Cannabis Science
A 606,434-person cohort study published in The BMJ found that GLP-1 receptor agonists — including semaglutide — were associated with a 14% lower risk of developing cannabis use disorder compared to SGLT-2 inhibitors, with a 50% reduction in substance-related mortality among those who already had a substance use disorder. The finding cuts across every major substance type tested, which raises a biologically plausible question: are GLP-1 drugs dampening the craving signal itself?
GLP-1 Drugs Reduced Cannabis Use Disorder Risk 14% in 606,000-Person BMJ Cohort: What the Data Actually Say
When a study of 606,000 people finds that a diabetes medication cuts cannabis use disorder risk — and cuts substance-related deaths in half — it demands careful reading. The numbers are striking. The mechanism is biologically plausible. The limitations are real. Here’s how to hold all of that at once.
This study matters directly to the CED patient population. Cannabis use disorder affects a meaningful subset of medical cannabis patients, GLP-1 medications are already widely prescribed for metabolic conditions that frequently co-occur with cannabis use, and the ECS-GLP-1 receptor overlap is a clinically active area of investigation. Both prevention and harm-reduction signals appear in the data.
Cannabis Use Disorder
Addiction Medicine
Endocannabinoid System
BMJ Cohort Study
- What the BMJ cohort study actually found — by substance type, with exact hazard ratios
- Why the effect appears consistent across six different addictive substances, and what that may mean biologically
- Where the study is strong and where its design creates limits that headline writers ignored
- How the endocannabinoid system and GLP-1 receptor pathways may interact in the brain’s reward circuitry
- What this means — and does not mean — for cannabis patients in Massachusetts and beyond
- A BMJ cohort study of 606,434 US veterans found GLP-1 drugs reduced cannabis use disorder risk by 14% compared to SGLT-2 inhibitors over three years
- Among those with pre-existing substance use disorders, GLP-1 use was associated with a 50% reduction in SUD-related mortality and a 39% reduction in overdose
- The cross-substance pattern — alcohol, cannabis, cocaine, nicotine, opioids all showing similar signal — suggests GLP-1 drugs may act on the craving pathway rather than any single substance
- This is observational data in a specific population (veterans with diabetes) — promising, but not yet a basis for prescribing GLP-1 drugs to treat addiction
| Study Type | Retrospective cohort study (target trial emulation using EHR data) |
| Population | 606,434 US veterans with type 2 diabetes; followed up to 3 years |
| Intervention | GLP-1 receptor agonists (primarily semaglutide, liraglutide, dulaglutide) |
| Comparator | SGLT-2 inhibitors (active comparator design, not placebo) |
| Key Finding (Cannabis) | HR 0.86 (95% CI 0.81–0.90); NRD −2.25 per 1,000 people at 3 years |
| Key Finding (SUD Mortality) | HR 0.50 (95% CI 0.32–0.79) in those with pre-existing SUD |
| Journal | The BMJ, March 4, 2026 (PMID: 41781010) |
| Institution | Washington University School of Medicine; VA Saint Louis Health Care System |
| Funding | US Department of Veterans Affairs; funders had no role in study design or analysis |
| Limitations | Observational design; veterans population (predominantly male); all participants had T2DM; residual confounding possible |
Why This Matters
Cannabis use disorder affects an estimated 9 to 11 percent of regular cannabis users — a figure that has likely grown since legalization expanded across Massachusetts and much of the country. At CED Clinic, many patients who seek medical cannabis certification carry co-occurring metabolic conditions: obesity, type 2 diabetes, hypertension. GLP-1 drugs are already part of the metabolic care conversation for a meaningful fraction of this population.
This study puts both conversations in the same room. If the associations hold in future trials, it would mean a single medication could address two conditions that frequently co-occur and reinforce each other. That’s a different kind of clinical utility than treating one problem at a time — and it’s worth understanding precisely, which means understanding what this study can and cannot claim.
What the Study Found
Researchers at Washington University School of Medicine and the VA Saint Louis Health Care System analyzed electronic health records from 606,434 US veterans with type 2 diabetes. Using a target trial emulation framework — a method designed to extract causal inference from observational data with a level of rigor approaching a clinical trial — they compared people who initiated GLP-1 receptor agonists (primarily semaglutide, liraglutide, or dulaglutide) to people who initiated SGLT-2 inhibitors, another active diabetes medication class. The use of an active comparator rather than no-treatment control is methodologically important: it reduces the tendency for sicker or healthier patients to be concentrated in one arm.
