Topic
To evaluate whether metformin use is associated with a reduced risk of developing glaucoma.
Clinical Relevance
Glaucoma is a leading cause of irreversible blindness worldwide. Identifying systemic medications that may modify glaucoma risk could have important implications for prevention strategies in patients with diabetes, a population frequently treated with metformin.
Methods
This systematic review and meta-analysis of observational cohort studies, registered on PROSPERO (CRD420250655975), was done through PubMed, Scopus, Web of Science, and Google Scholar until June 16, 2025. Eligible studies compared glaucoma incidence among metformin users versus nonusers or users of other antidiabetic drugs (ADDs). Risk of bias was assessed using the Newcastle–Ottawa Scale. Binary outcomes were pooled using random-effects models to calculate odds ratios (ORs), and time-to-event outcomes were synthesized using hazard ratios (HRs). Subgroup analyses explored confounder adjustment methods and comparator types. Certainty of evidence was graded using GRADE framework.
Results
Twelve retrospective cohort studies ( n = 1,247,325; 732,423 metformin users; 513,292 controls) were included. The pooled crude OR showed no association between metformin use and glaucoma risk (OR = 0.96; 95% CI, 0.87-1.06; I² = 79.14%; low certainty). A leave-one-out sensitivity analysis excluding a study with a nonmetformin active comparator resulted in a modest but significant reduction in risk (OR = 0.92; 95% CI, 0.87-0.98; low to moderate certainty). No effect modification was detected by the confounder adjustment method (propensity-score matched vs regression; P =.20; low certainty) or comparator type (no metformin vs other ADDs; P =.34; very low certainty). The follow-up duration did not significantly modify the effects. Four studies contributed time-to-event analyses: pooled unadjusted HR indicated a lower risk among metformin users (HR = 0.86; 95% CI, 0.79-0.93; I² = 0%; moderate certainty), which persisted in adjusted models (aHR = 0.88; 95% CI, 0.80-0.96; I² = 0.01%; moderate certainty). There was no evidence of small-study effects (Egger’s P =.955).
Conclusion
Across observational cohorts, metformin use was not associated with a reduced glaucoma risk in crude analyses; however, sensitivity analyses suggested a possible protective effect. Time-to-event analyses consistently demonstrated a modest reduction in glaucoma risk, supported by moderate-certainty evidence after adjusting for confounders. Overall, certainty of evidence ranged from very low (comparator analyses) to moderate (time-to-event analyses).
INTRODUCTION
G laucoma is the leading cause of irreversible blindness worldwide, affecting an estimated 76 million individuals in 2020. It accounts for approximately 8.4% of global blindness and 1.4% of moderate to severe visual impairment (MSVI). ,,,, Primary open-angle glaucoma (POAG), the most common subtype, comprises more than two-thirds of cases. While elevated intraocular pressure (IOP) is the only modifiable risk factor, disease progression occurs in up to 45% patients despite IOP reduction, and approximately one-third of patients develop glaucoma with IOP within the normal physiological range, indicating that non-IOP-related mechanisms also contribute to disease pathogenesis.
As global life expectancy increases, so does the burden of polypharmacy, particularly among individuals with chronic metabolic conditions such as diabetes mellitus. This trend has increased interest in the ocular effects of systemic medications and their potential to modulate glaucoma risk. ,, Metformin, a first-line treatment for type 2 diabetes, has attracted attention due to its activation of longevity-associated pathways (eg, AMP-activated protein kinase—AMPK signaling) and its potential neuroprotective properties independent of glycemic control. ,, Several observational studies have reported an association between metformin use and reduced glaucoma risk, , with some suggesting dose-dependent benefits.
However, findings across studies have been inconsistent. While certain cohorts have observed lower glaucoma risk with higher cumulative metformin exposure, , others, including a meta-analysis by Kim et al, found no significant association. , Notably, that meta-analysis included only five studies and did not distinguish between crude and adjusted estimates, nor did it incorporate detailed subgroup or sensitivity analyses.
To address these gaps, we conducted a systematic review and meta-analysis to comprehensively evaluate the association between metformin use and glaucoma risk. Our objectives were to assess both crude and adjusted effect estimates, investigate methodological sources of heterogeneity, and evaluate potential effect modifiers, thereby providing a more robust understanding of metformin’s potential role in glaucoma risk modulation.
