Metformin can reduce seizure-related hospital admissions, falls and coded seizures in people with epilepsy who had no evidence of diabetes, a large international cohort study has found.
The findings, published in the Journal of Neurology, Neurosurgery & Psychiatry [link here] stopped short of showing the drug prevented epilepsy from developing after brain injury, but the signal for symptom control was striking enough for the authors to call for formal repurposing trials.
The study, led by Dr Gashirai Mbizvo from the University of Liverpool, UK, drew on real-world electronic health data from more than 250 million patients across 130-plus healthcare organisations in the TriNetX global network. A total of 16,695 participants were included across two cohorts.
In the epilepsy cohort (847 matched pairs), metformin add-on therapy was associated with a 22% lower hazard of coded seizures, a 30% lower hazard of emergency department attendances or hospital admissions, and a 32% lower hazard of falls, injuries or burns over 10 years, compared with licensed weight-management comparators including GLP-1RAs, orlistat and bariatric surgery.
The composite “net potentially seizure-related morbidity” outcome favoured metformin with a hazard ratio of 0.729 (95% CI 0.626 to 0.848), corresponding to an approximate number needed to treat of six over 10 years.
Mortality did not differ between groups (HR 1.039, 95% CI 0.594 to 1.816), which the authors said supported comparable background health status between the matched cohorts.
Critically, the study restricted inclusion to people without any evidence of diabetes, excluding participants based on ICD codes, medication records, HbA1c of 5.7% or higher, or blood glucose of 100 mg/dL or higher. The use of an active comparator, rather than placebo or no treatment, was designed to reduce the likelihood that findings simply reflected better access to clinical care.
“Metformin add-on therapy to usual antiseizure medication care in people with epilepsy and no evidence of diabetes seems to be associated with lower net seizure-related outcomes,” the authors wrote.
The rationale for investigating metformin in epilepsy draws on preclinical evidence that the drug activates adenosine monophosphate-activated protein kinase (AMPK) signalling and inhibits mechanistic target of rapamycin (mTOR) pathways, both frequently dysregulated in epilepsy. Animal studies have shown it attenuates seizure generation and delays epilepsy onset.
In humans, a small randomised trial in tuberous sclerosis complex showed improvements in seizure control. But a 60-patient Indian trial of 500 mg twice daily metformin in active epilepsy found no significant reduction in seizure frequency versus placebo at six months, though seizure frequency declined significantly from baseline in the metformin arm. The authors of that trial called for larger studies with longer follow-up and higher doses.
The first cohort, comprising people without diabetes who experienced an incident brain insult (ischaemic stroke, intracranial haemorrhage, head injury or brain tumour), showed no significant benefit from metformin for epilepsy prevention. HRs for first seizure, first epilepsy diagnosis and first antiseizure medication prescription all failed to reach significance.
The authors noted this antiepileptogenic analysis may have been underpowered because metformin was rarely started immediately after brain insult, limiting assessment of the earliest post-injury period when seizure risk is highest; and the heterogeneous mix of brain insults may have reduced sensitivity to detect insult-specific effects.
Approximately 55% of people with epilepsy are overweight or obese, the authors noted, making metformin, already widely prescribed off-label for weight management, a pragmatically attractive candidate for further study.
“If replicated, these findings could support prioritising metformin as a candidate for repurposing trials in epilepsy, particularly in individuals with comorbid obesity, where weight management is already a clinical consideration,” they wrote.
The authors drew an explicit parallel with fenfluramine, originally a weight-loss drug that was later found to have antiseizure properties and ultimately received regulatory approval for Dravet syndrome and Lennox-Gastaut syndrome.
A word of caution
An editorial commentary accompanying the paper, by Associate Professor Wesley Kerr and Katherine McFarlane from the University of Pittsburgh, urged readers to scrutinise the study’s methodology before drawing clinical conclusions [link here].
The commentary used directed acyclic graphs to map potential causal and non-causal pathways. Kerr and McFarlane pointed out that the study’s exclusion criteria, which removed anyone with coded pre-diabetes, abnormal glucose or HbA1c, relevant medications, or antidiabetic ICD codes, likely captured fewer than 10% of real-world metformin users.
They also raised the possibility that metformin adherence may improve adherence to antiseizure medications, meaning ASM adherence rather than metformin itself could account for some of the observed benefit. “As ASMs have a direct causal effect on seizures, there may or may not be an additional direct metformin effect,” they wrote.
On the primary outcome of time to first seizure code (R56), the commentary noted the code’s limitations. In the epilepsy cohort, an R56 code could reflect a breakthrough seizure, but could equally indicate billing complexity, ASM refill justification, electroencephalography, or adherence counselling. “Unfortunately, EHR data struggle to resolve those possibilities,” Kerr and McFarlane wrote.
Their bottom line was measured: “The observed association of metformin use with improved outcomes was exciting, and the lack of pathophysiological mechanistic understanding limits conclusions about direct causation.”
Limitations acknowledged by the authors included inability to account for metformin dose or frequency, an assumption of treatment adherence, potential immortal time bias from the per-protocol-type exposure definition, and the inherent limitations of ICD-coded seizure outcomes in administrative data.
The authors estimated that a future confirmatory trial using the composite morbidity endpoint and the observed HR of 0.729 would require approximately 400 participants with 10-year follow-up.