Not once you have it. A systematic review screened 10,037 records and pooled 82 studies. It found people on these drugs less likely to develop Parkinson’s disease, at a pooled hazard ratio of 0.70 (95% CI 0.53 to 0.92) [2]. In people who already had it, motor symptoms did not improve: a mean difference of −3.29 points on the standard rating scale, on an interval from −7.84 to 1.26. The prevention figure comes from observational cohorts and the null comes from randomized trials, which is the wrong way round. A Bayesian network of nine cohorts covering 712,287 patients found no antidiabetic class with a statistically significant difference in risk at all [1].
Neuroprotection is the most exciting claim anyone makes about these drugs and the least established. The pattern below is the reason: prevention signals keep appearing in observational data, and treatment effects keep failing to appear in trials. The same split runs through the dementia evidence, where a risk score moved and no clinical outcome did.
What the prevention evidence is made of
The neuropsychiatric review applied GRADE across outcomes, which is more rigor than most syntheses in this field manage [2]. Its Parkinson’s incidence figure of 0.70 rests on observational studies. People prescribed a GLP-1 differ from people not prescribed one in age, wealth, insurance, comorbidity and how recently they entered care.
Parkinson’s develops over decades and is often preceded by years of subtle symptoms. A drug given more often to healthier, better-followed patients will look protective against a slow neurodegenerative disease whether or not it is. The split between prevention and treatment is set out in the neuropsychiatric review.
What was pooled in the network analysis
Nine observational cohort studies, 712,287 patients, searched to August 2025, comparing Parkinson’s incidence across antidiabetic drug classes in a Bayesian network that allowed indirect comparisons [1]. Results were expressed as risk ratios with 95% credible intervals, with prespecified subgroups by age, sex and cardiovascular disease.
In the authors’ own words in the results section: no antidiabetic class demonstrated a statistically significant difference in Parkinson’s risk.
What the analysis did produce was rank probabilities. SGLT2 inhibitors tended to rank lowest for Parkinson’s risk among people aged 75 and over, and GLP-1 agonists lowest among those under 75. Metformin, sulfonylureas and alpha-glucosidase inhibitors ranked consistently higher than the newer agents.
The sentence that does not follow
The conclusion states that SGLT2 inhibitors and GLP-1 agonists were consistently associated with lower Parkinson’s risk, suggesting potential neuroprotective effects.
That is a stronger claim than the results support, and it is the sentence that will travel. Nothing was statistically significant. The word doing the work is “consistently,” which describes the rankings rather than the estimates. Reporting rank probabilities is normal practice and not the problem. Converting them into an association is the step that does not follow, and it is the same move as reading a result into an estimate the comparison cannot carry.
What happened when a neurodegenerative trial did report
Parkinson’s has no large completed randomized readout. Alzheimer’s does. Two phase 3 trials tested oral semaglutide 14 mg in early-stage symptomatic Alzheimer’s disease and did not slow cognitive decline [3]. That is the closest thing this field has to a test of the neuroprotection story, and it failed.
A post hoc analysis of the SELECT trial did move a proteomic dementia risk signature in older adults with obesity and cardiovascular disease [4]. A signature is a predictor, not a diagnosis, and no clinical dementia outcome was measured. Both readouts are covered in the Alzheimer’s trials and the dementia risk score.
What to do with the neuroprotection story
Wait for trials in Parkinson’s specifically. Until those report, the honest summary is that nobody has shown these drugs prevent neurodegenerative disease, and neither analysis here is an exception. The one result in the review that is both randomized and countable concerns binge eating disorder, where liraglutide cut binge frequency by 1.28 episodes (95% CI −1.67 to −0.89).
No seller on this roster claims a neurological benefit, and none should. The tell to watch for, if one ever does, is the shape of this paper. A results section that found nothing sits beside a conclusion that found something. The other thing to look for is whether anything is published at all.