Among the 524,817 participants with no prior substance use disorder at baseline, GLP-1 users showed lower rates of developing every substance use disorder examined over three years. For cannabis specifically, the hazard ratio was 0.86 (95% confidence interval 0.81 to 0.90), translating to approximately 2.25 fewer cannabis use disorder diagnoses per 1,000 people over three years. Alcohol showed an 18% reduction (HR 0.82), opioids 25% (HR 0.75), cocaine 20% (HR 0.80), and nicotine 20% (HR 0.80). The composite across all six substance types reached HR 0.86 with a net risk difference of minus 6.61 per 1,000.
The second protocol, examining 81,617 participants who already had a diagnosed substance use disorder, produced even more striking numbers. Among those with pre-existing SUD, GLP-1 use was associated with a 31% reduction in substance-related emergency department visits (HR 0.69), a 26% reduction in SUD-related hospitalizations (HR 0.74), a 50% reduction in SUD-related mortality (HR 0.50, CI 0.32 to 0.79), and a 39% reduction in drug overdose (HR 0.61). Suicidal ideation or attempt was also lower in the GLP-1 group (HR 0.75). These outcome reductions, if replicated, would represent a substantial public health impact at scale.
What This Paper Does Not Show
This is not a randomized controlled trial. The target trial emulation framework is methodologically sophisticated, but the comparator is SGLT-2 inhibitors — not placebo. SGLT-2 drugs are metabolically active medications that influence weight, glucose, and cardiovascular risk, and their specific effects on brain reward pathways or substance use have not been rigorously characterized. The observed benefit attributed to GLP-1 drugs could be partly or entirely relative to a comparator that has its own distinct effects.
The population is exclusively US veterans with type 2 diabetes — predominantly older, predominantly male, with a high comorbidity burden and access to VA health care. The findings may not generalize to younger cannabis users, women, people without diabetes, or uninsured patients. Substance use disorder diagnoses in electronic health records are known to undercount actual use: many patients do not disclose substance use to their providers, and EHR coding is variable across sites.
The study does not establish that GLP-1 drugs should be prescribed to treat cannabis use disorder. No addiction indication is FDA-approved for GLP-1 medications. The absolute risk differences are modest at the individual level (2.25 fewer cannabis SUD events per 1,000 users over three years), even if the population-level implications of such numbers at scale are significant.
The Biology Behind the Pattern: GLP-1 Receptors and the Reward System
What makes this data compelling beyond the numbers is the cross-substance consistency. If GLP-1 drugs were reducing, say, alcohol use disorder specifically, the mechanism might be explained by drug-specific pharmacology — alcohol metabolism, GABA modulation, something narrow. But the signal appears across cannabis, alcohol, cocaine, nicotine, and opioids simultaneously. That pattern points toward something upstream of any single substance: the craving pathway itself.
GLP-1 receptors are present not only in pancreatic beta cells and the gut — where they modulate insulin and satiety — but in brain regions central to reward processing, including the ventral tegmental area and the nucleus accumbens. These are the structures where dopamine-mediated reinforcement is generated, and where the craving that sustains addiction lives. The hypothesis, grounded in receptor distribution and emerging preclinical data, is that GLP-1 agonists dampen the dopaminergic reward signal, quieting what the senior author Ziyad Al-Aly, MD, called “drug noise” — the persistent preoccupation with the addictive substance that drives use regardless of substance type.
This intersects meaningfully with endocannabinoid system research. The ECS — particularly CB1 receptors in the mesolimbic dopamine pathway — is itself a major modulator of reward processing and addiction. THC and other cannabinoids drive much of their intoxicating and reinforcing effects through CB1 activation in these same circuits. There is emerging preclinical evidence that GLP-1 receptor signaling and endocannabinoid signaling converge in these reward-relevant regions. The interaction has not been fully characterized in humans, but the overlap is not coincidental. It raises a genuinely interesting question for cannabis medicine: whether the therapeutic effects of cannabinoids in conditions like pain and anxiety operate partially through ECS pathways that GLP-1 drugs are also influencing — and whether that has clinical implications for patients on both medication classes.
For cannabis use disorder specifically, the endocannabinoid system is not a bystander. The development of cannabis dependence involves progressive desensitization and downregulation of CB1 receptors in mesolimbic regions. If GLP-1 agonists are modulating the same reward architecture, the mechanism by which they might reduce cannabis craving is at least biologically coherent — even if it remains to be demonstrated in properly controlled studies. Earlier smaller analyses, including a retrospective study of semaglutide in cannabis use disorder populations published in Molecular Psychiatry in 2024, pointed in the same direction. This BMJ cohort — larger by orders of magnitude — gives the signal more statistical weight, though not yet the causal rigor of a clinical trial.
The GLP-1 story keeps expanding, and this is one of the most interesting chapters yet. What I find clinically meaningful here isn’t just the cannabis use disorder risk reduction — it’s the pattern. When the same drug class shows a consistent signal across alcohol, cannabis, cocaine, nicotine, and opioids simultaneously, and when that signal appears in both prevention and harm reduction, it starts to look like something targeting the shared biology of craving rather than a substance-specific mechanism. The brain has one reward system. It doesn’t build six separate pathways for six different addictive substances. If GLP-1 drugs are quieting that system’s overactivity, they may be doing so in a way that’s agnostic to what the person is addicted to.
For the patients I see at CED Clinic, most of whom are using cannabis medically for pain, sleep, anxiety, or other conditions, the practical question is nuanced. Many already have or will have metabolic reasons to discuss GLP-1 medications with their primary care physician. If those medications happen to carry a modest protective signal for cannabis use disorder, that’s relevant information for the clinical conversation — not a reason to start GLP-1 drugs without a metabolic indication, but a reason to be thoughtful. Cannabis use disorder affects a real subset of medical cannabis patients, and right now our pharmacological toolkit for treating it is thin. Any plausible signal in a 600,000-person dataset deserves careful attention. The evidence here is early and observational, but the question it raises is worth taking seriously: are we approaching a era in which GLP-1 drugs serve the endocannabinoid system not by mimicking it, but by modulating the same reward circuits it controls? We need randomized controlled trials to know. But the biology doesn’t feel like a coincidence.
What a Careful Reader Should Take Away
A 606,000-person observational study with an active comparator design is genuinely informative — more so than most studies that make similar headlines. The finding that GLP-1 receptor agonists were associated with a 14% lower hazard of cannabis use disorder, alongside reductions in alcohol, cocaine, nicotine, and opioid use disorders, is consistent with a coherent biological mechanism and is not easily explained away as noise.
At the same time, this is not proof of causation, does not represent a treatment trial, and involves a population (veterans with diabetes) with specific characteristics that may not transfer to the broader cannabis medicine population. The absolute risk differences are modest per individual, even if the population-level implications of modest per-person reductions at scale are significant. The SGLT-2 comparator, while methodologically defensible, is not a neutral control, and its own potential effects on substance use behavior have not been established.
The most defensible read: this is a large, methodologically careful observational study pointing consistently toward a hypothesis that deserves to be tested in a randomized controlled trial. It adds meaningful weight to a growing body of evidence on GLP-1 drugs and addiction. It does not settle the question. And if you or a patient is asking whether to start Ozempic to treat cannabis use disorder, the answer from this data is: not yet — but the biology is interesting enough that the question will keep being asked.
The Same Study Can Mean Different Things Depending on the Question Being Asked
Scientific papers rarely answer a single question. Patients, clinicians, researchers, policymakers, and critics often read the same data differently. The perspectives below explore how this study looks through several evidence-based lenses.
Patient Takeaway
If you use cannabis and are also taking a GLP-1 drug like Ozempic, Wegovy, or Mounjaro for diabetes or weight, this study raises something worth knowing: the medication you’re already on may also be associated with a lower risk of developing a problematic relationship with cannabis. This doesn’t mean the drug cures or prevents cannabis use disorder for everyone, and the reduction observed in this study was modest at the individual level. But for people who already have reasons to be on GLP-1 medications metabolically, a secondary protective signal in the addiction space is clinically meaningful context.
If you’re currently struggling with cannabis use disorder, the study’s second protocol offers a different kind of signal: among people who already had a substance use disorder, GLP-1 users had fewer emergency department visits, fewer hospitalizations, and significantly lower rates of substance-related death. That’s a striking finding, even accounting for the study’s design limitations.
What this study does not mean: it is not a reason to ask your doctor for Ozempic to treat cannabis use disorder. There is no approved indication, the evidence is observational rather than from a clinical trial, and GLP-1 drugs have real side effects. What it does mean is that if your metabolic health warrants a conversation about GLP-1 medications, the potential intersection with substance use is worth raising as part of that conversation. Your doctor can help you weigh the full picture.
Clinician’s POV
The methodological choice of SGLT-2 inhibitors as the active comparator is the study’s most important structural feature. It reduces the healthy-user bias and channeling bias that plague comparisons of treated versus untreated patients. The fact that this comparator is itself a metabolically active drug — not a placebo — means the GLP-1 benefit is measured against a meaningful pharmacological baseline, not nothing. That strengthens confidence in the finding’s direction, even while keeping the effect size uncertain.
