Association of metformin use with primary open-angle glaucoma using data from the National Institutes of Health All of Us Program

Abstract

Background

Few studies have assessed the impact of metformin use on glaucoma risk. The purpose of this study was to examine the association between metformin use and the incidence of primary open-angle glaucoma (POAG) in a diverse and large nationwide cohort.

Methods

We included a retrospective cohort study of 18 440 participants in the National Institutes of Health All of Us Research Program aged 40 years or older, with a diagnosis of diabetes mellitus and without a diagnosis of POAG prior to diabetes diagnosis or metformin use. Bivariate logistic regression, multivariable logistic regression and survival analysis were used to analyse the association between ever use of metformin and incidence of POAG.

Results

Within the cohort, 240 participants acquired a diagnosis of POAG during all available follow-up time, while 18 200 did not. In regression-based bivariate analysis, metformin use was significantly associated with a lower odds of developing POAG (OR 0.35, 95% CI 0.26 to 0.47, p<0.001). In multivariable regression analysis, metformin remained protective against POAG (OR 0.33, 95% CI 0.21 to 0.50, p<0.001), while the use of other diabetic medications was associated with an increased odds of developing POAG (OR 2.39, 95% CI 1.48 to 3.90, p<0.001). In survival analysis, the probability of developing POAG was significantly lower for the participants using metformin than for the participants not using metformin (log-rank p<0.001, Cox proportional HR 0.38, 95% CI 0.29 to 0.51).

Conclusions

This study provides additional large-scale observational health data supporting the protective role of metformin in the development of POAG. However, limitations include the study’s observational design and lack of data on metformin dosage and duration, glaucoma severity and ocular exam findings. Despite these limitations, our findings contribute to the growing body of evidence suggesting a potential protective effect of metformin against POAG.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• Limited studies have assessed the relationship between metformin use and primary open-angle glaucoma (POAG) and they offer conflicting evidence.

WHAT THIS STUDY ADDS

• Our study contributes to addressing this gap with a nationwide cohort of patients with diabetes with a large sample size and inclusion of minority populations who are disproportionately impacted by glaucoma.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• Our findings offer evidence of the potential protective effect of metformin in mediating POAG risk and motivate future investigation on the topic.

Introduction

Glaucoma is a leading cause of irreversible blindness worldwide, and its global prevalence is expected to increase with the ageing population.^{1 2} Primary open-angle glaucoma (POAG) is the most common form, with an estimated global prevalence of 2.4%, or 68.6 million, among the population aged 40 years old and older.^{3 4} Intraocular pressure (IOP) is the only modifiable risk factor and lowering IOP is the current mainstay of therapy. However, POAG can occur with or without increased IOP, and lowering IOP may not decrease the risk of glaucoma in all patients.¹ Consequently, there is an emerging interest in the use of other therapeutic modalities to reduce the development of glaucoma or the rate of progression in patients with glaucoma.

Metformin is an antihyperglycaemic drug that exerts its effects by reducing hepatic glucose production.⁵ Although metformin is typically used for the management of type 2 diabetes mellitus, it is being increasingly explored for its health benefits outside of diabetes, including neuroprotection in the eye.⁶ Multiple experimental studies and two clinical studies have suggested that metformin plays a protective role in the development of glaucoma.6,8 A retrospective cohort study by Lin et al of patients aged 40 years or above showed that use of metformin resulted in reduced risk of developing open-angle glaucoma over a 2-year period and demonstrated a dose-dependent response.⁸ In contrast, other studies offer evidence that metformin use does not decrease glaucoma incidence or progression.9,11 In a study of patients with diabetes mellitus, there was no association found between metformin use and 6-year incidence of POAG in multivariate analysis.¹¹

Given the limited studies focusing on this topic and the conflicting evidence, there is a need for further research to help elucidate the potential role and benefits of metformin use in the prevention of glaucoma. Using the nationwide All of Us Research Program dataset,¹² this study investigated the association between metformin use and the incidence of POAG in a large and diverse population. The use of a database with a large sample size, nationwide scale and diverse population that accounts for demographic variability can help address the gap in detailed knowledge regarding the impact of metformin on glaucoma.

