Retinal vascular diseases, such as neovascular age-related macular degeneration (nAMD) and diabetic macular edema (DME), have long relied on vascular endothelial growth factor (VEGF) inhibitors as the first-line standard of care. By neutralizing VEGF, these therapies successfully reduce fluid leakage and halt abnormal blood vessel growth. However, clinicians frequently encounter a frustrating therapeutic plateau: approximately 30% to 40% of DME patients and a substantial portion of nAMD patients fail to respond adequately, experiencing persistent subretinal fluid and recurrent leakage.
This phenomenon of anti-VEGF resistance presents a major clinical hurdle. When fluid accumulation becomes chronic, it causes irreversible damage to the retinal architecture, leading to subretinal fibrosis and permanent vision loss. To overcome this resistance, recent research has focused on uncovering the alternative molecular pathways driving non-response, developing multi-target therapeutics, and leveraging deep learning to predict patient outcomes before irreversible damage occurs.
The Molecular and Structural Drivers of Non-Response
To understand why single-target VEGF inhibition fails, we must look at the broader microenvironment of the diseased retina. In advanced or recalcitrant stages of vascular disease, leakage is rarely driven by a single growth factor. For instance, ignoring secondary pathways like Angiopoietin-2 (Ang-2) leaves underlying vascular destabilization unaddressed. As noted by Franziska Eckardt et al. in their 2025 study, targeting VEGF alone is often insufficient to stabilize the vascular bed in all patients. Similarly, Viktoria Deiters et al. observed that in chronic DME, persistent fluid despite long-term therapy indicates that VEGF ceases to be the sole driver of vascular permeability (Real-World Data on Morphological and Functional Responses After Switching to Faricimab in Recalcitrant, Chronic Diabetic Macular Edema).
Beyond Ang-2, other molecular cascades contribute to this resistance. Alternative targets, such as the specific signatures of transfer RNA-derived small RNAs (tsRNAs) identified by Yingqian Peng et al. (Small RNA Sequencing Reveals Transfer RNA-derived Small RNA Expression Profiles in Retinal Neovascularization) and the Delta-like 4 pathway explored by Xianchun Yan et al. (Endothelial cells-targeted soluble human Delta-like 4 suppresses both physiological and pathological ocular angiogenesis), highlight a highly redundant angiogenic network that can bypass standard anti-VEGF blockades.
Furthermore, specific disease phenotypes introduce unique mechanical and structural challenges:
- Choroidal Thickening: Conditions characterized by abnormal choroidal thickening, such as pachychoroid neovasculopathy (PNV) and central serous chorioretinopathy (CSCR), exhibit heightened resistance due to distinct mechanical and inflammatory drivers (Benedikt Schworm et al., Response of neovascular central serous chorioretinopathy to an extended upload of anti-VEGF agents). In these thick-choroid diseases, standard dosing schedules fail to clear subretinal fluid effectively (Nikolaus Luft et al., Ranibizumab non-response in pachychoroid neovasculopathy: Effects of switching to aflibercept).
- Systemic Factors: Ocular drug response does not occur in a vacuum. For example, Wei Wu et al. demonstrated that systemic renal dysfunction correlates with highly variable clinical responses to intravitreal conbercept therapy (Renal dysfunction associated with clinical response to intravitreal conbercept therapy for diabetic macular edema).
- Drug-Induced Paradoxes: Most strikingly, Mingyue Luo et al. showed that when retinal vascular blockages are induced as a side effect of systemic anti-angiogenic tyrosine kinase inhibitors, local anti-VEGF injections can be ineffective or counterproductive, because the underlying pathology stems from direct vascular damage rather than excess growth factor signaling (Central retinal vein occlusion in patients with metastatic solid tumors on tyrosine kinase inhibitors: a report of case series and literature review).
Overcoming Resistance via Multi-Targeting and Dose Escalation
To break through the therapeutic plateau, researchers have shifted toward two primary strategies: increasing the molecular dose or expanding the therapeutic target profile.
The first approach involves intensifying the anti-VEGF signal to overcome receptor saturation or high fluid turnover. Michael Hafner et al. utilized deep learning to demonstrate that escalating the dose of aflibercept from 2 mg to 8 mg significantly reduces persistent intraretinal fluid and pigment epithelial detachment in therapy-resistant nAMD patients, allowing for extended treatment intervals (Deep learning-assisted analysis of biomarker changes after increase of dosing from aflibercept 2 mg to 8 mg in therapy-resistant neovascular age-related macular degeneration).
