A 2026 study analyzing data from over 200,000 participants across two major biobanks found no significant link between cannabis use and accelerated cognitive decline or dementia in older adults. However, the crude way cannabis exposure was measured and the healthy-volunteer bias in both cohorts mean this is not a clean bill of health, particularly for heavy or current users of high-potency products.

Cannabis use is rising faster among adults over 60 than in any other age group, yet clinicians have had virtually no large-scale evidence to guide conversations about long-term cognitive safety. Patients are asking whether cannabis will affect their memory, and until now the evidence base has been limited to small studies or populations too young to inform dementia risk. This study, drawing on two of the world’s largest biobanks and a genetic causal analysis, represents the most comprehensive attempt to date to answer that question, making its findings and its limitations directly relevant to everyday clinical practice.

As cannabis legalization expands globally, older adults represent the fastest-growing segment of new and returning cannabis users, yet nearly all prior research on cannabis and cognition has focused on younger populations or relied on small samples with short follow-up. This study leverages two of the world’s largest biobanks to address the question directly. In the UK Biobank, self-reported lifetime cannabis use was linked to cognitive performance across five domains measured at baseline and again at follow-up. In the Million Veteran Program, ICD-diagnosed cannabis use disorder was linked to incident all-cause dementia over a mean follow-up of 4.3 years. The investigators also applied two-sample Mendelian randomisation, using genetic variants associated with cannabis use as instruments for causal inference.

At baseline, cannabis users showed modestly higher scores on numeric memory (beta = 0.07 SD) and fluid intelligence (beta = 0.12 SD), but these small cross-sectional differences disappeared entirely in longitudinal analyses, suggesting confounding rather than benefit. Cannabis use disorder was not significantly associated with incident dementia (HR = 1.11, 95% CI 0.97 to 1.26, p = 0.12), though the upper confidence bound does not exclude a clinically meaningful 26% increase in risk. Mendelian randomisation analyses found no evidence of a causal relationship in either direction. The authors conclude that prior or occasional cannabis use does not appear to substantially accelerate cognitive aging, but they emphasize that the crude exposure measures, healthy-volunteer bias, and relatively short dementia follow-up mean that dose-dependent harms, particularly from heavy or current use of high-potency products, remain uncharacterized and require dedicated prospective study.

Cannabis and the Aging Brain: Large Study Finds No Accelerated Cognitive Decline, But Key Gaps Remain

With cannabis use rising faster among adults over 60 than in any other age group, the question clinicians are increasingly asked, “Will this hurt my memory?”, has until now had no satisfying evidence-based answer. A new study using two of the world’s largest biobanks and a novel genetic causal analysis offers the most comprehensive attempt yet to answer that question, and the headline finding is reassuring. But the details tell a more complicated story. What this paper actually tested was whether people who reported ever having used cannabis, often decades earlier and at unknown doses, showed faster cognitive decline than those who never used it, and whether veterans with a clinical diagnosis of cannabis use disorder developed dementia at higher rates over roughly four years. The convergence of null findings across two independent cohorts, five cognitive domains, incident dementia, and Mendelian randomisation is genuinely meaningful. Before any criticism, this study’s contribution should be acknowledged: it fills a real void in a literature that has been starved of large-scale, longitudinal data in older populations, and the authors are admirably honest about what their data can and cannot show. The cross-sectional finding that cannabis users had slightly better baseline cognitive scores is properly flagged as likely confounding rather than celebrated as a benefit, a responsible interpretive choice that not all research teams would make.