Materials and methods

Study population

The All of Us Research Program, launched by the National Institutes of Health in May 2018, represents a nationwide effort to recruit a diverse participant pool of one million or more people.¹² The programme focuses on creating a robust research database through the inclusion of historically excluded and under-represented populations.¹² The inclusion criteria are participants who are 18 years or older and currently reside in the USA or a territory of the USA.¹³ The exclusion criteria are individuals without the capacity to consent or those who are currently prisoners.¹³ Participants in the programme reflect diversity in race, ethnicity, age group, region of residency, gender identity, sexual orientation, socioeconomic status, education, disability and health status.¹³

After consenting, the individuals complete a baseline demographics survey, are evaluated for physical measurements and biospecimen data and can authorise sharing of their electronic health record (EHR) data.¹² Participants may also be invited to complete additional optional health surveys or participate in additional data collection.¹² All participant data are then deidentified through a rigorous protocol including removal of identifiers and date-shifting by a random number of days.¹³ Deidentified data can be accessed through the All of Us Researcher Workbench, a cloud-based data analysis platform available to researchers who have completed required training and have active data use agreements in place at their institution.¹³ The study adhered to the tenets of the Declaration of Helsinki.

Data processing

The present study was a retrospective cohort study of participants in the All of Us Research database. To evaluate the association between metformin use and risk of developing POAG, the study cohort consisted of participants in the All of Us Research Program aged 40 years or older with a diagnosis of diabetes mellitus. The baseline date was based on the date of first diagnosis code for diabetes mellitus and was, thus, different for each participant. Participants with a diagnosis of POAG prior to diabetes diagnosis or metformin use were excluded in order to assess newly incident glaucoma and to establish temporal validity between the exposure of interest (metformin use) and the outcome (development of POAG). The All of Us Cohort Builder was used to define the cohort according to the previously mentioned inclusion and exclusion criteria using the All of Us version 5 dataset.

The main exposure of interest (ever use of metformin) was based on EHR medication order data. Therefore, metformin use was treated as a binary variable, since duration/dates and dosage data were inconsistently available in the database. The outcome variable was diagnosis of POAG during all available follow-up time based on qualifying Systematised Nomenclature of Medicine (SNOMED) codes. These codes provide a standardised way to identify POAG cases across different EHR systems by linking different coding systems to a common standard.¹⁴ The study workflow and cohort definition are depicted in figure 1.

Figure 1. Study workflow and cohort definition using the All of Us Research Program. ICD-9-CM, International Classification of Diseases, Ninth Revision, Clinical Modification; POAG, primary open-angle glaucoma; SNOMED, Systematised Nomenclature of Medicine.

Based on prior literature,^{7 8} additional covariates for which we extracted data included age, gender, race, income, education, haemoglobin A1c (HbA1c), diagnosis of comorbid ocular diseases (exudative age-related macular degeneration (AMD), non-exudative AMD, cataracts or pseudophakia or aphakia, diabetic retinopathy), diagnoses of comorbid medical conditions (hyperlipidaemia, obesity, depression, hypertension), type of diabetes, cataract surgery status, and use of other common diabetes medication classes (sulfonylureas, alpha glucosidase inhibitors, thiazolidinediones, dipeptidyl peptidase 4 inhibitors, glucagon-like peptide-1 analogues, sodium-glucose co-transporter 2 inhibitors and insulin and analogues). Data for covariates were only included if they were present before the outcome (diagnosis of POAG) to establish temporal validity. Covariates were treated as binary variables and not as time-varying variables. Sets of qualifying codes were created for each covariate using queries of associated codes in the All of Us Researcher Workbench (online supplemental table S1).

Statistical analyses

Analyses were performed in R notebooks within the All of Us Workbench environment and are available in our workspace for registered All of Us users.¹⁵ Descriptive statistics were used to analyse the demographics and characteristics of the study cohort. Continuous variables were reported using means and SDs, while categorical variables were reported using frequencies and percentages. To compare the group that developed POAG and the group that did not develop POAG, t-tests were used for continuous variables (after ensuring that all assumptions of parametric hypothesis testing were met) and χ² analyses were used for categorical variables.

Spearman correlation coefficients were generated for each of the covariates to evaluate if any covariates were highly correlated. Missing values for categorical data were imputed with the mode, while missing values for continuous variables were imputed with the mean.

