Signals of Drug-Related Retinal Fibrosis: Findings From a 20-Year Disproportionality Analysis of Global Spontaneous Reports

Abstract

Purpose

This study aimed to identify potential drugs associated with drug-related retinal fibrosis (RF) and to analyze their risk characteristics systematically.

Methods

This study used a retrospective disproportionality analysis to identify significant drug signals associated with RF from over 17 million reports in the U.S. Food and Drug Administration (FDA) Adverse Event Reporting System (FAERS) database. The identified drugs were categorized by therapeutic class, risk level, latency period, and key subgroups (e.g., age, gender, off-label use).

Results

Sixteen drugs showed a significant association with RF, including anti-neovascularization agents, corticosteroids, and immunosuppressants. Risk severity varied: Verteporfin, ranibizumab, and brolucizumab had the highest risk, and vigabatrin and sildenafil had a relatively lower risk. The latency periods differed significantly, ranging from a median of 950 days for fluocinolone acetonide to 13 days for triamcinolone. Subgroup analysis revealed that off-label use increased the risk of RF and uncovered distinct susceptibility patterns: Adolescents (0–17 years) faced a higher risk when using specific drug classes such as corticosteroids, whereas elderly patients (≥65 years) were more susceptible to RF when using anti-neovascularization agents. Females showed a higher risk associated with immunosuppressants.

Conclusions

This study presents a list of drugs associated with RF, providing a valuable reference to guide rational drug use, preserve visual acuity, and prevent irreversible structural damage.

Translational Relevance

Identifying specific drugs associated with RF is crucial for preventing drug-related RF and developing personalized monitoring strategies.

Introduction

Retinal fibrosis (RF) is a common pathological change in various retinal diseases, including diabetic retinopathy and age-related macular degeneration (AMD).¹ This pathological process can lead to alterations in retinal structure, visual decline, and even retinal detachment, resulting in irreversible vision loss.^2,3 In developed countries, RF is considered a relatively common cause of visual impairment.⁴ Globally, its incidence is increasing, causing severe visual harm and imposing a significant burden on healthcare systems.⁵ RF is typically a repair response triggered by stimuli such as retinal inflammation, ischemia, trauma, or surgery; yet, its mechanisms remain incompletely understood.^6–9 As the choice and complexity of drugs in clinical treatment grow, the potential risk of drug-related RF is gaining increasing attention. Notably, our research team has previously demonstrated that metformin can inhibit pathological retinal neovascularization in experimental neovascular age-related macular degeneration (nAMD) but paradoxically promotes the fibrotic process.¹⁰ This finding provided an important impetus for our current study. RF is characterized by an insidious onset and is difficult to reverse. Because effective treatments remain limited, identifying drug-associated risks and implementing proactive preventive measures are particularly crucial.

In recent years, advancements in pharmacovigilance and real-world evidence studies have provided a more robust basis for drug safety assessment. Our research team, along with others, has previously utilized data from the publicly available U.S. Food and Drug Administration (FDA) Adverse Event Reporting System (FAERS) to evaluate the relationships between specific drugs and associated diseases.^11–13 FAERS is a comprehensive global platform for post-marketing drug adverse event surveillance, compiling spontaneous reports from healthcare professionals, pharmaceutical companies, and patients.^14–16 It provides a valuable real-world data resource for studying drug-related adverse reactions, thereby enhancing our understanding of drug safety in these settings and serving as a crucial reference for clinical practice.

To systematically evaluate the potential association between commonly used drugs and the risk of RF, this study leveraged the FAERS database to conduct large-scale real-world data mining, applying various signal detection algorithms to identify risk signals. Through further classification and analysis of these suspected signals, a list of potentially high-risk drugs was ultimately generated for clinicians’ reference. The value of this work lies in transforming real-world data into key evidence for clinical decision-making, assisting physicians in identifying and mitigating the risk of drug-related RF. Concurrently, it also provides a significant evidence base for regulatory agencies to formulate more precise pharmacovigilance strategies.

