Topic
This systematic review and meta-analysis evaluated whether gene therapy provides safe and clinically meaningful efficacy for patients with neovascular age-related macular degeneration (nAMD). The clinical question addressed outcomes in patients with nAMD receiving gene therapy (primarily adeno-associated virus–based anti-vascular endothelial growth factor (anti-VEGF) constructs) compared with baseline or standard care contexts, focusing on visual acuity, anatomical response, treatment burden, and safety. Anti-VEGF intravitreal injections remain the current standard of care but require frequent administration and long-term adherence.
Clinical Relevance
nAMD is a major cause of irreversible vision loss in older adults and imposes substantial treatment burden due to repeated injections. Gene therapy aims to achieve sustained intraocular therapeutic protein expression after a single or infrequent administration, potentially reducing injection frequency while maintaining disease control. Establishing safety and functional efficacy is critical before translation into routine retinal practice.
Methods
Following Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines, databases (PubMed, Embase, Scopus, Web of Science, Google Scholar, and Cochrane Library) were searched from inception to February 1, 2026. Eligible studies were prospective interventional clinical studies evaluating gene therapy in neovascular age-related macular degeneration, including randomized early-phase trials and open-label dose-escalation cohorts. Primary outcomes included best-corrected visual acuity (BCVA), central subfield thickness (CST), rescue anti-VEGF requirement, mortality, and adverse events (AEs). Risk of bias was assessed using the revised Cochrane Risk of Bias tool version 2 (RoB 2) and Risk Of Bias In Non-randomized Studies of Interventions (ROBINS-I) tools. Multilevel random-effects meta-analyses using restricted maximum likelihood (REML) accounted for clustering of multiarm cohorts.
Results
Eight prospective interventional studies comprising 203 treated participants were included, including randomized early-phase trials and open-label dose-escalation cohorts. The primary multilevel REML analysis showed no significant pooled BCVA improvement (mean difference (MD) 0.54 Early Treatment Diabetic Retinopathy Study (ETDRS) letters; 95% confidence interval (CI), −7.38 to 8.46), despite favorable fixed-effect sensitivity estimates. CST demonstrated significant anatomical reduction (MD 37.13 µm; 95% CI, 26.63–47.62). Approximately 44% of treated eyes required rescue anti-VEGF injections. Safety outcomes showed low-to-moderate cumulative event probabilities, although estimates varied by model, endpoint, and vector platform. Publication bias was detected for BCVA and CST but was minimal for most safety outcomes.
Conclusion
Gene therapy for neovascular age-related macular degeneration demonstrates encouraging anatomical efficacy and manageable short- to midterm safety signals but lacks consistent functional visual improvement. Current evidence supports a treatment-burden-reducing adjunctive role rather than replacement of conventional anti-VEGF therapy. Evidence strength is limited by early-phase designs, small sample sizes, and clinical heterogeneity; ongoing phase 3 trials are required to define long-term efficacy and clinical positioning.
INTRODUCTION
Age-related macular degeneration (AMD) is a leading cause of vision loss in individuals over 60. Approximately 10% of patients develop neovascular age-related macular degeneration (nAMD), which is characterized by macular neovascularization that leaks plasma into the macula. This leakage causes rapid retinal swelling and potentially reversible vision loss. The condition can progress to subretinal fibrosis without treatment, leading to permanent vision loss. With the global population aging, the rising prevalence of AMD presents significant challenges to healthcare systems and socioeconomic structures. Over the past two decades, genomics and molecular biology advancements have significantly enhanced the understanding of nAMD, elucidating its complex genetic and environmental underpinnings.
Genetic variations, particularly in complement factor H and other complement pathway components, highlight the pivotal role of chronic inflammation in nAMD pathogenesis. Furthermore, environmental factors, including smoking, dietary habits, and UV light exposure, have been identified as key contributors to disease progression. These insights have paved the way for targeted therapeutic strategies to modify disease progression rather than merely manage symptoms. The therapeutic landscape for nAMD primarily revolves around anti-VEGF agents, including ranibizumab, aflibercept, and bevacizumab. Administered via intravitreal injections, these agents inhibit abnormal blood vessel growth and reduce macular edema, preserving vision in many patients. However, these treatments are constrained by the need for frequent injections, high costs, and variable patient responses. This highlights the urgent need for innovative approaches to address nAMD more effectively and sustainably.
