Cannabis Use and Long-Term Cognitive Aging

In This Article

• Key Findings
• Why Long-Term Cognitive Data Have Been Scarce
• Study Design: A 44-Year Cohort
• The Headline Result and Why It Demands Caution
• Initiation Age and Frequency: Two Null Signals
• Reconciling the Finding With the Broader Literature
• What the Preclinical Mechanisms Do and Do Not Explain
• Methodological Constraints That Shape Interpretation
• Clinical and Research Implications
• References

Key Findings

• In a cohort of 5162 Danish men reassessed on the same intelligence test after a mean interval of 44 years, ever-use of cannabis was not associated with greater age-related cognitive decline [1].
• Against a mean decline of 6.2 IQ points across the full sample, users declined 1.3 points less than nonusers in the fully adjusted model, a difference equal to roughly 7% of a standard deviation that the authors characterize as potentially without clinical significance [1].
• Among users, neither age of initiation nor years of frequent use was significantly associated with steeper decline [1].
• The cohort reflects adult-life, largely former use: 92.4% of users had not used cannabis in the year before follow-up, a profile distinct from the adolescent-onset persistent use linked to decline elsewhere [1].

Why Long-Term Cognitive Data Have Been Scarce

The short-term cognitive effects of cannabis are comparatively well characterized: acute intoxication impairs attention, working memory, and psychomotor speed, and regular use in adolescence has been associated with measurable deficits in verbal learning and executive function. What has remained far less settled is whether cannabis use leaves a durable mark on the trajectory of cognitive aging, the slow decline in fluid abilities that unfolds across the adult lifespan.

The reason is structural. Detecting an effect on aging requires a cognitive measurement taken before substantial use begins, a second measurement taken decades later, and a study population large enough to survive the attrition that any multi-decade follow-up imposes. Few datasets satisfy all three conditions at once, which is why the existing longitudinal evidence has been described as sparse and mixed [1]. The Danish study by Høeg and colleagues, published in Brain and Behavior in 2024, is notable precisely because its design clears that bar in a way most prior work could not [1].

Study Design: A 44-Year Cohort

The investigation draws on the Danish Aging and Cognition (DanACo) cohort, assembled by pooling two follow-up studies of identical design [1]. Its central asset is the Børge Prien’s Prøve (BPP), the group-administered intelligence test that all Danish men historically completed at conscription, typically around age 20. Because that baseline score sits in a national archive, the researchers could recontact the same men in late midlife, readminister the identical instrument at a mean age of 64, and compute an individual change score spanning a mean of 44 years [1]. BPP scores were converted to a standard IQ scale (mean 100, SD 15) using the conscription distribution as the reference.

A useful way to picture the design is a single ruler applied to the same plank twice, decades apart: because the measuring instrument does not change, the difference between the two readings reflects change in the plank rather than change in the ruler. The long interval is also a methodological strength rather than merely a logistical feat, because it minimizes the practice effect that shorter retest windows introduce, and because the baseline captured fluid intelligence near its lifetime peak [1].

Cannabis exposure was assessed at follow-up. Ever-use was treated as a binary variable, with further self-reported detail on age of initiation (under 18, 18 to 25, over 25) and, in one of the two sub-studies, on frequency across defined age bands [1]. Frequent use was operationalized as roughly twice weekly or more. The analysis proceeded through five nested linear regression models, culminating in a fully adjusted specification that accounted for age at follow-up, retest interval, baseline IQ, education, extreme binge drinking, smoking, use of other illicit drugs, psychiatric history, and somatic comorbidity [1].

The Headline Result and Why It Demands Caution

The defensible takeaway from this study is narrow and worth stating plainly before anything else: across 44 years, in this large sample, cannabis use was not associated with accelerated age-related cognitive decline [1]. That null is the result clinicians and researchers can lean on.

The study’s more eye-catching number, that users declined 1.3 IQ points less than nonusers in the fully adjusted model, requires considerably more caution, and the authors extend exactly that caution themselves [1]. Three considerations restrain any protective reading.

First, the effect is small. A 1.3-point difference on a scale with a standard deviation of 15 amounts to the roughly 7% of an SD the authors report, an effect they note may carry no clinical significance [1]. Differences of this magnitude are easily generated by residual confounding rather than by any property of the exposure.

Second, the groups differed systematically in ways that plausibly track cognitive trajectory. Cannabis users in this cohort had slightly higher baseline IQ and more years of education, but also smoked more, drank more heavily, and carried a substantially higher burden of psychiatric diagnoses and other illicit drug use [1]. The direction of these confounders is not uniform, which is part of the problem: when the measured covariates pull in opposing directions, unmeasured ones can plausibly do the same, and the residual is unlikely to be zero. The authors are explicit that the association may reflect characteristics of the people who used cannabis rather than an effect of cannabis itself [1].

