select search filters
briefings
roundups & rapid reactions
Fiona fox's blog
A study published in JAMA Network Open looks at the association between people with obesity and diabetes taking weight loss drugs, and risk of neurodegenerative diseases, stroke, and all-cause mortality.
Dr Sarah Marzi, Senior Lecturer in Neuroscience and UK DRI Group Leader, UK Dementia Research Institute at King’s College London, and Institute of Psychiatry, Psychology and Neuroscience, King’s College London, said:
Is this good quality research? Are the conclusions backed up by solid data?
“This is retrospective study in over 60000 individuals with type 2 diabetes and obesity who were using antidiabetic drugs between 2017 and 2024. The authors looked at the incidence of neurological diseases and mortality. They showed that people taking glucagon-like peptide 1 receptor agonists (GLP-1Ras), such as semaglutide, was associated with a lower incidence of dementia, stroke and all-cause mortality, but not associated with Parkinson’s disease or mild cognitive impairment. The hazard ratio for developing dementia with GLP1-RA treatment compared to other diabetic drugs was 0.63. Or maybe more easily interpretable: The cumulative probability of developing dementia on GLP1-RA after 7 years was 1.63%, whereas it was slightly higher (1.98%) in the group with other antidiabetics. The study seems well executed and open about the limitations. There could have been some more detail on the methods, but I suspect that has to do with the format of the publication.
How does this work fit with the existing evidence?
“It has been hypothesised that GLP1-RAs may have protective effects in the brain, particularly in the context of dementia, possibly through lowering neuroinflammation or promoting neurogenesis. There is increasing epidemiological evidence that supports this, for example this meta analysis of clinical trials of GLP1-RAs: https://jamanetwork.com/journals/jamaneurology/fullarticle/2831975
Have the authors accounted for confounders? Are there important limitations to be aware of?
“They used propensity weighting to account for various factors that might bias the outcome, like sex, age, ethnicity, BMI or hospitalisation. This is good and what should be done in these type of observational studies. If there is a difference in the two populations that receive the different drugs, that could easily affect their risk to develop neurodegenerative or other neurological conditions. For example, the proportion of GLP1-RA users who were within hospital inpatient care was much higher than in the comparison group – and this could indicate worse diabetes symptoms or other health complications that may increase risk for neurological disease. The propensity weighting should account for these differences. However, it only works for variables that were actively measured, and may overlook other relevant factors. The authors are clear about the limitations in their discussion, also saying that only a randomized controlled trial would establish causality and that it would be important to investigate underlying biological mechanisms. One thing I would also note is that the studied population is slightly young for the investigation of neurodegenerative diseases. Late onset Alzheimer’s disease typically starts after the age of 65 and the probability increases as people age. The study population here was around 58 years of age on average when originally recruited, so should have been around 65 at 7-year follow-up. This would be when people are only about to start to develop some of these diseases.
What are the implications in the real world? Is there any overspeculation?
“If shown to be protective for neurodegenerative diseases in future trials, GLP1-RAs could potentially be used clinically in disease prevention in the future, so this is definitely important – but we are not there yet. No overspeculation on behalf of the authors.”
Dr Richard Oakley, Associate Director of Research and Innovation, Alzheimer’s Society, said:
“It is well established that diabetes and obesity can increase your risk of developing dementia. This study retrospectively examines whether GLP-1RAs drug, such as semaglutide and tirzepatide which are used to treat diabetes, can also reduce a person’s dementia risk.
“This study supports existing evidence that shows these drugs may reduce dementia risk, particularly for people aged 60 and over who are living with Type 2 diabetes and obesity.
“Although interesting, we can’t draw conclusions from this study alone as it is an observational study, only a small number of people who took part went on to develop dementia and as the impact of these drugs on different types of dementia is not clear.
“There are clinical trials currently looking at whether drugs like these can be used to treat early-stage Alzheimer’s disease, so this is a really exciting area being explored in the research fight against dementia.”