Gene therapy offers a potentially durable approach to disease management by introducing or modifying genetic material within cells to achieve sustained therapeutic effects. This technology has shown success in treating severe monogenic disorders, such as spinal muscular atrophy, β -thalassemia, primary immunodeficiencies, specific vision impairments, and B-cell cancers resistant to conventional treatments. In recent years, the US Food and Drug Administration has approved several gene therapy products, underscoring the expanding role of this modality in modern medicine. Gene therapy holds considerable promise for nAMD management, with the potential for long-term therapeutic benefits through single or infrequent administration, thereby reducing treatment burden while maintaining sustained therapeutic protein expression. Its advantages include targeting underlying molecular mechanisms, reducing the treatment burden, and delivering durable efficacy.
Recent advances in vector technology, particularly adeno-associated virus (AAV) vectors, have significantly improved the safety and efficiency of gene delivery, making gene therapy a viable option for nAMD. ,, Delivery methods, including subretinal, suprachoroidal, and intravitreal routes, enable precise administration to the retina. Gene therapy for nAMD primarily targets fundamental pathological processes, such as abnormal neovascularization and chronic inflammation. Therapies focusing on VEGF pathways aim to inhibit angiogenesis through sustained protein expression, while others target the complement cascade to modulate excessive inflammatory responses. , The immune-privileged status of the retina and its accessibility via minimally invasive techniques further enhance its suitability for gene therapy. Encouraging results from preclinical models and early-phase clinical trials have laid a robust foundation for larger-scale investigations. ,
Despite its potential, gene therapy for nAMD faces several challenges. The clinical and genetic heterogeneity of nAMD necessitates personalized therapeutic approaches to optimize outcomes. Safety concerns, such as immune reactions and off-target effects, require rigorous attention through advanced vector design and delivery techniques. Long-term studies are essential to evaluate the durability of therapeutic effects and monitor for potential late-onset adverse events (AEs). Furthermore, gene therapy’s high cost and technical complexity pose accessibility challenges, raising ethical and practical concerns regarding equitable implementation.
This systematic review and meta-analysis critically examines current clinical evidence on gene therapy for neovascular age-related macular degeneration, focusing on visual acuity, anatomical response, rescue anti-VEGF requirement, and safety. This review provides a comprehensive overview of advancements in this rapidly evolving field by consolidating findings from recent clinical studies. It also highlights significant challenges and future directions, underscoring the importance of interdisciplinary collaboration to unlock the full potential of gene therapy in transforming the nAMD treatment paradigm.
METHODS
We performed this systematic review and meta-analysis by adhering to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. The protocol was registered with the International Prospective Register of Systematic Reviews (PROSPERO) under the registration ID CRD42024619992. Compliance with the Declaration of Helsinki’s principles reinforced this review’s ethical foundation. This review exclusively analyzed secondary data from existing studies and, therefore, qualifies for exemption from informed consent or Institutional Review Board approval.
search strategy
A comprehensive search strategy was employed to identify relevant studies. Databases searched included PubMed, Embase, Scopus, Web of Science, Google Scholar, and the Cochrane Library. The bibliographic search used MeSH terms related to gene therapy and nAMD. The search was carried out from database inception through February 1, 2026, using a predefined strategy that combined terms such as “neovascular age-related macular degeneration,” “macular degeneration,” “nAMD,” “gene therapy,” “gene transfer techniques,” “genetic therapy,” “gene-based therapy,” “AAV,” “AAV,” “safety,” “treatment outcome,” “efficacy,” “adverse effects,” and “clinical trials.” These terms were searched using Boolean operators (AND, OR, NOT) to maximize the sensitivity and specificity of the search. The search aimed to include all studies investigating the effectiveness and safety of gene therapy for nAMD. Reference lists of included studies and relevant systematic reviews were manually screened for additional eligible studies. This iterative process ensured a comprehensive identification of pertinent literature.
study selection and eligibility criteria
Records identified through database searches were imported into EndNote (Clarivate) for systematic organization and to identify duplicates, which were manually reviewed and excluded. Two independent reviewers, K.Y.C. and H.C.C., screened the titles and abstracts for relevance, and full-text articles of potentially eligible studies were retrieved and assessed against the inclusion criteria. Any disagreements during the process were resolved through consensus or consulting a third reviewer, C.M.C., when necessary.