Third, and most fundamentally, this is observational cohort data. It can establish the absence of a strong harmful signal, which is a genuine contribution, but it cannot license a causal claim in the opposite direction. An apparent benefit in a non-randomized comparison is the precise situation in which confounding-by-indication and selection effects most often masquerade as a treatment effect. The viral-tier reading of this paper, that cannabis protects the aging brain, is not what the data support, and treating the 1.3-point figure as anything more than a confounded, sub-clinical association would overstate what a cohort of this kind can deliver.

Initiation Age and Frequency: Two Null Signals

Beyond the primary comparison, the study tested two dose-related questions that matter clinically. Neither yielded a significant association with decline [1].

Among users, age of initiation, whether before 18, between 18 and 25, or after 25, did not significantly predict the degree of cognitive decline [1]. Likewise, years of frequent use, examined in the sub-study where frequency data existed, showed no significant gradient: men with more than ten years of frequent use did not decline measurably more than those with no frequent use [1].

These nulls deserve a calibrated reading rather than an enthusiastic one. The initiation and frequency analyses were restricted to users only, which shrinks the sample, and the frequency analysis was limited to the smaller of the two sub-studies (roughly 1100 men) with relatively few heavy, sustained users [1]. A null in an underpowered subgroup is weak evidence of no effect; it is the absence of a detectable signal, which is not the same thing. The honest framing is that the study did not find a dose-response relationship, while acknowledging it may have lacked the statistical power to detect a modest one.

Reconciling the Finding With the Broader Literature

Placed alongside comparable work, the Danish null is neither an outlier nor a vindication. It sits within a genuinely mixed literature, and the pattern of agreement and disagreement is itself informative.

The result aligns closely with the longitudinal studies of most similar design. McKetin and colleagues followed 1897 middle-aged Australians over eight years and found that cannabis use was not associated with accelerated cognitive decline, though mid-life users showed poorer verbal recall unrelated to their current use level [2]. Lyketsos and colleagues, tracking 1318 adults under 65 in Baltimore across twelve years, similarly found no significant difference in cognitive decline between heavy users, light users, and nonusers [3]. Both concordant studies share the Danish cohort’s defining feature: they examine adult-life use rather than developmental-window exposure.

Two further studies sharpen the picture in a way worth dwelling on, because each isolates a distinction the headline Danish result tends to blur. Auer and colleagues, using the CARDIA cohort of roughly 3400 American adults followed for 25 years, found that cumulative lifetime exposure was associated with poorer verbal memory specifically, while processing speed and executive function showed no significant association after adjustment [4]. Because that outcome was a single midlife assessment of cognitive level rather than a measured decline slope, it speaks to a slightly different question than Høeg does; the transferable lesson is domain specificity, since a global IQ change score can mask a localized deficit in a single domain, and a null on the composite does not guarantee a null everywhere beneath it. Watson and colleagues, following 297 older adults with HIV for up to a decade, found that occasional users performed better on global cognition cross-sectionally, yet the rate of decline and everyday function did not differ by use level [5]. That dissociation, a better starting point but the same downward slope, is a clean illustration of why a snapshot of cognitive level should never be read as evidence about the trajectory of decline, which is the actual question here. Notably, the vulnerable domain recurs across studies: McKetin’s poorer verbal recall and Auer’s verbal-memory association both implicate verbal memory specifically, a localized deficit that a composite IQ change score of the kind used here would be structurally unable to isolate [2,4].

The most prominent discordant result comes from the Dunedin birth cohort. Meier and colleagues followed 1037 New Zealanders from birth to age 38, with neuropsychological testing at age 13, before any cannabis use, and again at 38, after persistent use had developed in some members [6]. There, persistent cannabis use was associated with broad neuropsychological decline, the effect was dose-dependent, and critically, the impairment was concentrated among those who began in adolescence and did not fully resolve with cessation [6].

The apparent contradiction largely dissolves once the exposure windows are compared rather than conflated. Dunedin captured use that began during ongoing brain maturation and persisted, whereas the Danish cohort captured predominantly adult-onset use that had, for most participants, ceased: 92.4% reported no use in the year before follow-up [1]. The relevant distinction is developmental timing, and the analogy is closer to a sapling than a mature tree. A bend imposed while the trunk is still forming can set into the grain; the same force applied to grown timber leaves little trace once removed. Several studies cited by the authors report that cannabis-related cognitive deficits recover substantially after sustained abstinence, which would further explain why a largely former-using cohort shows no lasting decrement [1]. The Danish data, in other words, do not refute Dunedin so much as describe a different exposure.

What the Preclinical Mechanisms Do and Do Not Explain

Because the “less decline” signal invites a mechanistic story, it is worth being precise about what the laboratory evidence supports and at what confidence. The animal work the Danish authors gesture toward is real, but it belongs to a different evidentiary register and should not be imported to explain an epidemiological association.