Prof Tara Spires-Jones, Director of the Centre for Discovery Brain Sciences at the University of Edinburgh, Group Leader in the UK Dementia Research Institute, and Past President of the British Neuroscience Association, said:
“This is a very interesting study adding to evidence that GLP1 receptor agonists are associated with a lower risk of dementia in people with type 2 diabetes and obesity. This study by Lin and colleagues looked at data from over 60,000 people and found an association between taking GLP1 receptor agonists semaglutide or tirzepatide for 7 years and reduced risk of dementia, stroke, and all-cause mortality (death). This type of study cannot determine whether the drugs reduced disease risk by directly protecting the brain. It is highly likely that effectively treating type 2 diabetes and obesity would reduce dementia and stroke risk as they are known risk factors for these conditions. Further work is needed including randomised clinical trials to confirm these drugs are protective in people with diabetes and obesity and other trials are needed to determine whether these drugs will be protective in people who do not have type 2 diabetes and obesity.”
Dr Coco Newton, Senior Research Fellow, Institute of Cognitive Neuroscience, UCL; and Health Systems Group, University of Cambridge, said:
“This is a rigorous study and suggests important therapeutic effects of GLP-1RAs beyond glycemic control. However, the protective effects against dementia should be taken with caution. Three types of dementia outcomes were investigated – Alzheimer’s, vascular, and ‘other’. Although there was an overall lower risk of dementia associated with GLP-1RAs, the sub-group analysis revealed that this was only the case for ‘other’ dementia, but not for Alzheimer’s disease or vascular dementia – the two most common forms of dementia. What constitutes ‘other’ dementia is unclear. The relatively short average follow-up of 1.7 years is far less than the time it takes to develop symptoms of a dementia disease and access a diagnosis, so a longer follow-up time should be investigated before making claims around dementia protection.”
Prof Kevin McConway, Emeritus Professor of Applied Statistics, Open University, said:
“This study adds to previous evidence suggesting that, in people who have type 2 diabetes and are overweight, taking the newer GLP-1RA drugs to manage and alleviate those conditions might also lead to benefits in terms of reduced rates of some neurological conditions such as dementias, and of stroke.
“I think it’s a careful and competent study of its type. But it doesn’t yet come near showing with any certainty that talking these drugs definitely causes reduced risk of these neurological and brain conditions. Also, since everyone in the study already had type 2 diabetes and obesity, and was aged 40 or over, the results can’t tell us anything direct about people who aren’t in that group.
“That’s why the brief press release, and the abstract (summary) of the research paper, rightly don’t go beyond a suggestion that these GLP-1RA drugs might have a protective effect, even in people with diabetes and obesity, but instead say that their results mean that further clinical trials are called for.
“The newer GLP-1RA drugs being studied are semaglutide (marketed as Ozempic, Rybelsus or Wegovy) and tirzepatide (marketed as Zepbound or Mounjaro).
“The researchers for this study are based in Taiwan. For the study they used data from deidentified health records from 67 US health care organisations, made available through a research network called TriNetX. The researchers used data on people aged 40 and over with type 2 diabetes and obesity, who had started as new users of semaglutide, tirzepatide, or other antidiabetic drugs between 2017 and 2024. They excluded from their analysis patients who had previously been prescribed one of the earlier GLP-1RA drugs.
“The primary outcomes that were analysed were new diagnoses of neurodegenerative diseases, including dementia, mild cognitive impairment, and Parkinson’s disease, and also diseases of blood circulation to the brain, including strokes (where a blood clot blocks the blood supply to part of the brain) and intracerebral haemorrhage (bleeds in the brain).
“The study found that there were fewer new cases of several, but not all, of these conditions in people who had started taking semaglutide or tirzepatide, compared to people who had started on a different antidiabetic drug that was not a GLP-1RA.
“However, this was an observational study – so not like a randomised clinical trial where people are allocated at random to one of the drug treatments. That means that there will, inevitably, be some other differences between the people taking the GLP-1RA drugs and people taking other kinds of drug, apart from which antidiabetic drug they were taking. So it would remain possible that any difference in diagnosis rates, for the conditions they were looking at, between those on GLP-1RAs and those on other drugs, was caused by one of these other factors and not by the drugs themselves.
“Of course the researchers were aware of this possibility, and they tried to allow for it using a statistical procedure called propensity score matching. They found factors, that were recorded on their database, that were associated with the chance of being prescribed a GLP-1RA drug, and used them to construct a statistical model giving a score for how likely each person was to be prescribed a GLP-1RA drug. Then each of the more than 30,000 patients who was prescribed a GLP-1RA was matched with a patient who was prescribed a different drug, on the basis of this score. Here the so-called propensity scores were based on people’s age, sex, ethnicity, BMI and various other aspects of their lives and their previous health. Then in the statistical analysis, each patient was primarily compared with the person they were matched with.