Bivariate and multivariable logistic regression were performed to analyse the association between metformin usage and incidence of POAG. The glm method in R was used for regression. ORs, 95% CIs and p values were generated and interpreted to determine whether metformin use was independently associated with decreased risk of POAG incidence. P values of less than 0.05 were considered statistically significant. For χ² and regression modelling, controls (ie, participants with diabetes who did not develop POAG) were randomly sampled using the sample function in R at a 4:1 control: case ratio to avoid class imbalance. Prior studies have additionally shown that ratios exceeding 4:1 do not lend additional statistical power for analyses.¹⁶

The cumulative incidence of POAG over time in the metformin use group and the no metformin use group among all participants with diabetes was analysed using Kaplan-Meier curves over 20 years of follow-up with log-rank comparisons and Cox regression modelling. For survival analysis, no participants were censored from the baseline cohort, and the survival, ggfortify and survminer packages, which censored participants over time as they developed POAG, were used in R.

Patient and public involvement

Patients or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.

Results

We identified a study population of 18 440 participants in the All of Us Research Program aged 40 years or older with a diagnosis of diabetes mellitus and without a diagnosis of POAG prior to diabetes diagnosis or metformin use. Within this cohort, 240 participants subsequently acquired a diagnosis of POAG during all available follow-up time, while 18 200 did not. Missing data for covariates (education, income and HbA1c) were present in 2.4% of POAG cases and 1.9% of controls (online supplemental figures S1 and S2). The characteristics of the study population are displayed by POAG diagnosis (table 1). Additionally, we presented the characteristics of the study sample used in χ² and regression modelling, which consisted of 240 cases and 940 controls (online supplemental table S2).

Table 1. Characteristics of adults in the All of Us Research Program diagnosed with diabetes without pre-existing primary open-angle glaucoma at the time of diabetes diagnosis or start of metformin use.

	Diagnosis of primary open-angle glaucoma (n=240)		No diagnosis of primary open-angle glaucoma (n=18 200)		P value
Current age (years)
Mean±SD	71.10±9.33		65.55±10.79		<0.001
Gender
Female	135	56.25%	10 941	60.12%	<0.001
Male	105	43.75%	7259	39.88%
Race*
White	82	34.17%	7944	43.65%	<0.001
Black or African American	98	40.83%	5345	29.37%
Other	60	25.00%	4911	26.98%
Ethnicity
Not Hispanic or Latino	194	80.83%	13 928	76.53%	<0.001
Hispanic or Latino	46	19.17%	4272	23.47%
Annual income
<US$10 000–US$25 000	81	33.75%	6057	33.28%	<0.001
US$25 000–US$50 000	30	12.50%	2802	15.40%
US$50 000–US$100 000	41	17.08%	2828	15.54%
>US$100 000	28	11.67%	2021	11.10%
Not indicated	60	25.00%	4492	24.68%
Type of diabetes
Type 1	50	20.83%	2622	14.41%	0.001
Type 2	223	92.92%	17 421	95.72%	0.01
Use of metformin
No	152	63.33%	7015	38.54%	<0.001
Yes	88	36.67%	11 185	61.46%

P values of statistically significant variables were shown in bold.

^*Certain race categories were collapsed/aggregated for the reporting of descriptive analysis (due to the All of Us Data and Statistics Dissemination Policy preventing any display of data that can be calculated to <20) but were treated separately for regression analysis.

Among the 240 participants in our cohort who developed POAG, the mean (SD) age of participants at time of analysis was 71.1 (9.33) years. Female participants (n=135) comprised 56.3% of the cases, and black or African American participants (n=98) represented 40.8%. The majority of the participants who were diagnosed with POAG were not Hispanic or Latino (n=194, 80.8%). The most common annual income reported among participants with POAG was <US$10 000–US$25 000 (n=81, 33.8%). Dates of diabetes diagnosis among participants in the cohort ranged from 1996 to 2020. 20.83% of participants had type 1 diabetes (n=50) and 92.9% of participants had type 2 diabetes (n=223), with some participants having both diagnoses, termed ‘double diabetes’.¹⁷ Most of these participants did not use metformin (n=152, 63.3%). Specifically, 32.1% used metformin in addition to other diabetic medications (n=77), 4.6% used metformin but no other diabetic medications (n=11), 31.7% did not use metformin but used other diabetic medications (n=76), and 31.7% used neither metformin nor other diabetic medications (n=76).