Methods

Data Source

The data for this study were sourced from the FAERS database, from which we extracted and analyzed all data records from the first quarter of 2004 to the fourth quarter of 2024. FAERS is a global passive surveillance database, with data primarily submitted spontaneously by healthcare professionals, pharmacists, and consumers.^15,17 The data structure of the system is comprised of seven core modules: DEMO (patient demographic and administrative information), DRUG (associated drugs and biologics), REAC (adverse events), OUTC (patient outcomes), RPSR (report sources), THER (drug therapy start and end dates), and INDI (indications for drug use), thus providing multidimensional data support for this research. To minimize confounding bias from concomitant medications, such as the concurrent use of anti-vascular endothelial growth factor (VEGF) agents and steroids, we included only drugs listed as the “Primary Suspect” (PS) in the disproportionality analysis. We utilized the DrugBank (https://go.drugbank.com/) database to query the official package inserts for each drug and determine their approved indications.¹⁸ Consequently, any reported indication that did not match the approved uses listed in DrugBank was categorized as “off-label use.”

Adverse Drug Reaction Identification

In this study, all adverse drug reaction events were standardized according to the Medical Dictionary for Regulatory Activities (MedDRA) terminology (version 27.0; http://www.meddra.org/).¹⁹ We first encoded all ADRs using MedDRA Preferred Terms (PTs). Subsequently, following a method established in previous research,²⁰ we applied a narrow-scope Standardized MedDRA Query (SMQ) that included the terms retinal scar (PT code: 10038895), macular scar (PT code: 10063185), macular fibrosis (PT code: 10071392), subretinal fibrosis (PT code: 10062958), epiretinal membrane (PT code: 10064697), and retinal fibrosis (PT code: 10071391) to identify all PTs related to RF precisely.

Statistical Analysis

To systematically mine for potential association signals between drugs and RF, this study applied four mainstream imbalance analysis algorithms based on previous studies: two non-Bayesian methods (reporting odds ratio [ROR] and proportional reporting ratio [PRR]) and two Bayesian methods (Bayesian confidence propagation neural network [BCPNN] and multi-item gamma Poisson shrinker [MGPS]).^21,22 Among the non-Bayesian methods, the PRR is known for its high sensitivity, but its drawback is a tendency to generate false-positive signals, particularly when case numbers are small.²³ In contrast, the ROR, as a more reliable estimator of the rate or hazard ratio, exhibits less bias.²⁴ The Bayesian methods, however, offer optimized solutions for these limitations. For example, the advantage of the BCPNN lies in its ability to maintain result stability even with a limited number of reports, effectively addressing the small sample size problem.²⁵ The MGPS, in turn, demonstrates superior performance in identifying signals related to rare or unexpected events.²⁶ All of these algorithms are based on a 2 × 2 contingency table and compare the target drug–event combination against all other drug–event pairs (see Supplementary Tables S1 and S2 for details).

To ensure the robustness and reliability of the signals, an association between a drug and RF was classified as a positive signal of interest only if it simultaneously met the threshold criteria for the following four methods: (1) ROR ≥ 3 and lower limit of the 95% confidence interval (CI) > 1; (2) PRR ≥ 3 and lower limit of the 95% CI > 1; (3) BCPNN information component at the 2.5th percentile (IC₀₂₅) > 0; and (4) MGPS empirical Bayesian geometric mean at the 5th percentile (EBGM₀₅) > 2 and a > 0. All positive signals identified through the above criteria were subjected to further characteristic analysis. First, we evaluated and compared the differences in the latency period of RF associated with different drug classes. Second, the risk of RF was stratified based on the calculated BCPNN values. All data processing, statistical analysis, and visualization were performed using SPSS Statistics 26.0 (IBM, Chicago, IL), Prism 10.1.2 (GraphPad, Boston, MA), Excel 2019 (Microsoft, Redmond, WA), and R 4.2.2 (R Foundation for Statistical Computing, Vienna, Austria). Analyses in R were primarily conducted using the ggplot2 (v3.4.4), ggrepel (v0.9.4), dplyr (v1.1.4), and DescTools (v0.99.52) packages. All statistical tests were two sided, and P < 0.05 was considered statistically significant.