“This is a standard statistical procedure these days, but it doesn’t get the researchers off the hook of not being able to conclude that the different type of drug actually cause differences in the risk of being diagnosed with one of the diseases they were interested in.
“That’s partly because there’s no way to be sure that all relevant factors are included in the statistical model that produces the propensity scores. For instance, the researchers couldn’t include factors that are not recorded in the database they had – they mention the patient’s frailty as one example of something quite possibly relevant that was not on the database.
“And basically that’s why the researchers, rightly, don’t go further than suggesting that their findings are a reason for doing clinical trials rather than just more observational studies.
“The research found evidence that was reasonably solid statistically of a reduced risk of diagnosis of dementia and of stroke in patients who were prescribed semaglutide or tirzepatide, compared to patients prescribed another antidiabetic drug. But don’t forget that they can’t show that these associations are one of cause and effect. They might be, but they might not be.
“Also, all these findings apply only to patients like those in the study – that is, people aged 40 or more who already had both type 2 diabetes and obesity.
“They did not, however, find good statistical evidence of a reduced risk of Parkinson’s disease, or mild cognitive impairment, or bleeds in the brain in people taking GLP-1RA drugs.
“That can’t be taken to mean that the drugs definitely don’t lead to reductions in the risk of those conditions. It’s possible that they don’t lead to risk reductions or risk increases. But it’s also possible that the study, despite the large number of participants, didn’t provide enough evidence one way or the other. Only just over 100, out of the over 60,000 people studied, had a Parkinson’s diagnosis and that’s not really enough to come to clear conclusions. Or it’s also still possible that the effect of other unrelated factors, not accounted for by the propensity scores, disguised an association that would otherwise be detectable. That’s always a risk with observational studies.
“The study made one other interesting finding, which actually arose from a restriction in the data tools the researchers had available. Imagine that, for some reason, patients on the GLP-1RA drugs had a higher death rate than patients on the other antidiabetic drugs. Then perhaps the GLP-1RA patients would have a lower risk of being diagnosed with one of the diseases being studied, simply because they would have been more likely to die of something else first. There are standard statistical methods for getting round this issue, but they could not be used with the available database.
“Therefore the researchers decided to use death from any cause (so-called all-cause mortality) as a secondary outcome of this study, as well as the primary outcomes about neurological conditions, strokes and brain bleeds. In fact. they found that patients on the GLP-1RA drugs had a lower risk of death, during the study, than patients on the other antidiabetic drugs, not a higher risk, again using the propensity scoring method. So the lower diagnosis rates for stroke and dementia, that they found in their primary data analyses, weren’t simply an odd consequence of differences in mortality rates.
“This conclusion about death rates is subject to the same provisos as the other conclusions – we can’t conclude that the difference in death rates is actually caused by the different drugs that people were taking for their diabetes, though it certainly doesn’t rule that possibility out.
“And it raises the interesting question of whether the associations between the drugs people were talking and their risks of diagnoses of the specific conditions of interest could look different, possible stronger, if differences in risk of death from any cause could have been taken into account directly in measuring those associations.”
‘Neurodegeneration and Stroke After Semaglutide and Tirzepatide in Patients With Diabetes and Obesity’ by Huan-Tang Lin et al. was published in JAMA Network Open at 16:00 UK time on Tuesday 15 July 2025.
DOI: 10.1001/jamanetworkopen.2025.21016
Declared interests
Dr Sarah Marzi: “No conflicts of interest on my part (no industry funding etc).”
Dr Richard Oakley: “Nothing to declare.”
Prof Tara Spires-Jones: “I have no conflicts with this study but have received payments for consulting, scientific talks, or collaborative research over the past 10 years from AbbVie, Sanofi, Merck, Scottish Brain Sciences, Jay Therapeutics, Cognition Therapeutics, Ono, and Eisai. I am also Charity trustee for the British Neuroscience Association and the Guarantors of Brain and serve as scientific advisor to several charities and non-profit institutions.”
Dr Coco Newton: “No interests to declare.”
Prof Kevin McConway: “I have no conflicts of interest to declare.”