The inclusion criteria for this study adhered to the population intervention comparator outcomes study design framework. The population (P) included patients diagnosed with neovascular age-related macular degeneration, with no restrictions on patient age, location, or language of publication, ensuring comprehensive inclusion of diversepopulations affected by neovascular age-related macular degeneration. The intervention (I) of interest was gene therapy, evaluated regardless of its type, dosage, or route of administration. This inclusive approach enabled a broad assessment of various gene therapy modalities targeting nAMD, ranging from vector-based approaches to innovative delivery systems. Comparators (C) included any reference intervention, such as placebo, standard treatments (eg, anti-VEGF agents), or alternative therapeutic approaches, facilitating a relative evaluation of gene therapy against established and experimental methods. However, even single-arm trials were included.
The primary outcomes were change in best-corrected visual acuity (BCVA), change in central subfield thickness (CST), rescue anti-vascular endothelial growth factor (anti-VEGF) requirement, mortality, serious adverse events (SAEs), ocular inflammation, retinal hemorrhage, retinal pigmentary changes, subconjunctival hemorrhage, and overall AEs. Studies that did not provide data on these outcomes were excluded or partially included based on the availability of information. Secondary outcomes, discussed narratively, included biological efficacy, long-term consequences, and variability in response. Eligible studies included prospective interventional clinical studies conducted between 2015 and 2026, encompassing randomized early-phase trials and open-label dose-escalation cohorts. Randomized early-phase trials were included when available, and early-phase open-label or dose-escalation interventional studies were also eligible because they represent the foundational developmental period of ocular gene therapy. This design framework allowed inclusion of the complete translational timeline of gene therapy development for nAMD, from initial phase I safety studies (2015) to recent phase I/II expansion cohorts reported in 2026. Animal or in vitro studies, nonrandomized observational studies, and those lacking sufficient quantitative or qualitative analysis data were excluded. Rescue anti-VEGF injection was defined as any additional intravitreal anti-VEGF treatment administered following gene therapy to control persistent or recurrent disease activity.
data collection and risk of bias assessment
Two reviewers, K.Y.C. and H.C.C., performed data extraction independently using a standardized data collection form. Extracted data were entered into a pretested Microsoft Excel sheet for consistency and organization. Extracted information included study characteristics such as author names, country of study, design, sample size, age, gender, follow-up period, interventions used, route of administration, outcomes measured, and key findings.
Risk of bias in randomized trials was assessed using the revised Cochrane Risk of Bias 2 (RoB 2) tool, whereas nonrandomized interventional studies were assessed using the Risk Of Bias In Non-randomized Studies of Interventions (ROBINS-I) tool. RoB 2 evaluates domains including the randomization process, deviations from intended interventions, missing outcome data, outcome measurement, and selection of reported results. ROBINS-I evaluates domains including confounding, selection of participants, classification of interventions, deviations from intended interventions, missing data, outcome measurement, and selection of reported results. Two reviewers independently conducted these assessments, resolving disagreements through discussion.