Results

Baseline Characteristics of RF Patients

The initial dataset, collected from January 2004 to December 2024, was comprised of 22,249,476 records. After the removal of duplicate reports, 18,627,667 individual adverse event reports were retained. Among these, 1578 patients and 455 drugs were suspected to be associated with RF events. After further excluding drugs that did not meet the criteria of the four disproportionality analysis methods and merging duplicate entries for the same drug listed under different trade names, a final total of 16 unique drugs were included in the analysis. The data filtering flowchart is presented in Figure 1.

Figure 1.

Raw data cleaning flowchart for drug-related RF. The screening process to extract data on patients with drug-related RF and their medication usage from the FAERS database was developed using Microsoft Visio 2021, illustrating the methodology for case identification and data analysis.

A total of 1578 cases of RF were reported in the FAERS database. The mean patient age was 63.7 ± 16.8 years, and the mean weight was 75.9 ± 21.2 kg. The age and sex distribution (Fig. 2A) showed a relatively balanced distribution across age groups, with no significant sex difference: Females accounted for 48.8% (n = 770) and males for 35.2% (n = 556). The number of drug-related RF reports has shown an overall upward trend since the beginning of data collection (Fig. 2B). Reports primarily originated from physicians (38.28%, n = 604), followed by consumers (36.44%, n = 575) (Fig. 2C). The most common route of administration was intraocular (16.54%, n = 261), followed by oral (16.48%, n = 260) (Fig. 2D). The most frequently reported outcomes were hospitalization (13.12%, n = 207) and disability (6.78%, n = 107) (Fig. 2E). Geographically, the United States reported the highest number of cases (42.4%, n = 669), followed by Germany (6.97%, n = 110), Japan (6.27%, n = 99), Switzerland (6.21%, n = 98), United Kingdom (5.01%, n = 79), and France (2.85%, n = 45). Countries such as China (1.01%, n = 16), Singapore (0.25%, n = 4), and Sweden (0.13%, n = 2) accounted for a smaller proportion of cases (Fig. 2F). Detailed demographic information is provided in Supplementary Table S3.

Figure 2.

Distribution of baseline data for patients reporting adverse events of drug-related RF in the FAERS database. (A) Age and gender distribution of RF patients. (B) Temporal trend of RF reports over the study period. (C) Reports based on reporter type. (D) Distribution of drug administration routes among RF cases. (E) Assessment of the seriousness of adverse drug reactions associated with RF. (F) Geographic distribution of RF reports.

Identification of Positive Drug Signals Through Disproportionality Analysis

Disproportionality analysis was performed on 455 drugs associated with RF that had three or more reported cases. After de-duplication of drugs listed under different trade names, 16 drugs with positive signals were identified (Fig. 3A). Among ophthalmic anti-neovascularization agents, ranibizumab had the highest reporting frequency, with 435 adverse event reports, followed by aflibercept (n = 236) and bevacizumab (n = 95). Other anti-neovascularization agents with fewer reports included verteporfin (n = 60), brolucizumab (n = 35), and faricimab (n = 16). Other drugs, such as latanoprost (n = 8), sildenafil (n = 8), vigabatrin (n = 7), ocriplasmin (n = 5), and exemestane (n = 3), were reported less frequently.

Figure 3.

Analysis of grouped drug signals for RF and their Anatomical Therapeutic Chemical (ATC) classification. (A) Grouped bubble chart of positive drug signals for RF, where different colors denote distinct drug categories and the size of each bubble is proportional to the number of adverse event reports (n). (B) Heatmap of signal values, in which a darker color corresponds to a higher signal value, indicating a greater associated risk of RF.

Corticosteroids were also significantly associated with RF, with dexamethasone (n = 50) and triamcinolone (n = 43) being the most reported, followed by fluocinolone acetonide (n = 10). Among immunosuppressants, two drugs showed strong positive signals for RF: fingolimod, which was the most reported (n = 34), and pegcetacoplan (n = 15).