Within the cohort of the 18 200 participants who did not develop POAG, the mean (SD) age was 65.6 (10.8) years. White participants (n=7944) represented 43.7% of this cohort, and 76.5% were not Hispanic or Latino (n=13 928). The majority of participants who did not develop POAG had type 2 diabetes (n=17 421, 95.7%) and were prescribed metformin (n=11 185, 61.5%). Specifically, 47.9% used metformin in addition to other diabetic medications (n=8724), 13.5% used metformin but no other diabetic medications (n=2461), 14.9% did not use metformin but used other diabetic medications (n=2714) and 23.6% used neither metformin nor other diabetic medications (n=4301).

Between the participants who developed POAG and the participants who did not, there were significant differences in the distribution of each of the predictors displayed in table 1: age (p<0.001), gender (p<0.001), ethnicity (p<0.001), annual income (p<0.001), type 1 diabetes (p=0.001), type 2 diabetes (p=0.01) and use of metformin (p<0.001). Notably, most participants who developed POAG had not used metformin, while most participants who did not develop POAG had used metformin.

Similarly, in regression-based bivariate analysis, demographic factors such as age, education level and annual income were significantly associated with increased odds of developing POAG (table 2). Additionally, comorbid ocular disease (exudative AMD, non-exudative AMD, cataracts/pseudophakia/aphakia and diabetic retinopathy) and comorbid medical conditions (hyperlipidaemia, obesity and hypertension) were associated with increased POAG incidence. Male gender, ‘other’ race, Hispanic or Latino ethnicity, higher HbA1c, type 2 diabetes and use of insulin analogues were associated with decreased odds of developing POAG. Notably, use of metformin (OR 0.35, 95% CI 0.26 to 0.47, p<0.001) was also associated with decreased POAG risk (table 2).

Table 2. Logistic regression analysis for odds of developing primary open-angle glaucoma among patients with diabetes mellitus aged 40 or above (n=1200).

	Bivariate analysis		Multivariable analysis
	Crude OR (95% CI)	P value	Adjusted OR (95% CI)	P value
Current age (years)	1.07 (1.05 to 1.08)	<0.001	1.03 (1.01 to 1.05)	0.01
Gender
Female	Reference		Reference
Male	0.11 (0.08 to 0.15)	<0.001	0.27 (0.17 to 0.41)	<0.001
Race
White	Reference		Reference
Black or African American	1.06 (0.69 to 1.62)	0.80	1.61 (0.93 to 2.80)	0.093
Asian	1.24 (0.41 to 3.91)	0.707	3.48 (0.91 to 14.17)	0.071
Other	0.06 (0.04 to 0.09)	<0.001	0.21 (0.09 to 0.48)	<0.001
Ethnicity
Not Hispanic or Latino	Reference		Reference
Hispanic or Latino	0.08 (0.06 to 0.11)	<0.001	0.92 (0.42 to 2.20)	0.843
Highest education earned	1.45 (1.27 to 1.64)	<0.001	1.02 (0.84 to 1.23)	0.884
Annual income	1.45 (1.27 to 1.64)	<0.001	1.19 (0.97 to 1.46)	0.10
Use of metformin
No	Reference		Reference
Yes	0.35 (0.26 to 0.47)	<0.001	0.33 (0.21 to 0.50)	<0.001
Haemoglobin A1c	0.64 (0.57 to 0.72)	<0.001	0.75 (0.64 to 0.88)	<0.001
Comorbid ocular diseases
Exudative AMD	5.41 (1.18 to 27.60)	0.028	1.32 (0.20 to 9.31)	0.775
Non-exudative AMD	3.54 (1.77 to 6.99)	<0.001	1.69 (0.66 to 4.30)	0.273
Cataracts/pseudophakia/aphakia	4.00 (2.98 to 5.38)	<0.001	2.55 (1.67 to 3.91)	<0.001
Diabetic retinopathy	1.92 (1.29 to 2.82)	0.001	2.36 (1.32 to 4.19)	0.004
Comorbid medical conditions
Hyperlipidaemia	1.58 (1.15 to 2.21)	0.006	0.90 (0.56 to 1.46)	0.677
Obesity	1.97 (1.08 to 3.46)	0.022	1.25 (0.57 to 2.67)	0.571
Depression	1.21 (0.84 to 1.73)	0.299	0.76 (0.45 to 1.26)	0.297
Hypertension	1.82 (1.25 to 2.70)	0.002	1.04 (0.61 to 1.81)	0.884
Type 1 diabetes
No	Reference		Reference
Yes	1.38 (0.96 to 1.96)	0.078	1.24 (0.74 to 2.07)	0.404
Type 2 diabetes
No	Reference		Reference
Yes	0.07 (0.02 to 0.18)	<0.001	0.25 (0.06 to 0.88)	0.041
Cataract surgery
No	Reference		Reference
Yes	1.77 (0.68 to 4.20)	0.212	0.95 (0.27 to 3.07)	0.940
Use of insulin analogues
No	Reference		Reference
Yes	0.68 (0.51 to 0.90)	0.007	0.96 (0.61 to 1.51)	0.847
Use of other diabetes medications (sulfonylureas, alpha glucosidase inhibitors, thiazolidinediones, dipeptidyl peptidase 4 inhibitors, glucagon-like peptide-1 analogues and sodium-glucose co-transporter 2 inhibitors)
No	Reference		Reference
Yes	1.13 (0.84 to 1.52)	0.423	2.39 (1.48 to 3.90)	<0.001