data synthesis and analysis
The statistical analysis was conducted using Comprehensive Meta-Analysis (CMA) software version 3.7. The mean difference (MD) with 95% confidence intervals (CIs) was employed to assess efficacy values, while the logit event rate was used to analyze AE values. The pooling of continuous estimates was weighted by study sample size to account for variations in precision. If only medians (with their respective interquartile ranges) were given, the values were converted to approximate means before weighting. Other outcomes were pooled using logit meta-analysis to generate summary incidence rates with 95% CIs. For studies reporting zero events, a continuity correction (0.5 added to numerator and 1 to denominator) was applied to enable logit transformation and inclusion in the meta-analysis. Due to inconsistent reporting of follow-up duration and lack of person-time data, incidence rates could not be reliably calculated, which precluded the application of incidence rate–based meta-analytic approaches for rare event data synthesis. Therefore, AEs were analyzed as cumulative event proportions using logit transformation, a commonly adopted approach in meta-analyses of rare AEs when time-to-event data are unavailable. In this framework, statistical testing evaluates whether pooled incidence differs from a reference probability of 50% (logit = 0). Therefore, statistical significance does not imply comparative safety benefit vs control interventions and should not be interpreted as evidence of risk reduction. Interpretation focused primarily on absolute pooled event probabilities and corresponding CIs. The primary estimand of this meta-analysis was the pooled treatment-associated effect within gene-therapy cohorts, reflecting the early-phase nature of available clinical evidence. Because most included studies were single-arm or dose-escalation trials, analyses primarily synthesized within-cohort outcomes rather than direct treatment-vs-control contrasts. When randomized data were available, these were incorporated within the same hierarchical framework without generating separate comparative effect estimates. Given the early-phase nature of ocular gene therapy trials conducted between 2015 and 2026, many studies included multiple dose-escalation or expansion cohorts derived from a shared parent protocol. These multiarm structures introduce statistical dependence that violates the independence assumptions underlying conventional pooled analyses. Therefore, multilevel random-effects models using restricted maximum likelihood (REML) estimation were prespecified as the primary inferential framework, as they explicitly account for clustering of treatment arms within studies and appropriately partition within-study and between-study variance. Conventional fixed-effect pooled analyses were additionally performed and are presented as descriptive sensitivity analyses to illustrate within-arm trajectory patterns and facilitate comparison with earlier early-phase ophthalmic meta-analyses. These fixed-effect estimates should not be interpreted as primary inferential results, as they assume independence across treatment arms and may inflate precision or event incidence when multiple cohorts originate from the same trial. Accordingly, interpretation throughout this manuscript prioritizes multilevel REML estimates, while fixed-effect results are presented for contextual and descriptive purposes only. Heterogeneity across studies was assessed using the Higgins I ² statistic, which evaluates the variability between studies that cannot be explained by chance alone. Heterogeneity was categorized as low ( I ² = 0%-25%), moderate ( I ² = 26%-50%), and high ( I ² > 50%), in accordance with conventional thresholds. Visual inspection of a funnel plot and Egger’s test were utilized to evaluate potential publication bias. The funnel plot was used to assess asymmetry, while Egger’s test provided a more standardized, quantitative evaluation of bias, with a P value of <.05 indicating significant bias. Together, these methods strengthened the robustness and interpretability of the meta-analysis. Given the limited number of eligible trials, heterogeneity in study design, and frequent use of single-arm early-phase studies, subgroup analyses by vector platform were considered exploratory and hypothesis-generating rather than confirmatory. Delivery-route subgroup analyses were not emphasized because of sparse data. Exploratory subgroup comparisons were conducted using mixed-effects models in Comprehensive Meta-Analysis (CMA) software and multilevel random-effects models using REML estimation implemented in R with the metafor package. Multiarm studies were modeled hierarchically to account for clustering of dose cohorts within studies. Between-subgroup differences were evaluated using the Q -test for moderators (QM statistic), while residual heterogeneity was quantified using QE, τ ², and I ² statistics.
OUTCOMES
results
Study selection
This PRISMA 2020 flow diagram summarizes the systematic study selection process from both database searching and other methods. In the identification stage via databases and registers, a total of 12,849 articles were identified from PubMed ( n = 1797), Embase ( n = 4997), Scopus ( n = 2359), Google Scholar ( n = 1600), Cochrane Library ( n = 98), and Web of Science ( n = 1998). Before screening, 3499 duplicate records were removed, together with 994 records excluded by automation tools and 502 records removed for other reasons, resulting in 7854 records entering title and abstract screening. During screening, 5992 articles were excluded, and 1862 reports were sought for retrieval. Of these, 398 reports were not retrieved, leaving 1464 full-text reports assessed for eligibility. Full-text evaluation resulted in the exclusion of 1456 reports for predefined reasons, including invalid outcomes ( n = 597), invalid patient population ( n = 401), invalid study design ( n = 295), and invalid intervention ( n = 163). In parallel, 33 additional records were identified through other methods, including websites ( n = 21), organizations ( n = 4), and citation searching ( n = 8). Among these, 2 reports were not retrieved, and 31 reports underwent eligibility assessment; all 31 were excluded because they did not meet inclusion criteria. Following completion of the identification, screening, retrieval, and eligibility phases, a total of 8 studies met the predefined criteria and were included in the final qualitative and quantitative synthesis. This flow diagram reflects a rigorous, transparent, and methodologically robust study selection process consistent with PRISMA 2020 standards. The study selection process is illustrated in the PRISMA flow diagram ( Figure 1 ).
Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 flow diagram of study selection for gene therapy in neovascular age-related macular degeneration.
Baseline characteristics
Eight studies published between 2015 and 2026 were included, representing the early clinical development era of gene therapy for neovascular age-related macular degeneration. These consisted of prospective early-phase interventional studies, including randomized early-phase trials and open-label dose-escalation cohorts. The elapsed period from 2015 to 2026 reflects progressive maturation of vector design, delivery routes, and dose-escalation strategies across successive trial generations. Most studies were conducted in the United States, with additional trials from Australia and China. Sample sizes ranged from 6 to 68 treated participants, with follow-up durations spanning 6 weeks to 36 months.
The interventions included diverse gene therapy agents, such as RGX-314, rAAV.sFlt-1, and others, delivered via subretinal or intravitreal injection. Safety and tolerability were the primary outcomes assessed in most studies. At the same time, some also evaluated efficacy measures such as BCVA, CST, and the frequency of rescue anti-VEGF injections. Although individual cohorts were relatively small, the cumulative enrolled population across the eight included prospective interventional studies was 203 participants, reflecting the early-phase developmental nature of ocular gene therapy studies conducted between 2015 and 2026. Sample sizes, therefore, represent trial-level enrolled populations rather than pooled extension cohorts or repeated safety populations, ensuring transparent and reproducible denominator reporting across studies.
Notably, Campochiaro et al reported sustained suppression of VEGF-A and vision improvement with RGX-314, while Constable et al demonstrated the long-term stability of nAMD over 3 years with rAAV.sFlt-1. Overall, the studies suggested favorable safety and anatomical efficacy signals, although visual outcomes, rescue-treatment requirements, and follow-up durations varied. Detailed characteristics of the included studies are summarized in Table 1 .
TABLE 1
Baseline Characteristics of Included Prospective Interventional Studies (2015-2026)
| Study | Country | Design | Sample Size | Age | Gender (M:F) | Follow-up | Intervention | Route | Outcomes | Findings |
|---|---|---|---|---|---|---|---|---|---|---|
| Constable et al | USA | RCT | 8 | 79 (74-84) | 4:4 | 36 mo | rAAV.sFLT-1 |
Subretinal
Injection |
Safety and tolerability, sFLT-1 protein concentration in aqueous fluid, BCVA by ETDRS protocol, CRT measured by central subfield thickness on SD-OCT | Subretinal injection of rAAV.sFLT-1 was well-tolerated with no safety concerns and demonstrated stability of nAMD over 36 mo. |
| Rakoczy et al | Australia | RCT | 9 | 79 (74-85) | 5:4 | 12 mo | rAAV.sFLT-1 |
Subretinal
Injection |
Safety and tolerability of gene therapy | A single subretinal injection of rAAV.sFLT-1 was safe, well-tolerated, and reduced the need for rescue anti-VEGF injections in patients with nAMD. |
| Heier et al | USA | RCT | 17 | 71 (69-86) | 6:11 | 52 wk | AAV2-sFLT01 | Intravitreal injection | Safety and tolerability of gene therapy | A single intravitreal injection of AAV2-sFLT01 was safe and well-tolerated, with variable sflt01 expression and fluid reduction observed in patients with advanced nAMD. |
| Campochiaro et al | USA | RCT | 21 | 79.4 | 10:11 | 6 wk | EIAV vector RetinoStat |
Subretinal
Injection |
Safety and tolerability of gene therapy | Subretinal injection of the EIAV vector RetinoStat was safe, well-tolerated, and achieved sustained expression of endostatin and angiostatin for up to 4 y in patients with advanced nAMD. |
| Rakoczy et al | Australia | RCT | 44 | 80 ±7 | 30:14 | 36 mo | rAAV.sFLT-1 |