The central methodological problem, however, is the quality of exposure measurement. The UK Biobank asks only whether participants ever used cannabis and captures a rough lifetime frequency, with no information on dose, potency, THC-to-CBD ratio, recency, route, or duration. The Million Veteran Program captures only clinically diagnosed cannabis use disorder, a severe phenotype covering a small fraction of all users. To put this in perspective, imagine studying whether exercise affects heart disease by asking people only whether they have “ever exercised,” without knowing if they ran marathons for twenty years or walked once in college. You would be unable to detect any dose-dependent relationship, and your null finding would say very little about whether intensive exercise is protective or harmful. That is precisely the situation here. A dose-response relationship showing cognitive harm only at high doses, or only with prolonged heavy use of today’s high-THC products, would be entirely invisible in these data. Both cohorts are also subject to healthy-volunteer and survivor bias: people who are healthy enough to volunteer for a biobank visit or maintain veteran health system engagement are systematically different from those most affected by heavy substance use. The study is measuring the least vulnerable users and generalizing to everyone.

So what would I actually say to a patient asking about this? I would say the best available large-scale evidence does not show that having used cannabis in the past, especially occasionally, dramatically accelerates memory loss or significantly raises dementia risk. But I would be clear that the research cannot tell us about heavy, frequent, or current use of today’s stronger products, and I would not present this as a green light. To a colleague, I would describe this as a well-executed null finding that adds to a sparse literature but cannot rule out dose-dependent harms, and I would not change my counseling approach on the basis of this paper alone. To a policymaker, I would say that public health monitoring systems urgently need to capture actual exposure characteristics, not just whether someone ever used cannabis, if we want evidence that can genuinely inform regulation. The durable lesson from this study is one that applies far beyond cannabis research: null findings from large observational studies are only as strong as the quality of exposure measurement. When the biologically relevant variables are dose, potency, and recency, a study that cannot measure them cannot definitively rule out harm. It can only report that harm was not detectable with the instruments available.

This study occupies an important position in the research arc on cannabis and neurocognitive aging. Prior work has been dominated by small cross-sectional studies in younger populations, retrospective analyses with significant selection bias, and a handful of neuroimaging studies with limited clinical outcome data. The addition of longitudinal cognitive assessments from a large general-population cohort, incident dementia data from a veteran cohort, and Mendelian randomisation represents a meaningful step forward in methodological rigor. However, the study is best understood as a necessary early stage of evidence development rather than a definitive answer, particularly because it cannot characterize the exposure parameters most relevant to risk.

From a pharmacological standpoint, the study’s inability to distinguish between THC-dominant and CBD-dominant products, between smoked and ingested routes, and between occasional historical use and daily current use represents a substantial gap for clinical translation. Older adults frequently use cannabis alongside multiple prescription medications, and potential interactions with anticholinergic drugs, benzodiazepines, and other CNS-active agents are not captured in this analysis. Safety in polypharmacy contexts remains entirely unaddressed. The one concrete recommendation clinicians can take from this study is to continue routinely asking older patients about cannabis use history, including dose, frequency, and product type, and to counsel them honestly that current evidence neither confirms nor rules out risk from heavier or more frequent use patterns.

This is a peer-reviewed original research article published in BMJ Mental Health combining cross-sectional and longitudinal observational cohort analyses from two large national biobanks with supplementary two-sample Mendelian randomisation. As an observational study, it sits below randomized controlled trials in the evidence hierarchy and can identify associations and test for consistency with causality but cannot definitively establish causal relationships. The single most important inference constraint is the crude measurement of cannabis exposure, which prevents characterization of dose-response relationships.

The null finding for longitudinal cognitive decline is broadly consistent with several smaller observational studies that have failed to detect large cognitive harms from cannabis in middle-aged and older populations, though prior work has been limited by sample size and follow-up duration. The finding contrasts with some neuroimaging studies suggesting structural brain differences in chronic cannabis users, a discrepancy that may reflect the difference between detectable neural changes and measurable functional impairment, or may simply reflect the inability of this study to capture heavy chronic users. The application of Mendelian randomisation to the cannabis and dementia question appears to be novel and adds a methodologically distinct perspective that did not previously exist in this literature. The non-significant but non-trivial dementia hazard ratio of 1.11 is consistent with population-level analyses suggesting that if cannabis increases dementia risk, the effect size is likely modest and may require decades of follow-up or more precise exposure measurement to detect reliably.