P values of statistically significant variables were shown in bold.

AMD, age-related macular degeneration.

All covariates had correlation coefficients of below 0.9 and were thus retained for multivariable modelling. In multivariable analysis, age, cataracts/pseudophakia/aphakia and diabetic retinopathy remained significantly associated with increased odds of developing POAG. Additionally, use of other diabetes medications was associated with increased POAG incidence. Male gender, ‘other’ race, increased HbA1c and type 2 diabetes remained associated with decreased risk of POAG. Even after adjustment for demographic and medical covariates, use of metformin (OR 0.33, 95% CI 0.21 to 0.50, p<0.001) remained associated with decreased odds of developing POAG.

The Kaplan-Meier curves comparing metformin users versus non-users with diabetes across the entire cohort are displayed in figure 2. Consistent with the results of the regression models, the probability of developing POAG was significantly lower for the participants using metformin than for the participants not using metformin throughout 20 years of follow-up (log-rank p<0.001, Cox proportional HR 0.38, 95% CI 0.29 to 0.51).

Figure 2. Probability of primary open-angle glaucoma (POAG) diagnosis in metformin users and non-metformin users.^a n < 1,200 due to censoring and exclusion of patients with unspecified diabetes start date.

Discussion

The present study examined the association between metformin and the incidence of POAG using the All of Us database, notable for its inclusion of underrepresented populations. Key findings are that metformin use was independently associated with a decreased risk of developing POAG, even when controlling for sociodemographic factors and other ocular and medical risk factors. Moreover, time to diagnosis of POAG was delayed with metformin use. This study further validates the findings of Lin et al, offering evidence from an additional data source with increased diversity that metformin has a protective effect on the development of POAG.

Metformin has recently garnered a great deal of attention for its potential antiageing and anticancer effects and is associated with reduced incidence of diabetes, cardiovascular disease, cancer and neurodegenerative disease.¹⁸ Metformin has also been reported to be significantly associated with lower incidence of ocular conditions such as diabetic retinopathy.¹⁹ Ongoing clinical trials are investigating the influence of metformin on POAG progression through measuring changes in visual field loss, retinal nerve fibre layer thickness, visual acuity and cup/disc ratio.^{20 21} Given that IOP is the only proven modifiable risk factor associated with glaucoma and that reducing IOP does not decrease risk of POAG in all patients, there is a need for additional approaches to treatment. Understanding how antidiabetic drugs are associated with POAG may help uncover new therapeutic approaches for POAG.