Subretinal
Injection |
Safety and tolerability, sFLT-1 protein concentration in aqueous fluid, BCVA by ETDRS protocol, CRT measured by central subfield thickness on SD-OCT, and safety and the need for supplemental anti-VEGF-A injections up to wk 106 | Study was unable to unequivocally confirm the existence of a biologic efficacy signal; however, it confirmed that rAAV.sFLT-1 gene delivery was well tolerated among the elderly. |
| Khanani et al | USA | RCT | 30 | 79.0 (±9.57) | 15:15 | 24 mo | Ixo-vec dose (2 × 10 11 vs 6 × 10 11) | Intravitreal injection | Safety and tolerability of gene therapy | A single intravitreal administration of Ixo-vec was well-tolerated, maintained vision, and reduced anti-VEGF injection frequency by 80%-98% in patients with nAMD. |
| Campochiaro et al | USA | RCT | 68 | 80 (74-85) | 46:22 | 2 y | RGX-314 |
Subretinal
Injection |
Safety and tolerability, RGX-314 protein concentration in aqueous fluid, BCVA by ETDRS protocol, CRT measured by central subfield thickness on SD-OCT, and safety and the need for supplemental anti-VEGF-A injections up to wk 106 | Subretinal RGX-314 gene therapy demonstrated good safety and efficacy in patients with nAMD, offering sustained VEGF-A suppression, improved vision, and reduced treatment burden with no clinically significant immune responses. |
| Sun et al | China | Phase I, open-label, single-center, dose-escalation clinical trial | 6 patients (6 eyes) | Mean 70 y | 3:3 | 12 mo (52 wk) | LX102-C01 (AAV2.7m8 vector encoding VEGF-Trap); single injection after one aflibercept pretreat | Intravitreal |
Primary: Safety (ocular/systemic AEs)
Secondary: BCVA, CST, rescue injections Exploratory: Macular atrophy, MCT, CVI |
No systemic-related AEs; mild ocular inflammation (dose-dependent); no vasculitis/endophthalmitis; 2 transient hemorrhages resolved; no rescue injections required; BCVA maintained; CST decreased; MCT significantly decreased ( P =.03); CVI unchanged; no macular atrophy progression. |
Sample sizes represent enrolled participants reported in the primary trial publications. Where studies included multiple dose cohorts or extension phases, participants were counted once at the trial level to avoid denominator inflation.
AMD = age-related macular degeneration; AE = adverse event; anti-VEGF = anti-vascular endothelial growth factor; BCVA = best-corrected visual acuity; CRT = central retinal thickness; CST = central subfield thickness; CVI = choroidal vascularity index; EIAV = equine infectious anemia virus; ETDRS = Early Treatment Diabetic Retinopathy Study; Ixo-vec = ixoberogene soroparvovec; MCT = mean choroidal thickness; nAMD = neovascular age-related macular degeneration; NR = not reported; rAAV = recombinant adeno-associated virus; RCT = randomized controlled trial; SD-OCT = spectral-domain optical coherence tomography; sFLT-1 = soluble fms-like tyrosine kinase-1; VEGF = vascular endothelial growth factor.
Table 1 summarizes the baseline characteristics of the included prospective interventional studies. The included studies span 2015 to 2026 , capturing the earliest human gene-therapy investigations through more recent expansion cohorts. The included studies comprised randomized early-phase trials and open-label dose-escalation interventional cohorts, predominantly in elderly populations representative of typical patients with neovascular age-related macular degeneration. Despite shared disease indication, the studies differed substantially in vector platform, transgene target, dosing strategy, and delivery route, highlighting considerable clinical heterogeneity across interventions. This heterogeneity in intervention design and follow-up duration is an important contextual factor when interpreting pooled efficacy and safety estimates.