The most consequential analytic choice was the operationalization of cannabis exposure as a simple binary variable (ever vs. never in UKB) or a clinical disorder diagnosis (in MVP). Had the investigators been able to stratify by dose, frequency, duration, recency, or product potency, a dose-response relationship showing harm at higher exposure levels could potentially have emerged even within a dataset that shows a null overall association. Additionally, the mean follow-up of 4.3 years for dementia outcomes in MVP is short relative to the typical latency of neurodegenerative disease. An analysis with 15 to 20 years of follow-up, or one that used prodromal cognitive biomarkers rather than clinical dementia diagnosis, might have yielded different results. The Mendelian randomisation analyses, while valuable, relied on small instrument sets of 8 to 23 SNPs, and stronger instruments from larger cannabis GWAS could meaningfully alter MR estimates in future studies.

The most likely and most consequential misreading of this study is the conclusion that “cannabis is safe for older adults’ brains.” The study does not demonstrate safety. It demonstrates absence of detected large harm in populations with crude exposure measurement and healthy-cohort selection. A non-significant p-value of 0.12 for the dementia hazard ratio, with the upper confidence interval reaching 1.26, is a finding with insufficient power to exclude clinically meaningful risk, not evidence of no effect. Similarly, the cross-sectional finding that cannabis users had slightly better baseline cognitive scores should not be interpreted as evidence that cannabis improves memory. The authors themselves attribute this to residual confounding and healthy-user bias, and it vanishes entirely in longitudinal analysis. Finally, the Mendelian randomisation results do not “prove” that cannabis does not cause dementia; they are consistent with no causal effect given the instruments available, but are limited by small instrument sets and untestable pleiotropy assumptions.

This study contributes the largest and most methodologically diverse analysis to date of cannabis use and cognitive aging, providing convergent null findings across two independent cohorts and Mendelian randomisation. It offers limited reassurance that occasional or historical cannabis use does not produce large, detectable accelerations in cognitive decline or dramatically elevated dementia risk. It does not establish safety for heavy, current, or high-potency use, cannot characterize dose-response relationships, and does not exclude a modest increase in dementia risk. Clinical practice should continue to include routine assessment of cannabis use patterns in older patients, with honest counseling that important uncertainties remain.

Does this study prove that cannabis is safe for my brain as I get older?

No. The study found no clear evidence that having used cannabis in the past is linked to faster memory decline or significantly higher dementia risk. However, the way cannabis use was measured was very basic, capturing only whether someone had ever used it, with no details about how much, how often, or how recently. The study cannot address whether heavy or current use of today’s stronger products poses a risk, so it should not be taken as proof of safety.

Did this study find that cannabis improves memory?

Cannabis users in the UK Biobank did score slightly higher on some cognitive tests at baseline, but this difference was very small and disappeared when researchers tracked changes in cognition over time. The most likely explanation is that people who choose to use cannabis tend to differ from non-users in ways that also affect cognitive test performance, such as education, lifestyle, and overall health. The researchers themselves caution against interpreting this as a benefit of cannabis.

Should I stop using cannabis based on this study?

This study alone does not provide a reason to stop or start using cannabis. If you are using cannabis for medical purposes, the findings offer some reassurance that occasional use is unlikely to cause dramatic cognitive harm, but they do not address your specific situation, dose, or product type. The best course of action is to discuss your cannabis use openly with your physician, who can help weigh the potential benefits and risks in the context of your overall health.

What does the Mendelian randomisation part of this study mean?

Mendelian randomisation is a technique that uses genetic variants associated with a behavior (in this case, cannabis use) to test whether that behavior might have a causal effect on a health outcome. Because genes are assigned randomly at conception, this approach can help distinguish true causal effects from coincidental associations. The MR analyses in this study found no evidence that cannabis use causally affects cognition or dementia risk, but the genetic instruments available were limited, so this should be considered suggestive rather than definitive.

Have thoughts on this? Share it: