Metformin and AMD: Does a 1.6-Million-Patient Cohort Support Repurposing?

A US TriNetX retrospective electronic-health-record analysis summarized by Jindal and colleagues (2025) evaluated associations between metformin prescribing and age-related macular degeneration (AMD) in more than 1.6 million adults aged 65 years or older receiving ophthalmic care. After propensity-score matching, the overall risk ratio for incident AMD was 0.90 (95% CI 0.86–0.94), a modest observational association that is statistically below 1.0 but does not establish a clinically meaningful protective drug effect. Among patients with early or intermediate AMD, progression estimates were imprecise for geographic atrophy (RR 0.87, 95% CI 0.76–1.01) and close to null for neovascular AMD (RR 1.03, 95% CI 0.91–1.17). Prescription-based exposure, residual confounding, and diagnostic-code outcomes limit causal inference. Current evidence does not justify initiating metformin specifically to prevent or treat AMD.

This was an observational, retrospective cohort study using the TriNetX network of 70 US healthcare organizations, with records spanning January 2013 through June 2025 as described in the supplied review. The source document reports 1,635,007 older patients receiving ophthalmic services. The incident-AMD comparison included 189,725 propensity-matched patients in each exposure arm; the progression comparison included 9,244 matched patients per arm with baseline early or intermediate nonexudative AMD. These are distinct analytic cohorts and should not be conflated with a randomized treatment sample.

Propensity-score matching balanced recorded demographics, cardiometabolic disease, diabetes and diabetic retinopathy, glaucoma, retinal vascular occlusion, smoking codes, and baseline HbA1c strata (standardized mean difference below 0.10). An age-related cataract subgroup was used as a proxy for ophthalmic surveillance; a head-trauma negative-control cohort was also analyzed. Such methods reduce measured imbalance but cannot eliminate confounding from unmeasured health behavior, treatment indication, medication adherence, or differential diagnostic ascertainment.

The source reports overall incident AMD RR 0.90 (95% CI 0.86–0.94), five-year RR 0.94 (0.90–0.99), and ten-year RR 0.91 (0.87–0.94). In the cataract subgroup the overall RR was 0.93 (0.88–0.98). Matched onset event counts were 4,273 metformin-exposed versus 4,741 unexposed. These estimates suggest a small association in the recorded data, but the relative differences are sensitive to exposure definitions and surveillance; they do not demonstrate causality. Importantly, a confidence interval below 1.0 is conventionally statistically significant; the source document’s 0.90–1.10 clinical-relevance band should not be described as a formal proof of equivalence.

Among patients with baseline early/intermediate AMD, the overall GA progression estimate was RR 0.87 (95% CI 0.76–1.01), with 331 versus 379 events in matched exposed and unexposed arms. For nonfoveal GA, RR was 0.80 (0.67–0.97), based on 189 versus 235 events, but this exploratory signal attenuated in the cataract-restricted comparison (RR 0.83, 0.64–1.06). Foveal GA progression was RR 0.96 (0.80–1.14). The isolated nonfoveal result warrants investigation, not a clinical claim of protection; subgroup multiplicity and surveillance differences are relevant.

The overall estimate for conversion to neovascular AMD was RR 1.03 (95% CI 0.91–1.17), with 452 metformin-exposed versus 439 unexposed events. These results provide no convincing evidence that prescribing metformin reduces neovascular conversion in the studied population.

After excluding patients with metformin prescriptions before the index period, the reported incident-AMD estimate was RR 0.83 (0.80–0.87); in a cataract-defined ophthalmic-care subgroup it was RR 0.85 (0.80–0.89). The head-trauma negative control yielded RR 1.04 (0.93–1.16). These analyses provide useful context but cannot rule out residual bias or establish a causal preventive effect.

Experimental literature cited by the supplied review implicates AMPK-related metabolic regulation, reduced oxidative stress and inflammatory signaling, altered lipid handling, autophagy, and antiangiogenic pathways. These mechanistic hypotheses are biologically interesting but are not equivalent to demonstrated preservation of human visual acuity, OCT biomarkers, or geographic-atrophy area. Claims that metformin clears drusen in patients should not be inferred from preclinical observations.

For patients already taking metformin for established metabolic indications, this AMD analysis alone is not a reason to stop an otherwise appropriate medication. Conversely, retina clinicians should not prescribe metformin off-label solely for AMD prevention or slowing GA/neovascular conversion. Established AMD assessment, surveillance, risk modification, and indicated ocular treatments remain the clinical foundation. Any medication decision should account for systemic indications and contraindications.

The analysis is large and uses matching and sensitivity cohorts, but prescriptions do not confirm ingestion, cumulative exposure, or duration; outcomes rely on EHR diagnostic codes rather than centrally graded OCT or fundus images. Concurrent diabetes medications and longitudinal metabolic control may remain incompletely characterized. Matching does not correct all time-varying confounding, and a small relative association may coexist with detection bias. The supplied document also discusses target-trial emulation as a future design strategy; such methods can address specified biases when implemented well but do not automatically eliminate unmeasured confounding.

More informative work would prospectively define incident versus prevalent metformin use, adherence, dose, follow-up, and competing treatments; verify AMD stage by imaging; measure visual and anatomical endpoints; and emulate a prespecified target trial or conduct a randomized study where justified. Independent replication of the nonfoveal GA subgroup result is especially important.