Our analysis reinforces the findings of a recent study conducted by Lin et al, which used data from a nationwide healthcare claims database and showed that a cumulative dose of metformin above 1000 g resulted in a 25% reduction in the risk of developing open-angle glaucoma over a 2-year period.⁸ Furthermore, the study demonstrated a dose-dependent response, with every 1 g increase in cumulative metformin dosage associated with a 0.16% decrease in open-angle glaucoma risk.⁸ Our study is also consistent with another retrospective study of patients with type 2 diabetes, which showed that patients who used metformin had fewer ocular complications compared with patients who used other oral antihyperglycaemic agents (p<0.0001) and that metformin use was associated with a decreased odds of developing glaucoma (p=0.006).⁷ A key strength of our study was the length of follow-up data available, allowing survival analyses over a 20-year period (figure 2), which, to our knowledge, is the longest follow-up period in studies of metformin and glaucoma to date.

Our results, however, differ from other studies that offer evidence that metformin use does not decrease glaucoma incidence or progression. In a prospective cohort study of 905 patients with diabetes mellitus, there was no association found between metformin use and 6-year incidence of POAG after controlling for age, gender and region of residence.¹¹ Another study indicated that metformin did not reduce retinal nerve fibre layer thinning rates in 143 patients with POAG.⁹ Both studies, however, had substantially smaller sample sizes compared with our present study (n=18 440).

The racial and ethnic diversity of our study population offers new and deeper insight into the relationship between metformin and POAG. In the study conducted by Lin et al, 7.9% of the overall cohort was of African ancestry and 6.3% was of Latino ancestry.⁸ In contrast, 29.5% of the participants in our overall cohort identified as Black or African American, and 23.4% identified as Hispanic or Latino. The diversity of our study population, which is more reflective of the general US population than previous studies on the topic, contributes to the generalisability of our findings. African American individuals have the highest prevalence of POAG, and POAG progresses more rapidly and is more likely to result in subsequent visual impairment in this population.^{22 23} Similarly, POAG disproportionately impacts people of Hispanic or Latino ancestry. Individuals of Hispanic heritage have the highest age-dependent increase in POAG prevalence compared with other ethnic groups and by 2050, Hispanic POAG patients are anticipated to comprise 50% of all POAG patients.^{23 24} Because people of African American and Latino ancestry carry a disproportionate burden of glaucoma and glaucoma-related visual impairment, the broader representation reflected in this study uniquely adds to the literature on metformin and glaucoma.

In both our bivariate and multivariable analyses, male gender was associated with decreased odds of developing POAG. This contrasts with findings from other studies, including a recent meta-analysis and systematic review of studies published in the past 20 years, which reported that men were more susceptible to POAG than women.⁴ Proposed explanations include men having longer axial length and deeper anterior chamber depth.⁴ Additionally, female sex hormones, such as oestrogen, have been suggested to play a neuroprotective role in the eye.^{25 26} These contrasting findings suggest that the relationship between gender and POAG risk may vary by population, and our study’s racial and ethnic diversity may partially explain this discrepancy. Further research is needed to explore how these factors intersect to influence POAG susceptibility.

The biological mechanism driving the protective effect of metformin on POAG development remains unclear. Our analysis suggests that metformin may reduce the risk of developing POAG beyond glycaemic control. While Lin et al found that each unit of increase in HbA1c was associated with an 8% increase in POAG risk,⁸ our analysis revealed that higher HbA1c was associated with decreased POAG risk after adjustment for covariates. Some studies on this topic have found significant correlations between elevated HbA1c levels and factors associated with POAG such as high IOP, optic disc area, cup area and cup volume.^{27 28} Others, however, offer evidence of no association between HbA1c and POAG.^{29 30} One possible explanation for our findings is that better HbA1c control is linked with more frequent encounters with primary care providers.³¹ As a result, patients with lower HbA1c may attend medical visits more regularly and, therefore, may have more opportunities for referral for eyecare and POAG detection. For participants who developed POAG, another possible explanation is that only HbA1c values before their diagnosis were considered in order to preserve temporal validity of the analysis. Consequently, their HbA1c values do not reflect variations in HbA1c after diagnosis and may have been overestimated if these values decreased after POAG diagnosis. The conflicting evidence regarding the association between HbA1c and POAG risk warrants further research to better elucidate this relationship.