The cannabis-specific hazard ratio of 0.86 (95% CI 0.81 to 0.90) is statistically robust given the sample size but modest in absolute terms: 2.25 fewer cannabis SUD diagnoses per 1,000 over three years. For a clinician seeing 200 patients, that translates to less than one prevented case. The Protocol 2 mortality signal (HR 0.50) deserves more attention and more skepticism simultaneously: the confidence interval is wide (0.32 to 0.79), reflecting the smaller protocol 2 cohort, but a 50% reduction in SUD-related death is a clinically consequential number if confirmed.
For patients with co-occurring type 2 diabetes or obesity and cannabis use disorder, GLP-1 drugs may already be clinically indicated for metabolic reasons. This study adds plausible secondary framing to that conversation without changing the prescribing standard. It is not yet appropriate to prescribe GLP-1 drugs specifically for cannabis use disorder, but the shared decision-making discussion for a patient with both conditions is now richer than it was before this paper.
A Skeptical Read
The active comparator design is better than a no-treatment control, but SGLT-2 inhibitors are not a neutral baseline. These drugs affect glucosuria, weight, cardiovascular function, and potentially mood and energy. If SGLT-2 inhibitors have any effect — positive or negative — on brain reward circuits or substance use behavior, the “protective effect of GLP-1” observed here could be partly an artifact of what GLP-1 looks like relative to a comparator that isn’t passive. This hasn’t been studied. The implicit assumption throughout the paper — and certainly throughout the media coverage — is that SGLT-2 drugs are inert with respect to addiction. That assumption has not been demonstrated.
There is also the issue of differential care engagement. Patients prescribed semaglutide or liraglutide may differ from SGLT-2 patients in their overall engagement with healthcare, their relationship with their providers, their willingness to discuss lifestyle changes, and their access to ancillary support. All of these factors could reduce substance use disorder rates independently of the pharmacology. The inverse probability weighting attempts to balance observed covariates, but unobserved differences in patient motivation, care engagement, and social support cannot be adjusted away.
The SUD diagnosis outcome also relies on EHR coding. Veterans who do not disclose substance use to their VA providers — a well-documented phenomenon — will not appear as SUD cases regardless of their actual use patterns. If GLP-1 users have higher care engagement and therefore more substance use identified and coded, the apparent protective effect could be diluted or even reversed in magnitude. This is a structural limitation of any retrospective study using clinical coding for behavioral health outcomes.
Study Critic
Target trial emulation is a legitimate and increasingly well-validated methodology, but it does not eliminate the core limitation of observational data: unmeasured confounding. The study uses inverse probability weighting to balance the two groups on a large set of observed covariates, but residual confounding from unmeasured variables — health behaviors, social supports, motivational states, prescriber practice patterns — remains. In a study of this scale, small systematic differences in unobserved factors can produce the effect sizes seen here.
The unequal group sizes warrant scrutiny: 124,001 GLP-1 initiators versus 400,816 SGLT-2 initiators in Protocol 1. This 3:1 ratio is not inherently a problem statistically, but it reflects the real-world distribution of prescribing and raises questions about whether the two groups were truly comparable on all dimensions, including unmeasured ones, despite weighting. The cannabis-specific NRD of minus 2.25 per 1,000 over three years is a small absolute number, and the clinical significance of individual cannabis SUD diagnoses coded in EHR data — with all the ascertainment variability that implies — is not the same as a validated clinical outcome.
The follow-up of up to three years is adequate for capturing incident SUD diagnoses but may be insufficient to reveal time-varying effects: does the protective signal persist with long-term GLP-1 use, attenuate, or reverse? Adherence and discontinuation data would be important to interpret the cumulative effect curves fully. The paper reports adherence sensitivity analyses with “directionally consistent results,” which is reassuring but not definitive. Finally, the causal language in media coverage (“GLP-1 targets craving itself”) is substantially more confident than what the hazard ratios and study design can support.
Compared to Past Research
This BMJ cohort study sits at the top of a growing body of observational evidence connecting GLP-1 receptor agonists to substance use outcomes. A 2024 Molecular Psychiatry study by Wang and colleagues found that semaglutide was associated with reduced incidence and relapse of cannabis use disorder in a real-world retrospective cohort, using a design more limited in scale than the current BMJ paper but pointing in the same direction. Separately, studies of GLP-1 drugs and alcohol use disorder — several from Danish registry data — reported 18 to 27% risk reductions, a range consistent with this paper’s alcohol finding of 18%.