Classification Based on Therapeutic Purposes

Among anti-neovascularization agents, drugs associated with drug-related RF included verteporfin (ROR = 453.06; range, 349.56–587.21; P < 0.001), ranibizumab (ROR = 237.66; range, 213.02–265.14; P < 0.001), brolucizumab (ROR = 102.44; range, 73.26–143.26; P < 0.001), aflibercept (ROR = 93.63; range, 81.58–107.46; P < 0.001), faricimab (ROR = 55.96; range, 34.19–91.59; P < 0.001), and bevacizumab (ROR = 11.16; range, 9.07–13.72; P < 0.001). Corticosteroids included fluocinolone acetonide (ROR = 216.26; range, 115.91–403.5; P < 0.001), triamcinolone (ROR = 29.91; range, 22.09–40.49; P < 0.001), and dexamethasone (ROR=9.81; range, 7.4–12.99; P < 0.001). Immunosuppressants included pegcetacoplan (ROR = 87.32; range, 52.49–145.28; P < 0.001) and fingolimod (ROR = 4.39; range, 3.13–6.17; P < 0.001). Other drugs identified include ocriplasmin (ROR = 73.64; range, 30.58–177.33; P < 0.001), exemestane (ROR = 22.66; range, 7.3–70.37; P < 0.001), latanoprost (ROR = 8.42; range, 4.2–16.87; P < 0.001), vigabatrin (ROR = 7.14; range, 3.4–15.01; P < 0.001), and sildenafil (ROR = 4.03; range, 2.01–8.08; P < 0.001). More detailed information is available in Figure 3B and Supplementary Table S4. Furthermore, in a subgroup analysis limited to reports submitted by physicians, all 16 drugs also showed positive signals (Fig. 4A).

Figure 4.

Stratification analysis of adverse event reports for drug-related RF. (A) Heatmap of signal strengths, where darker colors correspond to higher signal values and thus greater associated risks. (B) A subgroup analysis of anti-angiogenic drugs, stratified by on-label and off-label use. (C) Subgroup analysis stratified by patient age and sex. (D) Chord diagram illustrating the distribution of adverse event reports across demographic subgroups (age and sex) for each implicated drug. In the analyses for panels (B) and (C), the displayed RORs indicate drugs that meet the criteria for a positive signal in the respective subgroups.

Subgroup Analysis Based on Drug Indications

We conducted a subgroup analysis to evaluate the risk of RF associated with on-label versus off-label use of six anti-neovascularization drugs with neovascular fundus disease as their primary indication. The results showed that the risk of RF was higher when the drugs aflibercept, verteporfin, brolucizumab, and faricimab were used off-label compared to their approved indications. Specifically, under conditions of off-label use, the ROR values for these drugs increased significantly, exceeding the positive signal thresholds for all four disproportionality analysis methods (Fig. 4B).

To further explore the influence of patient characteristics on RF, we stratified the study population into three main age groups: adolescents (0–17 years), adults (18–64 years), and elderly (≥65 years). In adolescent patients, triamcinolone and bevacizumab showed strong risk signals for RF. In the adult group, fluocinolone acetonide, verteporfin, ranibizumab, brolucizumab, and aflibercept were associated with a higher risk of RF. Among elderly patients, verteporfin, ranibizumab, ocriplasmin, pegcetacoplan, and aflibercept were linked to a relatively higher risk of RF. Furthermore, in the sex-based subgroup analysis, verteporfin, ranibizumab, fluocinolone acetonide, pegcetacoplan, and brolucizumab exhibited higher risk signals for RF in female subjects. In comparison, ocriplasmin and aflibercept showed stronger risk signals for RF in male subjects. Notably, the signals for ranibizumab and verteporfin were extremely strong across the adult, elderly, male, and female subgroups (Fig. 4C).

Analysis of Report Counts by Demographic Subgroup

After establishing the total number of adverse event reports for each drug, we further analyzed the composition of these reports across different population subgroups. The results indicated that the pediatric and adolescent population (0–17 years) accounted for a very small proportion of reports, with only a small number for triamcinolone, dexamethasone, and bevacizumab. For major anti-VEGF drugs (e.g., aflibercept, ranibizumab) and intraocular corticosteroids such as dexamethasone, adverse event reports primarily originated from the elderly patients, with this age group contributing a consistently high proportion. Additionally, differences were observed in the sex composition of the reports. For example, all reports for the two immunosuppressants associated with RF, fingolimod and pegcetacoplan, originated exclusively from female patients, whereas the proportion of males was higher among reports for aflibercept (Fig. 4D).