MATERIALS AND METHODS
PROTOCOL REGISTRATION
This systematic review and meta-analysis were conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. The review protocol was registered on PROSPERO (registration number: CRD420250655975).
LITERATURE SEARCH
On February 16, 2025, we systematically searched four electronic databases: PubMed, Scopus, Web of Science (WOS), and Google Scholar. Given the very large and nonreproducibly ranked result sets returned by Google Scholar, and its use as a supplementary rather than primary database in systematic reviews, we screened the first 200 records as a predefined stopping rule. This approach is consistent with published methodological guidance indicating that screening the first 200 to 300 Google Scholar results is generally sufficient and feasible for evidence synthesis. ,, The search terms were formulated using the PICO framework: participants were adult patients with or without diabetes, the exposure was metformin use, the comparator included nonmetformin users or users of other antidiabetic drugs (ADDs), and the primary outcome was the risk of glaucoma. The detailed search criteria, along with the adjusted queries based on the indexing and syntax rules of each database, can be found in (E-supplement 1). The database search was updated on June 16 th, 2025, yielding one additional study.
Following the identification of finally eligible studies, a manual search was conducted to identify potentially missing relevant articles. This included: (1) screening the reference lists of all finally included articles, (2) reviewing the ‘similar articles’ feature of each included record via PubMed, and (3) conducting targeted keyword searches on Google using the terms (metformin + glaucoma).
ELIGIBILITY CRITERIA
We included any original study reporting comparative data on the incidence rate or risk of glaucoma between metformin users and a comparator group (nonmetformin users or users of other ADDs), either as raw outcome data or as relative effect estimates. Eligible studies included both observational designs (eg, cohort, case-control, cross-sectional, and case series with more than 20 cases) and interventional studies (randomized and nonrandomized trials). No restrictions were placed on population demographics or comparator groups, and studies reporting any type of glaucoma were considered, including open-angle glaucoma (OAG), angle-closure glaucoma (ACG), and low-tension glaucoma (LTG). The full-texts and supplementary files of potentially relevant studies (ie, clinicodemographic risk factors or medications associated with glaucoma) were also checked to ensure the adequacy of the screening process.
We excluded studies based on the following criteria: (1) studies that focused exclusively on glaucoma progression rather than risk, (2) lack of specific data on glaucoma risk, (3) studies without a comparator group or without extractable relative risk estimates, (4) animal or in vitro studies, (5) nonoriginal articles (eg, reviews, editorials, commentaries, letters to editors, and guidelines), (6) case reports or case series with fewer than 20 cases, (7) studies not published in English, (8) conference abstracts or posters without full-text availability, and (9) duplicate records or studies with overlapping data sets identified by similar study location, population, and sample size.
SCREENING AND STUDY SELECTION
The records retrieved from the primary database search were imported into Rayyan AI (Rayyan Systems Inc., Cambridge, MA, USA; https://www.rayyan.ai ), a web-based platform designed to facilitate systematic review screening through blinded, independent review and collaborative conflict resolution. Duplicate entries were identified and removed prior to screening. The remaining records were screened in two steps: (1) title and abstract screening conducted within Rayyan, followed by (2) full-text assessment for eligibility based on predefined inclusion and exclusion criteria. Two independent reviewers (ZR and KA) conducted the screening process. Any discrepancies or conflicts during the screening stages were resolved through discussion with a senior author (MA) until a final decision was reached.
DATA EXTRACTION
A data extraction sheet was developed following the review of data reported in the finally included studies (AA), and it was structured into two main parts. The first part captured baseline study and population characteristics, including: author name and year of publication, study title, journal, study design, country, year of investigation, population description, temporality, and participant demographics (sample size, mean age ± standard deviation (SD), and male proportion). Sex and gender were recorded as reported in the included studies. In most studies, the term ‘sex’ referred to biological classification (male/female), although terminology was not always explicitly defined. We retained the original terminology used by each study and did not reclassify these variables. Data related to metformin exposure were also extracted, including inclusion and exclusion criteria, glaucoma definition, total number of diabetic and nondiabetic participants, glaucoma and nonglaucoma cases, metformin users, nonusers, users of other ADDs, specific ADD names, line of therapy (first-line, second-line, etc.), therapy duration, dosage, and diabetes duration.
The second part focused on the outcomes of interest. These included glaucoma type, metformin dose, exposure time, type of ADDs, glaucoma rate or count, and measures of glaucoma risk (eg, odds ratio (OR), hazards ratio (HR), risk ratio (RR)), along with corresponding 95% confidence intervals (CIs) and whether they were adjusted (aOR, aHR). Where available, confounding variables controlled for in multivariable models were also recorded. Outcomes were stratified by comparison groups (eg, metformin vs no metformin, metformin vs other ADDs), and data were extracted as raw numbers and total sample sizes per group.
QUALITY ASSESSMENT
We assessed the methodological quality of the included cohort and case-control studies using the Newcastle-Ottawa Scale (NOS). The NOS evaluates observational study quality across three domains: Selection, Comparability, and Outcome (for cohort studies), using eight specific criteria. Each item was scored using a star system, with a maximum of one star per item, except for comparability, which allows up to two stars. Based on the total score, each study was rated as having good, fair, or poor quality. The assessment was performed by one reviewer (AA), and all evaluations were reviewed and verified by another author (MA). In case of disagreement, a third reviewer (AME) was involved.
STATISTICAL ANALYSIS
All meta-analyses were conducted using STATA version 18 (StataCorp LLC, College Station, TX, USA). The primary outcome was the pooled risk of glaucoma in patients receiving metformin compared to those in the control group. Binary outcome data were synthesized using random-effects models to account for anticipated between-study variability. Pooled ORs and HRs were calculated with corresponding 95% CIs. Heterogeneity was assessed using the I² statistic, with values above 75% indicating substantial heterogeneity. Statistical significance of heterogeneity was determined using Cochran’s Q test, with a P -value threshold of <.05.
For studies reporting cumulative incidence rates between examined groups, pooled ORs were calculated for the crude risk of glaucoma. Subgroup analyses were performed to assess the effect of potential moderators, including the method of confounder adjustment (propensity score matching vs regression modeling), type of control group (no metformin vs other ADDs), and duration of follow-up. Sensitivity analyses were conducted using the leave-one-out method to evaluate the robustness of the pooled estimates.
For time-to-event data, HRs were pooled separately for crude and adjusted estimates. A fixed-effects model was used for these analyses when heterogeneity was low (I² < 25%). Funnel plots and Egger’s test were planned for publication bias assessment, but were not performed due to the limited number of included studies in several subgroups ( n < 10). All statistical tests were two-tailed, and P -values <.05 were considered statistically significant unless otherwise specified.
CERTAINTY OF EVIDENCE ASSESSMENT
To assess the certainty of the evidence, we applied the Grading of Recommendations Assessment, Development and Evaluation (GRADE) framework. Evidence from observational studies was initially rated as low certainty and could be downgraded for risk of bias, inconsistency, indirectness, imprecision, or publication bias, or upgraded for factors such as large effect size, dose-response relationship, or evidence of residual confounding likely reducing the observed effect. For each reported outcome (crude risk estimates, sensitivity analyses, subgroup analyses by adjustment method and comparator, and time-to-event analyses), certainty of evidence was graded as very low, low, or moderate. A Summary of Findings table ( Table 1 ) presents the detailed GRADE ratings for each outcome.
TABLE 1
GRADE Assessment of Certainty of Evidence for the Association Between Metformin Use and Glaucoma Risk Across Pooled and Subgroup Analyses.
