Scientific deep-dive

GLP-1s and Epilepsy Risk

People on GLP-1 drugs developed epilepsy less often than those on a different diabetes drug (hazard ratio 0.84). But the association was strongest at one year and weaker at five, which is backwards for neuroprotection.

By Nora Bissett · Pricing Editor
Editorially reviewed. Not medical advice · How we verify contentLast reviewed
5 min read·1 citation

A study in Neurology reports that, among people with type 2 diabetes, those on GLP-1 drugs developed epilepsy less often than those taking a different diabetes drug. The hazard ratio was 0.84, with semaglutide strongest at 0.68.[1] The journal grades it Class III evidence, and there is a pattern inside the numbers that points away from the neuroprotection explanation.

What was found

Compared with DPP-4 inhibitor users, GLP-1 users had a lower risk of developing epilepsy, and the association held at every follow-up point examined.[1]

Epilepsy risk on GLP-1 drugs against DPP-4 inhibitors, by follow-up duration.
Follow-upHazard ratio
Overall0.84 (95% CI 0.78–0.90)
1 year0.71 (95% CI 0.62–0.80)
3 years0.81 (95% CI 0.74–0.88)
5 years0.82 (95% CI 0.76–0.88)
Semaglutide specifically0.68 (95% CI 0.60–0.77)

Results held across age and sex subgroups, and a sensitivity analysis excluding patients with overlapping or switching exposure returned a similar figure at 0.71.[1] The comparator is an active one, which is the design property that separates a credible observational result from a misleading one. It is the same reason we treated the pneumonia study more seriously than the cancer cohorts.

The pattern that points the other way

Read the follow-up column downward. The association is strongest at one year and weaker at three and five.

A drug that protects neurons should look better with time, not worse. This looks better early.

That shape is hard to reconcile with the neurologic-benefit hypothesis the authors raise. Protection that works through brain physiology would be expected to accumulate over years of exposure. An effect concentrated in the first year fits other explanations more comfortably: differences in who is prescribed a newer drug, in how closely those patients are watched, or in what else was happening around the time of prescribing.

None of that makes the finding wrong. It means the proposed mechanism does not match the shape of the data, and that should be noticed before anyone builds a story on top of it.

Class III evidence, stated by the journal

Neurology formally classifies the strength of evidence in its papers, and this one is graded Class III.[1] That is the journal itself saying this is observational evidence with recognized limitations, not a demonstration of cause and effect. The authors also name residual confounding as a limitation directly.

A study that comes with its own grade is easier to use. Most observational papers leave the reader to work out how much weight a finding carries. A formal evidence class printed on the paper is a small piece of candor, and it makes this result easier to use responsibly than dozens of louder ones.

This study asks whether these drugs change your chance of developing epilepsy. A different study asks what happens to seizure recurrence in people who already have it, and runs into the same interpretive problem from the other direction (see seizures when you already have epilepsy).

What it means practically

  • It is not a reason to take a GLP-1. No GLP-1 is approved for anything neurological, and epilepsy is uncommon enough that a relative reduction is a very small absolute one.
  • If you already have epilepsy, it is at most weakly reassuring. The data point away from harm, not toward it, but they come from people developing epilepsy.
  • It says nothing about seizure control in existing epilepsy. The outcome was developing epilepsy, not managing it.
  • It joins a broader pattern of proposed neurologic benefits, one of which has now been tested directly and failed: see the Alzheimer’s trials.

That last point deserves weight. The strongest observational signals in this class have repeatedly not survived randomization. Epilepsy has not been tested, and until it is, this belongs in the interesting column rather than the established one.

Frequently Asked Questions

A study in Neurology found that, among people with type 2 diabetes, those on GLP-1 drugs developed epilepsy less often than those taking DPP-4 inhibitors, with a hazard ratio of 0.84, and 0.68 for semaglutide specifically. The journal grades it Class III evidence, meaning observational with recognized limitations.
Because the association was strongest at one year and weaker at three and five. A drug protecting neurons should look better with longer exposure, not worse. An effect concentrated early fits differences in prescribing or monitoring more comfortably than it fits neuroprotection.
No. Nothing in this class is approved for any neurological indication, and the most prominent neurologic hypothesis in this field, that semaglutide might slow Alzheimer's, was tested in two randomized trials that found nothing.
That is a question for your neurologist, and this study does not answer it. The outcome measured was developing epilepsy, not seizure control in people who already have it. The direction of the data is at least not suggestive of harm.
Small. Epilepsy is an uncommon outcome, so a relative reduction of this size corresponds to a very small absolute difference: a few cases across a large population rather than anything an individual would notice.

References

  1. 1.Cheng CY, Lo SC, Huang CN, et al. Association Between GLP-1 Receptor Agonist Use and Epilepsy Risk in Type 2 Diabetes Neurology. 2026. PMID: 41370744.

Semaglutide and Alzheimer's: The Trials Came Back Negative

Two randomized trials in 3,808 people with early Alzheimer's found no difference in cognitive decline at two years. Both were discontinued for negative clinical outcome. The 45% figure came from studies measuring something else.

6 min read

When the Primary Endpoint Fails and a Secondary Doesn't

A randomized trial in 72 people with major depressive disorder found no effect on executive function, its primary outcome. A secondary cognition measure cleared p=0.03, and that is the one that will be reported.

5 min read

Semaglutide and Motivation

Semaglutide reduced how heavily people with depression discounted effort, while leaving their sensitivity to probability unchanged. That dissociation is what makes the finding hard to dismiss.

5 min read

GLP-1 Drugs and Your Kidneys: Protective Over Years, Risky Over Days

Semaglutide cut the risk of kidney failure, a large drop in kidney function, or death from a kidney or cardiovascular cause by 24% over a median 3.4 years. The label also warns of acute kidney injury in people dehydrated by vomiting or diarrhea. Both are true.

7 min read

GLP-1s After a Kidney Transplant

A meta-analysis of 54,680 transplant recipients reports mortality nearly halved and kidney function improving. Four studies stand behind the mortality figure, and the size of the benefit is the reason to be careful with it.

6 min read

GLP-1s and Autoimmune Thyroiditis: An Association, and the Reason to Doubt It

A 290,770-pair cohort found autoimmune thyroiditis diagnosed 30% more often on a GLP‑1 than on a DPP-4 inhibitor. Everyone on a GLP‑1 gets a boxed thyroid warning and the comparison group does not, and this disease is found by testing.

6 min read

Where to get GLP-1 online, safely: sellers our editors have checked

These are telehealth sellers our editors have checked. For each one we hold a price, the form the drug comes in, and the states it reaches.

No insurance needed · vetted by our editors

Some of the links on this page earn us money. If you sign up with a provider after following one, that provider may pay GLP Watchdog a commission. Learn more

6.4

MyDrHank

An oral route if you will not self-inject

9.0

YourEra

Semaglutide at $99/month, well below the typical price

8.7

Eden

Coverage in all 50 states, which most sellers will not confirm

About this analysis

GLP Watchdog publishes independently. Each seller in our dataset is checked against that seller's own website, and the dataset is updated continuously. Read our full methodology →