Classification Based on Time to Onset of Adverse Reactions

Following the classification of drug-related adverse reactions, we conducted further analysis of the data using quartiles. Regarding the time to onset of these adverse reactions, the drugs with the shortest median latencies were triamcinolone (median = 13 days), dexamethasone (median = 29 days), and verteporfin (median = 38 days). Conversely, the drugs with the longest median latency were fluocinolone acetonide (median = 950 days), sildenafil (median = 882 days), bevacizumab (median = 851 days), and ocriplasmin (median = 721 days) (Fig. 5A). Furthermore, analysis of the cumulative risk curves and violin plots revealed no significant difference (P = 0.47) between the two major categories of anti-neovascularization agents (mean = 309.75 days) and non–anti-neovascularization agents (mean = 239.97 days) (Figs. 5B, 5C).

Figure 5.

Ranking of drugs according to the time to onset of RF, with a timeline of ocular adverse reactions associated with related drugs. (A) Median induction time for ocular adverse events related to each drug. (B) Kaplan–Meier curve illustrating the timeline of ocular adverse reactions associated with related drugs. (C) Mean induction time for these adverse reactions.

Subgroup Analysis Based on RF Subtypes

To further elucidate the impact of clinical heterogeneity on the associations between drug classes and specific fibrotic subtypes, we performed a subgroup analysis based on six MedDRA PTs: retinal scar, subretinal fibrosis, macular fibrosis, macular scar, epiretinal membrane (ERM), and retinal fibrosis. These results revealed significant heterogeneity in risk profiles across therapeutic categories (Supplementary Table S5). Among anti-neovascular agents, ranibizumab exhibited the broadest and most intense fibrotic signals, showing significant positive associations across all six subtypes. Notably, it displayed elevated risks for subretinal fibrosis (ROR = 800.85; range, 589.63–1087.74; P < 0.01) and macular scar (ROR = 516.35; range, 380.74–700.26; P < 0.01), whereas its signal for ERM was relatively lower (ROR = 48.06; range, 28.73–80.40; P < 0.01). Similarly, aflibercept and bevacizumab presented broad risk profiles. In contrast, verteporfin demonstrated high specificity for scarring outcomes, yielding exceptionally strong signals for subretinal fibrosis (ROR = 2004.63; range, 1311.59–3063.85; P < 0.01) and macular scar (ROR = 1200.11; range, 705.39–2041.79; P < 0.01). The newer agent, faricimab, exhibited a unique, localized risk profile, being associated only with subretinal fibrosis (ROR = 330.23; range, 168.81–646.02; P < 0.01) and ERM (ROR = 173.62; range, 71.31–422.71; P < 0.01), with no detectable signals in other subtypes. In the immunosuppressants group, pegcetacoplan showed a signal distribution distinct from that of anti-neovascular agents. Specifically, it was not associated with serious structural damage, such as subretinal fibrosis or macular scar, but it displayed a robust and specific signal for ERM (ROR = 596.37; range, 314.72–1130.09; P < 0.01). In the corticosteroids group, fluocinolone acetonide was significantly associated with macular fibrosis (ROR = 329.78; range, 147.38–737.90; P < 0.01). Additionally, ocriplasmin demonstrated a specific signal for macular fibrosis (ROR = 186.44; range, 77.27–449.87; P < 0.01).

Discussion

To systematically investigate the association between drugs and RF, this study conducted a comprehensive analysis of large-scale real-world data from the FAERS database, spanning from the first quarter of 2004 to the fourth quarter of 2024. Utilizing four disproportionality analysis algorithms (ROR, PRR, BCPNN, and MGPS), the study performed in-depth signal mining and cross-validation, ultimately identifying 16 drugs potentially associated with RF. Building on this, we conducted further in-depth analysis from multiple dimensions, including the therapeutic use of the drugs, the time to onset of adverse reactions, and their primary indications, aiming to provide a clinically relevant drug risk profile. Prior to this, few studies have utilized such a vast dataset from FAERS to systematically evaluate the risk of drug-related RF. Therefore, this study can be considered one of the first systematic assessments of drug-related RF based on FAERS big data.