In the present study, metformin remained associated with decreased risk of POAG when adjusted for factors such as HbA1c, while the use of other diabetic medications was associated with an increased risk of POAG in the multivariable model. A substantially larger portion of both the case and control cohorts were taking metformin in addition to other diabetic medications (32.1% and 47.9%, respectively) than were taking metformin alone (4.6% and 13.5%, respectively). Given that the majority of metformin users in our study also used other diabetic medications, the protective association of metformin with POAG development was likely not due to avoidance of other diabetic medications associated with an increased risk of POAG. These findings provide further evidence that the relationship between metformin and POAG may not be mediated by glycaemic control. One experimental study has proposed that metformin prevents fibrosis through entering human conjunctival fibroblasts and activating the AMPK/Nrf2 signalling pathway.³² Others have suggested that metformin inhibits inflammation and mitochondrial dysfunction of retinal ganglion cell axons,⁶ reduces the production of reactive oxygen species and protects against apoptosis, all of which are involved in glaucoma pathogenesis.³³ Regardless of the underlying biological mechanism, this study provides additional evidence of the potential protective effects of metformin on POAG development and suggests that the use of other antihyperglycaemic agents may potentially be associated with greater POAG risk.

Although metformin has historically been a first-line therapy for diabetes since the 1950s, the paradigm of type 2 diabetes treatment has shifted in recent years.³⁴ Based on emerging evidence regarding the cardiovascular benefits of sodium-glucose cotransporter 2 (SGLT2) inhibitors and glucagon-like peptide 1 receptor agonists (GLP-1 RAs), the American Diabetes Association began recommending the use of these classes of antihyperglycaemic medications for patients with long-standing, suboptimally controlled type 2 diabetes and atherosclerotic cardiovascular disease in 2017.^{35 36} As of 2022, the Standards of Medical Care in Diabetes recommend first-line therapy depending on patients’ comorbidities.³⁷ In patients with contraindications or intolerance to metformin, first-line therapy for type 2 diabetes may include an antidiabetic drug from a different class.³⁷ For instance, in patients with type 2 diabetes who have or are at high risk for atherosclerotic cardiovascular disease, heart failure or chronic kidney disease, SGLT2 inhibitors and GLP-1 RAs are recommended with or without metformin.³⁷ Additionally, comprehensive lifestyle modification, including exercise and diet, represents a crucial component of first-line therapy for type 2 diabetes. In patients with type 2 diabetes who are overweight or obese, weight loss has been associated with improved glycaemic control and reduced need for antihyperglycaemic medications.³⁸ In patients with clear and modifiable factors contributing to hyperglycaemia, a 3-month period of only lifestyle modification, without the initiation of pharmacological therapy, may be considered.³⁹ As the standard of care for diabetes medications evolves, it is critical that we understand how different antihyperglycaemic agents impact the risk of concomitant chronic diseases such as POAG.

A limitation of our study is that dosage and duration data for metformin use were not consistently available and therefore not considered in our analysis. Because we did not have comprehensive information regarding dates of metformin use, we were unable to determine if covariates were concurrent with metformin use. Additionally, data on glaucoma severity were limited in the database, so we were unable to assess the relationship between metformin use and disease severity. Moreover, ocular exam data are currently not well-phenotyped in the All of Us database, so the potential effects of metformin on IOP, ocular structure and function were not determined. However, there are ongoing efforts to improve ophthalmic data coverage in All of Us, and this could provide additional opportunities in the future to better understand the biological mechanisms underlying the effect of metformin on POAG. Additionally, although diabetic retinopathy was controlled for in multivariable analysis, it was characterised by diagnosis codes rather than directly through notes or images in which staging could be directly confirmed/validated. Similarly, the diagnosis of POAG was based on SNOMED codes, which may not reflect the exact timing of clinical diagnosis, as coding can occur after disease onset. Finally, given the observational design of the present study, only associative, not causative, relationships can be established. To establish temporal validity, we placed temporal constraints including limiting the cohort to those with a diagnosis of diabetes without a preceding or concurrent diagnosis of POAG and limiting covariates for modelling to only data available before the diagnosis of POAG.

Overall, our study provides additional large-scale observational health data supporting the potential protective effects of metformin in mediating POAG risk and motivates future investigation. Additional research, namely prospective studies and clinical trials with non-diabetic populations, detailed information regarding metformin dosage and duration, and visual and glaucoma-related end points, is warranted to validate these findings. Given the limited prevention and treatment strategies for glaucoma, validation of the protective effects of metformin on POAG development and progression would represent a major advancement in uncovering new therapeutic modalities for POAG.

Data availability statement

Data are available on reasonable request.