The second, more robust strategy relies on bispecific antibodies that simultaneously target multiple pathways. Faricimab, which inhibits both VEGF-A and Ang-2, has emerged as a powerful tool for recalcitrant cases. By dual-targeting, faricimab stabilizes the endothelial wall, tightens cell junctions, and reduces inflammation more effectively than pure anti-VEGF agents.
In a series of studies, Franziska Eckardt et al. showed that switching treatment-resistant nAMD patients to faricimab achieved significant anatomical improvements and extended the treatment interval by over 50% to 60%, drastically reducing patient burden (Retinal and choroidal efficacy of switching treatment to faricimab in recalcitrant neovascular age related macular degeneration; Deep learning assisted analysis of biomarker changes in refractory neovascular AMD after switch to faricimab). Similarly, Yingyi Lu et al. demonstrated the safety and efficacy of RC28-E, a novel bispecific antibody designed to prevent subretinal fibrosis and permanent tissue damage by moving beyond single growth factor targeting (Safety and Efficacy of Multiple Escalating Doses of RC28-E for Neovascular Age-Related Macular Degeneration: A Phase 1b Trial).
Standard Anti-VEGF Therapy:
[VEGF Signal] ───┤ [Anti-VEGF Agent] ───> Persistent Fluid / Destabilized Endothelium
Dual-Pathway Bispecific Therapy (e.g., Faricimab):
[VEGF Signal] ───┤ ┌─────────────────┐
│ Bispecific Agent│ ───> Dry Macula / Stabilized Endothelium
[Ang-2 Signal] ───┤ └─────────────────┘
Predicting Non-Response with Deep Learning and Biomarkers
Because chronic fluid causes cumulative, irreversible damage, identifying non-responders early is critical. Researchers are leveraging machine learning and advanced optical coherence tomography (OCT) imaging to predict therapeutic outcomes before structural failure occurs.
Rather than looking at fluid volume alone, recent models focus on structural integrity and inflammatory markers. Fengjiao Li et al. identified that baseline counts of hyperreflective foci (HRF)—microscopic indicators of inflammation visible on OCT—independently predict visual recovery following anti-VEGF therapy in DME (Predictive value of hyperreflective foci for anti-VEGF therapeutic outcomes in different subtypes of diabetic macular edema: a retrospective analysis).
To scale these insights, Xian Song et al. developed a deep learning pipeline to automatically quantify the integrity of the outer retinal layers, specifically the ellipsoid zone and external limiting membrane. The authors proved that photoreceptor disruption is a far more potent predictor of persistent disease activity than fluid volume alone (Automatic quantification of retinal photoreceptor integrity to predict persistent disease activity in neovascular age-related macular degeneration using deep learning).
By combining these structural OCT features with systemic blood biomarkers, multimodal machine learning models can now forecast anatomical responses with high accuracy. Wenrui Lu et al. built a predictive model achieving an area under the curve (AUC) of 0.90, allowing clinicians to flag potential non-responders before initiating standard first-line injections (A machine learning model for predicting anatomical response to Anti-VEGF therapy in diabetic macular edema).
Similarly, in challenging conditions like chronic central serous chorioretinopathy, Tae Keun Yoo et al. developed DeepPDT-Net, a two-stage transfer learning framework that outperforms human specialists in predicting treatment success (DeepPDT-Net: predicting the outcome of photodynamic therapy for chronic central serous chorioretinopathy using two-stage multimodal transfer learning). These AI-driven tools shift the clinical paradigm from reactive drug-switching to proactive, personalized intervention.
Key Foundational Work
Ranibizumab non-response in pachychoroid neovasculopathy: Effects of switching to aflibercept (2020) — contributed 2 concept elements to this direction.
Small RNA Sequencing Reveals Transfer RNA-derived Small RNA Expression Profiles in Retinal Neovascularization (2020) — contributed 1 concept elements to this direction.
Endothelial cells-targeted soluble human Delta-like 4 suppresses both physiological and pathological ocular angiogenesis (2015) — contributed 1 concept elements to this direction.