Highlights
- In silico power analyses demonstrate that incorporating a 9-month run-in phase can reduce the required sample size for a 90% powered RCT from 156 to just 26 participants.
- Single-arm trial designs, which compare treatment phases against a patient’s own run-in natural history, offer the highest statistical efficiency, potentially requiring as few as 14 participants.
- Mesopic microperimetry (MMP) significantly outperforms visual acuity as a functional outcome measure, requiring far smaller cohorts to detect significant disease progression.
- Emerging AI-driven longitudinal modeling and glial-cell-based biomarkers provide new avenues for precision patient stratification and growth rate prediction.
Background
Geographic atrophy (GA), the advanced non-neovascular stage of age-related macular degeneration (AMD), is a leading cause of irreversible blindness worldwide, affecting over 5 million people. Characterized by the progressive loss of retinal pigment epithelium (RPE), photoreceptors, and choriocapillaris, GA presents a formidable challenge for drug development. Historically, interventional retinal trials have faced high attrition rates; an analysis of trials conducted between 2015 and 2025 revealed that while Phase 2 studies often fail during interim analyses, Phase 3 trials frequently collapse due to an inability to meet primary efficacy endpoints despite massive geographic dispersion and high assessment burdens (PMID: 42082070).
One of the primary hurdles in GA research is the slow and variable nature of lesion enlargement. Traditional randomized controlled trials (RCTs) require large sample sizes and long follow-up periods to differentiate treatment effects from natural history. However, GA enlargement rates within a specific eye tend to be remarkably stable over time. Exploiting this intra-eye stability through innovative trial architectures—such as run-in phases and single-arm designs—represents a paradigm shift in clinical trial methodology.
Key Content
Methodological Advances: Run-In Phases and Single-Arm Designs
A pivotal 2026 diagnostic/prognostic study by Hou et al. utilized parameters from the GA Minocycline Trial to evaluate trial efficiency (PMID: 42133353). The researchers compared standard 1:1 RCTs with designs incorporating a run-in phase (to collect natural history data before randomization) and single-arm trials (where the run-in phase serves as the control).
The findings were transformative: for a 2-year trial with 90% power (assuming a 25% treatment effect), a standard RCT required 156 participants. Incorporating a 3-month run-in reduced this to 82; a 6-month run-in reduced it to 40; and a 9-month run-in slashed the requirement to 26 participants. Even more efficient was the single-arm design; a 2-year single-arm trial with a 9-month run-in required only 14 participants to achieve the same statistical power. These results suggest that using a patient as their own control leverages the inherent stability of GA enlargement, dramatically lowering the barriers to clinical research (PMID: 42133353).
Refining Functional and Anatomic Endpoints
Efficiency is not only a matter of trial structure but also of endpoint selection. While visual acuity (VA) remains a regulatory gold standard, it is often insensitive to early GA progression. Research into mesopic microperimetry (MMP) has shown that mean sensitivity (MS) thresholds provide a much more robust measure of retinal function (PMID: 40977307). Statistical modeling indicates that using MS as an endpoint requires significantly fewer patients (n=51 at 12 months) compared to VA (n=203 at 12 months) to show significant changes.
Furthermore, anatomic biomarkers are being refined. Perilesional fundus autofluorescence (FAF) patterns are not static; transitions between patterns like “diffuse nontrickling” and others are associated with changes in growth rates (PMID: 41439216). Recognizing these transitions can improve the accuracy of growth rate predictions and patient stratification.
AI and Precision Stratification
The integration of Artificial Intelligence (AI) is addressing the challenge of the “low-data regime” in GA trials. The SWAU-Net (Sliding-Window Attention U-Net) architecture has demonstrated the ability to integrate Transformer-based temporal modeling with U-Net spatial precision to predict GA growth trajectories accurately (PMID: 41752939).
Moreover, a shift toward a Müller glial cell-centered framework suggests that the integrity of the external limiting membrane (ELM) could serve as a critical stratification variable. Pre-ELM descent stages may be more suitable for neuroprotective interventions, while post-ELM stages might require regenerative or stabilization strategies (PMID: 41985834). Failure to account for these biological stages in past trials may have led to the dilution of treatment effects.
Expert Commentary
The shift toward small-cohort, high-efficiency trials is essential given the high cost and logistical complexity of global Phase 3 studies. However, experts caution that while single-arm designs are statistically powerful, they lack the rigorous control for systemic confounders and the “placebo effect” inherent in double-masked RCTs. The stability of GA enlargement is a strong biological rationale for using run-in data, but it assumes that the rate of change is linear and that no external factors (such as the sudden onset of neovascularization) intervene.
Additionally, the prevalence of nonexudative macular neovascularization (neMNV) must be considered. Although a “double-layer sign” (DLS) on OCT is often used as a surrogate for neMNV, recent multicenter studies show that only 40% of eyes with DLS actually harbor neMNV on OCT angiography (PMID: 41954906). This underscores the need for high-fidelity multimodal imaging to ensure that trial cohorts are truly homogeneous.
Conclusion
Improving efficiency in GA clinical trials is no longer just a statistical goal but a clinical necessity. By adopting run-in phases and exploring single-arm designs, researchers can exploit the intra-eye stability of GA enlargement to reduce sample sizes by over 80%. When coupled with sensitive functional measures like microperimetry and AI-driven predictive modeling, these methodological advances offer a pathway toward faster, more cost-effective, and ultimately more successful drug development for this devastating disease. Future research should focus on validating these in silico findings in real-world prospective trials and harmonizing regulatory requirements with these emerging high-efficiency designs.
References
- Hou J, von der Emde L, Mukherjee S, et al. Improving Efficiency in Geographic Atrophy Clinical Trials Using Run-In Phases or Single-Arm Designs. JAMA Ophthalmol. 2026;e261380. PMID: 42133353.
- Assi L, et al. Unsuccessful Clinical Trials in Retina: Lessons Learned. Am J Ophthalmol. 2026;S0002-9394(26)00229-1. PMID: 42082070.
- Acta Ophthalmol. Power and clinical utility of mesopic microperimetry analysis strategies in age-related macular degeneration. 2026;104(3):e292-e299. PMID: 40977307.
- Prog Retin Eye Res. Are Müller glial cells gatekeepers of neuroprotection and regeneration in age-related macular degeneration? 2026;112:101471. PMID: 41985834.
- Ophthalmol Sci. Perilesional Fundus Autofluorescence Patterns Are Not Static: Longitudinal Transitions in Geographic Atrophy and Association with Disease Progression. 2025;6(2):100995. PMID: 41439216.