Risk of bias
Domain-level judgments were made independently across all bias domains according to prespecified criteria. A detailed account of the methodological characteristics of each study, including what was explicitly reported and how these aspects informed the corresponding risk-of-bias judgments, is provided in Supplemental Table 1 (RoB 2 assessment for randomized trials) and Supplemental Table 2 (ROBINS-I assessment for nonrandomized studies). This supplemental table transparently outlines the rationale underpinning each domain-level and overall assessment.
Randomized early-phase trials
The methodological quality of the randomized early-phase trials was assessed using the Cochrane Risk of Bias 2 (RoB 2) tool across five domains, as shown in Figure 2 . Overall, most randomized early-phase trials ,,,, were judged to have ‘some concerns,’ primarily driven by Domain 1 (randomization process) and Domain 5 (selection of reported results). In several of these studies, insufficient detail was provided regarding allocation concealment and sequence generation, limiting confidence in the adequacy of randomization. Additionally, publicly accessible protocols or prespecified statistical analysis plans were not consistently available, raising minor concerns regarding selective reporting. Domain 3 (missing outcome data) showed some concerns in selected trials where attrition handling or imputation methods were not clearly described, although dropout rates were generally low. Domains 2 (deviations from intended interventions) and 4 (measurement of outcomes) were consistently rated as low risk, as interventions were clearly defined and outcomes such as BCVA, CST, and AEs were measured using standardized, objective methods. In contrast, the more recent multicenter trials , demonstrated low risk across all domains, reflecting improved reporting transparency and trial methodology. Overall, no study was judged to be at high risk of bias.
Risk-of-bias traffic light plot for randomized early-phase trials assessed using the revised Cochrane Risk of Bias tool version 2 (RoB 2).
Nonrandomized cohort studies
The nonrandomized, open-label, dose-escalation trial by Sun et al was assessed using the ROBINS-I tool as shown in Figure 3 . Overall, the study was judged to have moderate risk of bias , primarily due to confounding and participant selection concerns inherent to early-phase, single-arm designs. Bias due to confounding (Domain 1) was rated as moderate because the absence of a comparator arm limits the ability to distinguish treatment effects from natural disease progression or regression to the mean. Selection bias (Domain 2) was also judged moderate, as inclusion criteria and single-center recruitment may introduce systematic differences between enrolled patients and the broader nAMD population. Classification of interventions (Domain 3) was considered low risk given the clearly defined dosing protocol. Bias due to deviations from intended interventions (Domain 4) was rated low, as treatment delivery and monitoring followed a structured protocol. Missing data (Domain 5) were minimal and adequately reported. Measurement of outcomes (Domain 6) was judged low risk because standardized Early Treatment Diabetic Retinopathy Study (ETDRS) and optical coherence tomography assessments were used. Selection of reported results (Domain 7) raised moderate concerns due to limited publicly available prespecified analytic plans.
Risk Of Bias In Non-randomized Studies of Interventions (ROBINS-I) traffic light plot showing risk of-bias assessment for the nonrandomized phase I trial.
Efficacy and safety
To provide an integrated overview of pooled estimates across all evaluated endpoints, Table 2 summarizes the results of the multilevel random-effects meta-analyses incorporating clustering of multiarm studies using REML. Across efficacy outcomes, CST demonstrated a statistically significant pooled mean reduction, whereas BCVA showed no significant improvement under hierarchical modeling despite favorable fixed-effect estimates. Regarding safety endpoints, overall AEs and SAEs demonstrated moderate pooled incidences with negligible between-study heterogeneity. Inflammation, retinal hemorrhage, pigmentary changes, and subconjunctival hemorrhage showed low-to-moderate event probabilities with minimal residual heterogeneity after accounting for intrastudy correlation. Notably, between-study variance ( τ ² or σ ²) was negligible for most safety outcomes, indicating consistent event rates across vector platforms and dose cohorts. This consolidated summary table provides a global statistical framework to contextualize the outcome-specific analyses presented below and facilitates comparison between anatomical efficacy and safety signals across gene therapy platforms. Detailed arm-level raw data extracted from the included studies are provided in Supplemental Tables 3 to 12 to enhance transparency and enable identification of influential study arms.