From a therapeutic perspective, the findings of this study reveal a potential association between RF and several drug classes, including anti-neovascularization agents, corticosteroids, and immunosuppressants. This finding is consistent with results from both clinical and basic research. For example, in clinical practice, long-acting corticosteroids (e.g., dexamethasone, fluocinolone acetonide) are administered via intravitreal implantation to treat various types of macular edema.^27,28 However, despite their potent anti-inflammatory properties, complex pharmacological mechanisms of corticosteroids may also induce or exacerbate fibrotic processes under certain pathological conditions.^29,30 From a translational perspective, we hypothesize that this may be linked to their modulation of key fibrotic signaling pathways. Although corticosteroids primarily target glucocorticoid receptors, their off-target effects can non-specifically activate mineralocorticoid receptors on retinal pigment epithelial (RPE) cells.^31,32 This activation subsequently upregulates critical fibrotic mediators, specifically connective tissue growth factor (CTGF) and transforming growth factor-beta (TGF-β), and triggers epithelial–mesenchymal transition (EMT), which in turn affects retinal tissue remodeling.^30,33 This hypothesis is substantiated by basic research demonstrating that triamcinolone can induce angiogenesis and tissue remodeling in human RPE cells via TGF-β2 regulation, suggesting that dysregulation of this process under certain pathological conditions could ultimately accelerate RF and scar formation.³⁴ Moreover, the cessation of long-term corticosteroid therapy may precipitate a “rebound effect,” in which the sudden withdrawal of inflammatory suppression leads to a compensatory surge in profibrotic cytokines, thereby accelerating fibrosis and scarring.³⁵

Furthermore, the immunosuppressant pegcetacoplan is the first drug globally approved for the treatment of geographic atrophy.³⁶ Although it treats geographic atrophy in dry AMD by inhibiting the complement protein C3, this intervention in the complement pathway may disrupt the homeostasis of the retinal microenvironment.³⁷ Previous studies have indicated that complement activation products C3a and C5a activate microglia, forming a positive feedback loop that exacerbates retinal damage.³⁸ Similarly, Xu's research team found that the complement fragment C3a can induce EMT in retinal pigment epithelial cells, a critical mechanism of RF.³⁹ Concurrently, clinical studies have revealed that patients treated with pegcetacoplan face a significantly increased risk of developing nAMD. This suggests that some patients initially diagnosed with dry AMD may experience a “conversion” or “concurrent development” of nAMD following pegcetacoplan therapy. nAMD is the primary ocular condition responsible for RF.^40,41

More importantly, in the treatment of neovascular retinopathy, intravitreal injections of anti-VEGF drugs, such as ranibizumab and aflibercept, have become the primary therapeutic approach for conditions such as nAMD.^42,43 Although these drugs demonstrate remarkable efficacy in suppressing choroidal neovascularization, a potential clinical paradox is emerging: Their core therapeutic mechanism may be associated with fibrotic changes.⁴⁴ Clinical trials have also demonstrated the existence of this risk.⁴⁵ This phenomenon reveals that the most effective anti-angiogenic therapy currently available may, with long-term use, potentially correlate with RF and scar formation. This mechanism is often referred to as the “angio-fibrotic switch.” This may occur because potent VEGF inhibition, although inducing neovascular regression, disrupts the delicate balance between angiogenesis and tissue repair. Under physiological conditions, VEGF and CTGF are dynamically co-regulated. However, rapid pharmacological inhibition of VEGF may result in aberrantly elevated CTGF levels. CTGF is a pivotal factor driving intraocular fibrosis, and the disruption of the equilibrium between CTGF and VEGF is associated with the angio-fibrotic switch.^46,47 This imbalance subsequently triggers excessive extracellular matrix deposition and activates the fibrotic potential of multiple retinal cell types. Examples include EMT in RPE cells, alongside glial proliferation, activation, and migration in Müller cells.^48,49 The phenotypic transformation of these cells collectively contributes to the formation of RF and scarring.

However, it is crucial to interpret these associations with caution. In clinical practice, distinguishing whether RF is a specific drug-related adverse event or simply the natural progression of the underlying neovascular disease (e.g., nAMD) is challenging. Generally, fibrosis is often a natural sequela of the disease itself rather than a direct result of pharmacotoxicity. The high ROR observed in our study may partly reflect “protopathic bias,” where aggressive disease states predisposed to fibrosis are more frequently treated with these agents. Therefore, our findings should be viewed as safety signals identifying a potential statistical association, which does not equate to definitive causality.

Notably, specific anti-VEGF drugs showed stronger statistical signals in our analysis. Clinical evidence, for example, indicates that brolucizumab can elicit a more potent inflammatory response.⁵⁰ Given that inflammation is a critical factor in the fibrotic pathological process,⁵¹ this heightened inflammatory stimulus could further accelerate or aggravate the progression of fibrotic lesions. Given the large-sample evidence from this study, patients receiving these therapies should undergo regular ophthalmic follow-up. Furthermore, patients should be advised to remain vigilant in order to promptly identify early warning signs of potential RF, such as metamorphopsia, a progressive decline in central vision, or the appearance of a central scotoma. These results should not discourage the necessary use of anti-VEGF therapy for indicated conditions but rather encourage vigilant monitoring.

Risk stratification analysis revealed that RF was strongly associated with verteporfin, ranibizumab, and brocilizumab, whereas the risk for latanoprost, vigabatrin, and sildenafil was relatively low. Of greater clinical significance is the difference in latency: high-risk anti-neovascularization agents (e.g., verteporfin) tend to induce fibrosis with a shorter latency, suggesting a need for close monitoring of ocular symptoms and regular assessments for users. In contrast, sildenafil, despite its lower risk classification, is associated with fibrotic lesions that present with a longer latency period. This long-latency signal for sildenafil must be interpreted in the context of its overall ocular safety profile. Numerous studies have confirmed that the primary visual side effects of sildenafil (e.g., chromatopsia, blurred vision) are typically transient and fully reversible, with most cases resolving within hours to days.^52,53 When used appropriately under medical guidance, the drug rarely causes permanent retinal damage. Therefore, although this study identified a signal for a long-latency association with fibrosis, given its low overall toxicity, only routine follow-up is required for patients on long-term therapy.

The subgroup analysis in this study further confirms that off-label drug use is a significant risk-associated factor for a significantly elevated risk of RF. Although off-label use is often associated with the severity and complexity of the disease and may affect the RF risk signal (i.e., confounding factors arising from the disease itself), we propose that it also introduces distinct pharmacological risks that extend beyond the disease process. Unlike on-label use, off-label use usually operates within a zone of pharmacological uncertainty, where non-standardized dosing regimens may deviate from the safety margins established in clinical trials. As large-scale pharmacovigilance studies have confirmed, off-label use is significantly associated with a higher incidence of adverse drug events.⁵⁴ In the context of anti-VEGF therapy, clinicians may employ intensified dosing or shortened administration intervals to control the disease, which may inadvertently expose ocular tissues to cumulative drug concentrations that exceed the physiological safety threshold of the RPE. This toxicological stress, stemming from non-standardized exposure, may trigger fibrotic pathways independent of the natural progression of the disease. Therefore, this finding not only corroborates that off-label use is associated with a higher risk of adverse events but, more importantly, reminds us that in addition to maintaining vigilance regarding disease progression, we must also attend to the potential toxicity of unoptimized therapeutic regimens. This result provides strong evidence for the “drug-related RF” hypothesis and holds significant guiding importance for clinicians' risk assessment and prescribing decisions. Therefore, when considering an off-label treatment regimen, clinicians must fully weigh the potential therapeutic benefits against the risk of inducing serious ocular adverse events.

To evaluate the potential impact of clinical heterogeneity, we conducted a detailed subgroup analysis based on specific MedDRA PTs. The results confirmed distinct heterogeneity in the associations between different drug categories and fibrosis subtypes. For example, verteporfin showed high specificity for subretinal fibrosis and the macular scar, whereas the complement inhibitor pegcetacoplan showed a specific signal exclusively for ERM. However, despite variations in signal strength across subtypes, all 16 candidate drugs identified in our primary analysis displayed positive signals in at least one fibrosis subtype, with over half showing cross-subtype positive trends. This consistency indicates a robust fundamental association between these agents and RF risk. As reviewed by Shu and Lovicu,⁵⁵ although ERM, proliferative vitreoretinopathy, and subretinal fibrosis present with distinct clinical phenotypes, they share core pathophysiological mechanisms of aberrant fibroblast activation and extracellular matrix deposition. Although clinical heterogeneity reflects differential pharmacological effects on specific fibrotic pathways, it does not invalidate the aggregate RF analysis. Our subgroup data, combined with the shared mechanistic basis, justifies grouping these subtypes under the broad RF category.⁵⁵

Therefore, our study suggests that, despite clinical variations, the aggregate fibrosis signal remains a reliable indicator of potential fibrotic risk, although future prospective studies are warranted to further delineate specific drug-subtype interactions. This study further reveals significant age- and sex-based heterogeneity in the risk of drug-related RF. This finding provides crucial data to support stratified risk management in clinical practice. In terms of age stratification, heightened vigilance is warranted when using anti-VEGF drugs or specific corticosteroids in adolescent patients, whereas for elderly patients, the scope of priority monitoring should be expanded to include immunomodulators (e.g., pegcetacoplan). Regarding sex stratification, females exhibited a higher RF risk with verteporfin, ranibizumab, and fluocinolone acetonide, whereas males were more susceptible to ocriplasmin and aflibercept. Although the precise mechanisms underlying these differences remain to be fully elucidated, their clinical guiding significance is already clear. Therefore, when making prescribing decisions, clinicians must incorporate individualized factors such as the patient's age and sex, along with their medical history, into their considerations. When necessary, treatment regimens should be actively optimized, such as by selecting alternative drugs with more favorable risk profiles or adjusting dosages, to minimize the risk of treatment-related serious ocular adverse events.

One of the strengths of this study is its approach to the intricate association between drugs and RF, a complexity that single-level analyses often fail to capture. To address this challenge, this study innovatively constructed a multidimensional, systematic analysis framework that integratively assesses the therapeutic purpose of drugs, risk levels, latency periods, and individual patient factors such as age and sex. The value of this integrative research is twofold. On the one hand, it successfully distills vast amounts of real-world data into actionable strategies for individualized medication use and early intervention that clinicians can apply directly. On the other hand, it effectively fills a long-standing gap in the real-world evidence for drug-related RF, greatly deepening our understanding of its risk factors and patterns of occurrence. The conclusions of this study are derived from a rigorous analysis of over 17 million reports, and the scientific validity and reliability of its results will undoubtedly provide a solid foundation for future explorations in the fields of pharmacovigilance and ocular safety.

Although this study aimed for comprehensiveness, it is subject to several inherent limitations. First, the FAERS database lacks a definitive denominator representing the total number of patients exposed to a specific drug. Consequently, we cannot determine absolute incidence rates and can only quantify the strength of reporting associations. Second, inherent reporting biases may arise from selective reporting or shifts in public awareness. Additionally, residual confounding from unrecorded variables, such as patient pathophysiology or concomitant drug use, cannot be fully excluded. Specifically, although we restricted our analysis to drugs designated as the “PS” to minimize interference from concomitant medications, the influence of polypharmacy cannot be completely ruled out. Third, the disproportionality analysis employed in this study is a tool for mining statistical associations, not for establishing causality. Because FAERS data are based on spontaneous reports, the mere existence of a report does not inherently confirm that the drug caused the adverse event. Therefore, to make the critical leap from association to causation, the findings from this real-world data analysis must be validated by future, rigorously designed prospective cohort studies or randomized controlled trials.

Conclusions

This study systematically assessed the risk of drug-related RF through in-depth mining of real-world data from the FAERS database. A total of 16 significantly associated drugs were identified, and a multidimensional characterization of their risk profiles was conducted, revealing heterogeneity among different drug classes. For example, anti-neovascularization agents were associated with a higher risk, whereas corticosteroids exhibited a longer latency. These findings highlight the necessity of individualized risk monitoring, particularly in high-risk scenarios such as the use of high-risk drugs, off-label applications, and administration to specific susceptible populations. Therefore, clinicians should develop dynamic monitoring plans based on the risk signals identified in this study and individual patient characteristics. For example, patients receiving long-term anti-VEGF treatment should undergo regular optical coherence tomography and screening for early fibrosis biomarkers, such as subretinal hyperreflective material. The aim is to facilitate early detection and timely intervention, ultimately preserving visual acuity and preventing irreversible structural damage.