Nothing in this evidence says you cannot, and nothing says the drug will help the epilepsy. A matched cohort of 452,766 adults with type 2 diabetes recorded incident epilepsy in 2.35% of GLP-1 users against 2.41% of DPP-4 comparators [1]. That is a hazard ratio of 0.84 (95% CI 0.78 to 0.90). The 16% relative reduction is an absolute difference of 0.06 percentage points, or roughly one case avoided per sixteen hundred people treated. Everyone in it was free of epilepsy at baseline, so it is a question about developing the condition rather than living with it. No trial has tested one of these drugs in people who already have it.
This is a well-conducted study whose headline is nearly meaningless to an individual [1]. It is the clearest example on this site of why both numbers must be printed. The general problem is set out in the semaglutide and tirzepatide comparison.
Across 452,766 matched patients, epilepsy developed in 1,670 people on a GLP-1 drug and 1,886 on a DPP-4 inhibitor. That is 2.35% against 2.41%. The hazard ratio of 0.84 (95% CI 0.78–0.90) accounts for follow-up time and is correctly calculated; it will be reported as a 16% reduction in epilepsy risk.
The quality of the study and the size of its finding are separate judgments and should not be collapsed. This one is large, and the associations held at one, three and five years. They were consistent across age and sex, and sensitivity analyses supported them, comparable rigor to the venous thromboembolism cohort. Semaglutide showed the strongest association at a hazard ratio of 0.68. That rests on the same near-identical incidences, so it is a larger relative figure attached to the same tiny absolute one.
Whether there is a mechanism is genuinely open. Epilepsy and metabolic disease share pathways that these drugs plausibly touch. This is not the first neurological association to appear in observational data while trial evidence stays quiet. The same pattern is described in lower odds of getting Parkinson's, no help once you have it, where prevention signals and treatment nulls came from different study designs.
Where the rest of the neurological evidence sits
Two larger poolings frame this one. A systematic review screened 10,037 records and pooled 82 studies of neuropsychiatric outcomes, graded with GRADE [2]. Risk of developing Parkinson’s disease was lower at a pooled hazard ratio of 0.70, 95% CI 0.53 to 0.92, from observational data. Motor symptoms in people who already had it did not improve, mean difference −3.29, 95% CI −7.84 to 1.26, from trials. A Bayesian network meta-analysis of 712,287 patients found no antidiabetic class with a significant difference in Parkinson’s risk [3]. Its own conclusion nonetheless describes GLP-1 and SGLT2 drugs as consistently associated with lower risk. The gap between a paper’s results and its conclusion is set out in the Parkinson's evidence.
The practical translation is short. Nobody should choose or avoid a diabetes drug on epilepsy grounds from this evidence, and nobody with epilepsy should expect a GLP-1 drug to do anything for it. What the study contributes is a reassurance: these drugs do not appear to raise seizure risk. That is useful to have measured at this scale, in the same way the respiratory null in the respiratory adverse event review is useful.