TABLE 2
Summary of Multilevel Random-Effects Meta-Analyses (REML)
| Outcome | k | Model | Pooled Effect (Logit or MD) | SE | 95% CI | z | P Value | Tau² ( σ ²) | Q (df) | Q P Value |
|---|---|---|---|---|---|---|---|---|---|---|
| BCVA (mean change) | 11 | rma.mv | 0.5385 | 4.0399 | −7.3795 to 8.4565 | 0.1333 | .894 | 93.6673 | 176.34 (10) | <.0001 |
| CST (µm) | 15 | rma.mv | 37.1272 | 5.3544 | 26.6327-47.6217 | 6.9339 | <.0001 | 63.301 | 17.27 (14) | .242 |
| Mortality (logit) | 9 | rma.mv | −2.4016 | 0.3717 | −3.1301 to −1.6731 | −6.4611 | <.0001 | 0 | 2.49 (8) | .962 |
| Serious adverse events (SAEs) | 14 | rma.mv | −1.3143 | 0.1843 | −1.6756 to −0.9530 | −7.1302 | <.0001 | 0 | 5.18 (13) | .971 |
| Inflammation | 15 | rma.mv | −1.4056 | 0.1909 | −1.7798 to −1.0314 | −7.3617 | <.0001 | 0 | 7.11 (14) | .9306 |
| Rescue anti-VEGF required | 12 | rma.mv | −0.2423 | 0.2105 | −0.6549 to 0.1703 | −1.1511 | .2497 | 0 | 10.34 (11) | .5003 |
| Retinal hemorrhage | 11 | rma.mv | −1.9979 | 0.3232 | −2.6314 to −1.3643 | −6.1809 | <.0001 | 0 | 2.04 (10) | .996 |
| Overall AE incidence | 17 | rma.mv | −0.1601 | 0.1562 | −0.4662 to 0.1460 | −1.0252 | .3053 | 0.0069 | 8.38 (16) | .9367 |
Visual acuity
The fixed-effect subgroup meta-analysis ( Figure 4 ) evaluating BCVA change following gene therapy for nAMD demonstrated a pooled MD of 11.69 ETDRS letters (95% CI: 11.03-12.35; P <.001) across 11 treatment arms. Clinically, an improvement of ≥10 ETDRS letters corresponds to approximately two lines of visual gain and is widely regarded as functionally meaningful. Therefore, the fixed-effect pooled estimate suggests a favorable descriptive visual-acuity signal, but this estimate should not be interpreted as the primary inferential result. Subgroup analyses revealed relatively consistent treatment effects across platforms. The pooled MD was 11.61 letters for rAAV.sFlt-1 ( k = 4), 11.93 letters for ixoberogene soroparvovec (Ixo-vec) ( k = 4), 10.72 letters for LX102 ( k = 2), and 13.00 letters for RGX-314 ( k = 1). These fixed-effect sensitivity findings indicate broadly comparable descriptive estimates among AAV-based anti-VEGF constructs. Although multiple dose cohorts were evaluated (eg, HD-1 vs HD-2; 2 × 10¹¹ vs 6 × 10¹¹ vg), no clear linear dose-dependent trend was observed in pooled subgroup estimates, suggesting that visual response may not increase proportionally with higher vector doses and may instead reflect baseline disease characteristics or cohort size. Statistical heterogeneity within subgroups was low ( I ² ≈ 0%), and the overall Q -test was nonsignificant ( Q = 9.35, df = 10, P =.50), supporting homogeneity under the fixed-effect framework. Larger and more precise cohorts, particularly the Ixo-vec and RGX-314 arms, contributed greater statistical weight, thereby influencing pooled precision. Under the fixed-effect sensitivity model, BCVA change appeared favorable; however, this estimate should be interpreted descriptively only because it does not fully account for clustering of multiple treatment arms within parent studies or between-cohort heterogeneity. The primary multilevel restricted maximum likelihood model did not demonstrate a statistically significant pooled BCVA improvement, indicating that current evidence does not establish a consistent functional visual benefit.
