to Geographic Atrophy in AMD: Classification of Atrophy Meeting Report 5. Ophthalmol Retina. 2021;5(9):855-867.
- Tao LW, Wu Z, et al. Ellipsoid zone on optical coherence tomography: a review. Clin Exp Ophthalmol. 2016;44(5):422-430.