The rationale rests on the endocannabinoid system (ECS), the CB1 and CB2 receptor network that modulates neuroinflammation, synaptic plasticity, and oxidative stress. The hypothesis under investigation is that ECS signaling tone declines with age and that modest exogenous stimulation might partially offset age-related changes. In aged rats, the cannabinoid agonist WIN-55212-2 reduced hippocampal microglial activation and improved spatial memory [7]. In mice, a chronic low dose of THC restored learning and memory in older animals while impairing it in young ones, an age-dependent reversal accompanied by hippocampal gene-expression changes [8]; a separate group reported lasting improvement in aged mice from even a single ultra-low THC dose [9].

Three constraints keep this at the hypothesis level. First, the results are preclinical and in rodents, with doses and controlled delivery unlike human use; the “low” THC dose in one study was itself contested as far from low in translational terms [8]. Second, the effect is explicitly age- and dose-dependent, the opposite of a blanket “cannabis helps cognition” claim and an echo of the developmental-window concern, with young brains harmed where old brains were helped [8]. Third, an observational cohort cannot test mechanism: even were the Danish association causal, nothing in the data identifies a pathway. This literature is a reason to design better trials, not to read this one as confirmation.

Methodological Constraints That Shape Interpretation

Several features bound what the study can support, and they integrate naturally with the interpretation above rather than standing apart from it.

The participation rate was 14.3%, and participants were known to have higher test scores, lower morbidity, and more education than non-participants [1]. If men with the heaviest use or the steepest decline were less likely to enroll, the sample would be skewed toward healthier users, biasing the comparison toward a favorable or null result. The cohort is also exclusively male, and the authors caution that cannabis effects on cognition appear to differ by sex, so the findings should not be generalized to women [1].

Exposure rested on retrospective self-report gathered in late midlife, which introduces both recall limitations and the possibility of social-desirability underreporting on a sensitive topic, though prior work suggests users recall their own use histories reasonably accurately [1]. Residual confounding remains the central interpretive limit. The stark imbalance in other illicit drug use, 27.8% among cannabis users versus 0.7% among nonusers, is difficult to adjust away fully and could influence the frequency analysis in particular [1]. Finally, the study did not separate use occurring before the conscription test from use after it; frequent use around the baseline assessment could have depressed the baseline score and thereby distorted the apparent change [1].

None of these caveats overturn the primary null. They do, however, reinforce why the secondary “less decline” finding should not be read as protection.

Clinical and Research Implications

For clinicians counseling adult patients, the practical message is measured. The best long-term evidence to date does not indicate that adult-life cannabis use accelerates cognitive aging, and that is a reasonable thing to convey to a patient who began using in adulthood and uses intermittently. It is not evidence that cannabis benefits cognition, and the developmental signal from adolescent-onset cohorts remains the more clinically actionable concern, particularly for younger patients in whom the brain is still maturing.

Read as a piece of therapeutic groundwork rather than a therapeutic finding, the study’s main contribution to the research agenda is that it lowers a specific barrier. One persistent obstacle to studying cannabinoids in aging and neurodegenerative populations has been the unresolved worry that the exposure itself might hasten cognitive decline, which would complicate both recruitment and risk-benefit calculations. A 44-year null on accelerated decline does not remove that worry entirely, but it narrows it, and in doing so it makes the case for enrolling older adults into well-controlled cannabinoid trials somewhat easier to defend [1].

The abstinence-recovery signal points to a second, more pointed research question. If much of cannabis-related cognitive impairment is a state effect that recovers after sustained cessation rather than a fixed trait-level change, then the relevant therapeutic variables are timing, duration, and washout rather than lifetime exposure alone [1]. Disentangling reversible from durable effects is a question a prospective design with repeated assessment could answer and a retrospective cohort cannot. It also bears on population selection, since the risk signal concentrates in adolescent-onset use rather than in the older adults most relevant to therapeutic cannabinoid applications [6].

For researchers, the study models several design strengths worth emulating: an objectively archived pre-exposure baseline, an identical retest instrument, and a follow-up long enough to neutralize practice effects. It also defines the features the next generation of work will need: biologically verified exposure rather than self-report, contemporary high-potency formulations rather than historical use patterns, inclusion of women, and outcomes anchored to biomarkers such as amyloid, tau, or hippocampal volume rather than psychometric change alone [1]. The transfer problem is compounded by formulation: because cannabis potency has risen substantially since the decades these men used, even a definitive null in this historical cohort would not necessarily hold for today’s far stronger products. Those are the design features that would move the field from population-level association toward mechanistically interpretable, therapeutically relevant evidence, and they are the natural bridge from a study like this one to the controlled preclinical and clinical work that could actually test the ECS hypotheses outlined above.

This piece pairs with our existing examination of cannabinoid neuroprotection and cognitive health, which surveys the endocannabinoid-system mechanisms often invoked to explain cognitive effects, and connects to our forthcoming analysis of CBD as a candidate neuroprotective agent in Parkinson’s disease, where the question shifts from population-level association to a specific mechanistic hypothesis under active investigation.