| Outcome | No. of studies (n Patients) | Effect Estimate | Certainty of Evidence | Rationale |
|---|---|---|---|---|
| Glaucoma incidence (crude OR) | 11 studies (1,245,715) | OR = 0.96 (95% CI: 0.87-1.06); I² = 79% | ↓︎ Low | Downgraded for inconsistency (substantial heterogeneity). No clear protective effect; confidence interval crosses 1. |
| Glaucoma incidence (sensitivity analysis excluding Allan et al.) | 10 studies | OR = 0.92 (95% CI: 0.87-0.98) | ↓︎ Low–Moderate | Upgraded for consistent significant effect after exclusion. Still observational. |
| Subgroup by confounder adjustment (PSM vs regression) | 11 studies | PSM: OR = 1.02 (95% CI: 0.85-1.23, I² = 80%); Regression: OR = 0.91 (95% CI: 0.82-1.01, I² = 62%) | ↓︎ Low | Downgraded for inconsistency and residual confounding. Findings differ by analytic method. |
| Subgroup by comparator (no metformin vs other ADDs) | 11 studies | No metformin: OR = 0.94 (95% CI: 0.85-1.03); Other ADDs: OR = 0.69 (95% CI: 0.37-1.29, I² = 97%) | ↓︎ Very Low | Downgraded for inconsistency (extreme heterogeneity) and imprecision (wide CI). |
| Time-to-event risk (unadjusted HR) | 4 studies (∼large cohorts) | HR = 0.86 (95% CI: 0.79-0.93; I² = 0%) | ↑︎ Moderate | Upgraded for consistency (no heterogeneity) and precision. Observational design prevents high certainty. |
| Time-to-event risk (adjusted HR) | 4 studies | aHR = 0.88 (95% CI: 0.80-0.96; I² = 0.01%) | ↑︎ Moderate | Upgraded for strong consistency, precise effect, and adjustment for confounders. Still limited by observational design. |
HR: crude hazards ratio; aHR: adjusted hazards ratio; OR: odds ratio; CI: confidence interval; ADD: antidiabetic drugs; PSM: propensity-score matching.
RESULTS
LITERATURE SEARCH RESULTS
A total of 2413 records were retrieved from four databases: 44 from PubMed, 2121 from Scopus, 48 from WOS, and 200 from Google Scholar. After removing 218 duplicate records, 2195 records remained for title and abstract screening. Following the screening process, 2147 articles were excluded for not meeting the eligibility criteria. The full texts of 48 potentially relevant articles were retrieved and assessed for eligibility. Of these, 39 articles were excluded for the following reasons: not randomized or observational studies ( n = 15), wrong outcomes ( n = 9), or irrelevant topics ( n = 15). Notably, the study of Mauyad et al, although reporting relevant data was excluded as it analyzed the same dataset and patient population reported in the supplementary files of the study of Allan et al. Nine studies were identified through the formal database search ,,,,,,,, and three studies were identified with the manual search. ,, A total of 12 studies were meta-analyzed ( Figure 1 ).
A PRISMA flow diagram showing the results of the search and screening processes.
BASELINE CHARACTERISTICS OF INCLUDED STUDIES
All included studies were retrospective cohort in design, with four studies being done in the United States (US), ,,, while remaining studies were done in Korea, India, Denmark, Finland, , Bosnia, Jordan, and Taiwan. Overall, 1,247,325 patients were included, of whom 732,423 patients were metformin users and 513,292 were in the control group. Regarding diabetes status, eight of the nine included studies enrolled only patients with diagnosed diabetes mellitus. ,,,,,,, However, four studies included mixed populations of diabetic and nondiabetic individuals. ,,, In the study by Allan et al, diabetes was present in 83.9% of participants in the glucagon-like peptide 1 receptor agonists (GLP-1 RA) group and 83.3% in the metformin group. In the study by Funk et al, diabetes was reported in 29.6% of patients with glaucoma and 12.3% of those without glaucoma.
Seven studies reported OAG risk, ,,,,,, four reported POAG risk, ,,, and one reported LTG risk. The control group was no metformin in eight studies ,,,,,,, and other ADDs in the remaining four studies. ,,, The method of confounding control was propensity score matching in four studies ,,, and the remaining seven studies used regression modeling. ,,,,,, One study did not use either approach. Data on patients’ age and gender were reported only in four studies ( Table 2 ). ,,, The follow-up period was reported in nine studies, ,,,,,,,, ranging from one year to 10 years. , Although age-based eligibility criteria were reported at the study level (E-supplement 2), only four studies provided age data stratified by exposure group ( Table 2 ), and none reported age-specific glaucoma outcome data, precluding age-restricted subgroup meta-analysis.
TABLE 2
Baseline Characteristics of Included Studies Reporting the Risk of Glaucoma in Metformin Users.
| Author | Design | Country | YOP | Sample | Glaucoma type | Male (%) | Age (year) | DM (%) | FU (yr) | Comparison | Confounding Adjustment | ||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Total | Met | Ctrl | Met | Ctrl | Met | Control | Met | Control | |||||||||
| Kim et al | RC | Korea | 2024 | 20 646 | 10 323 | 10 323 | OAG | 54 | 54.3 | <50 (26.4%); 50-59 (33.9%); 60-69 (26.3%); ≥70 (13.4%) | <50 (26.5%); 50-59 (33.9%); 60-69 (26.1%); ≥70 (13.4%) | 100 | 100 | 10 | No Metformin | PSM |
Socioeconomic parameters, including age group (< 50, 50-59, 60-69, and ≥ 70 years), sex,
residential area (urban, metropolitan cities, and cities, rural, all other areas), and updated Charlson comorbidity index (CCI; score < 1, 1-2, and ≥ 3). |
| Lin et al | RC | USA | 2015 | 150 016 | 60 214 | 125 674 | OAG | – | – | – | – | 100 | 100 | 2 | Other ADDs | Regression | Age, sex, race, education, geographic region of residence, net worth, ocular comorbidities, other comorbidities, user of other diabetes medications |
| George et al | RC | India | 2021 | 4302 | 142 | 4160 | POAG | – | – | – | – | 100 | 100 (for ADD) | 6 | No Metformin | Regression | Age, gender, residence, IOP, and metformin use |
| Vergroesen et al | RC | Denmark | 2022 | 11 260 | 654 | 10 606 | OAG | – | – | – | – | 100 | 100 | 5 | Other ADDs | Regression | Age, sex, body mass index, treatment with statins, and treatment with antihypertensive medications |
| Virtanen et al | RC | Finland | 2022 | 244 100 | 69 837 | 174 263 | OAG | – | – | – | – | 100 | 100 | 10 | No Metformin | Regression | Diabetes, sex, age, socioeconomic status, medication, comorbidities, and hospital district |
| Allan et al | RC | USA | 2024 | 18 738 | 8943 | 8816 | POAG | 40.3 | 40.4 | 60.9 (7.5) | 61.2 (7.1) | 83.3 | 83.9 (for GLP-1 RA) | 5 | Other ADDs | PSM |
Age, sex, race (Black, Asian, and White), ethnicity (Hispanic), dyslipidemia, hypertension, diabetes, diabetic retinopathy and macular edema, overweight
or obese status, smoking status, BMI, and hemoglobin A1c. |
| Maleškić et al | RC | Bosnia | 2017 | 234 | 190 | 44 | OAG | 45.3 | 56.8 | 65.6 (10.5) | 72.1 (12.4) | 100 | 100 | 1 | Other ADDs | Regression | Analysis adjusted for gender, age, duration of T2D, serum concentration of cholesterol, smoking, BMI and presence of other diseases; Predictor: Metformin use; other therapy was used as a reference category |
| Funk et al | RC | USA | 2022 | 277 | 44 | 510 | LTG | 20.93 (entire population) | NR | No Metformin | Regression | Systemic hypertension | |||||
| Huang et al | RC | Taiwan | 2002- 2013 | 723 223 | 571 665 | 151 558 | OAG | 54.44 | 54.59 | 54.98 (12.15) | 57.81 (12.11) | 100 | 100 | 5 | No Metformin | Regression | Gender, age, income level, urbanization, DCSI score, and comorbidities |
| El-Zayyat et al | RC | Jordan | – | 62 | 41 | 21 | POAG | – | – | – | – | 100 | 100 | – | No Metformin | – | – |
| Loukovaara et al | RC | Finland | 2001- 2010 | 37 687 | 10 370 | 27 317 | OAG (XG, NTG, PG, CPOAG) | – | – | – | – | 48.20 (entire population) | 7 to 17 | No Metformin | PSM | Age, sex, start of follow-up year, and hospital district | |
| Zheng et al | RC | USA | 2007- 2014 | 36 780 | – | – | POAG | – | – | – | – | – | – | – | No Metformin | PSM | Age, gender, and region of residence |
Stay updated, free articles. Join our Telegram channel
Full access? Get Clinical Tree