Prior data on opioid overdose and GLP-1 drugs came primarily from insurance claims databases and showed mixed but generally favorable signals. The current study adds rigorous active comparator methodology and scale to those observations, and is the first to examine all major substance types simultaneously in the same cohort. That simultaneous cross-substance design is novel and is what gives the “targeting craving” hypothesis its current momentum.
What this study adds beyond prior work is not new mechanistic evidence — the biology remains hypothetical — but statistical scale and methodological rigor. It converts a pattern seen in smaller, less controlled studies into a large-scale signal that is difficult to attribute to chance alone, while still leaving causation unestablished. Prior comparison literature was reviewed for this analysis based on citations in the supplied paper; the broader systematic review of GLP-1 addiction literature was not independently conducted for this Lens Card.
Practical Considerations
For patients already prescribed GLP-1 medications for diabetes or obesity who also use cannabis, the most immediate practical question is whether to discuss substance use patterns more explicitly in that clinical relationship. This study doesn’t change what GLP-1 drugs are indicated for, but it does create a natural entry point for conversations that often don’t happen: the overlap between metabolic health and substance use is real, common, and undertreated.
Product variability matters here. The cohort included semaglutide, liraglutide, and dulaglutide — three different GLP-1 agonists with different receptor binding profiles, pharmacokinetic properties, and clinical trajectories. Whether the observed effects are class-level or drug-specific has not been established. Patients on tirzepatide (a dual GLP-1 and GIP agonist) are in yet a different pharmacological category, and no specific data on that drug class and cannabis use disorder is available from this study.
Cost and access remain significant real-world barriers. GLP-1 medications are expensive, and coverage for them varies considerably by insurance plan and indication. A patient seeking GLP-1 drugs specifically to address cannabis use disorder would find no approved indication and would face substantial access hurdles. For patients already prescribed these drugs for metabolic reasons, the potential secondary benefit adds value to an existing treatment plan rather than requiring a new clinical decision. Monitoring cannabis use patterns as part of ongoing metabolic care may be a low-cost practical application of this finding, pending further evidence.
Future Directions
The logical next step is a properly powered randomized controlled trial testing GLP-1 receptor agonists specifically against placebo in people with cannabis use disorder — ideally including participants without diabetes or obesity, to test whether the effect is metabolically mediated or more broadly applicable. The National Institute on Drug Abuse has reportedly been evaluating tirzepatide as a potential treatment for cannabis use disorder in a clinical trial; that kind of prospective, controlled evidence would be far more definitive than the cohort data here.
Mechanistic studies in humans that map GLP-1 receptor activity in mesolimbic reward circuits alongside endocannabinoid system markers would help clarify whether the addiction-relevant effect is truly mediated by craving suppression, or whether other pathways — weight change, improved metabolic function, reduced neuroinflammation — explain part of the signal. The ECS-GLP-1 receptor interaction in these circuits deserves dedicated investigation, particularly given how central CB1 receptor downregulation is to the development and maintenance of cannabis dependence.
Longer-term follow-up beyond three years is needed to understand whether the protective signal persists, plateaus, or changes character with sustained use. Studies in more diverse populations — women, younger adults, people without diabetes, people in cannabis-legal jurisdictions like Massachusetts — would substantially extend the generalizability of findings. Combination studies exploring GLP-1 drugs alongside evidence-based behavioral interventions for cannabis use disorder (motivational interviewing, cognitive behavioral therapy) would be immediately clinically relevant if pharmacological trials confirm efficacy. The field is moving, and this paper accelerates the pace of the question.
Misreadings & Bad-Faith Takes
Several distortions are already circulating, and a few more are predictable.
Distortion: “Ozempic cures weed addiction.” This is false. The study found an association, not a causal effect. It was conducted in veterans with type 2 diabetes, not in patients seeking addiction treatment. No RCT has tested GLP-1 drugs for cannabis use disorder. “Reduced risk of SUD diagnosis in an EHR cohort” is not the same as “clinical cure.”
Distortion: “This proves GLP-1 drugs should be prescribed for addiction.” No addiction indication exists for GLP-1 medications. The study’s authors themselves call for clinical trials before clinical application. Observational findings — however large and well-designed — cannot substitute for controlled evidence of efficacy and safety in a new indication.
Distortion: “Cannabis is so dangerous it needs Ozempic to control it.” This reading inverts the study’s implication. The majority of cannabis users do not develop cannabis use disorder. This study speaks to a subset of heavy users in a specific population. Cannabis’s therapeutic and risk profile is not changed by a study about a separate drug class’s pharmacological effects.
Distortion: “GLP-1 drugs work for all addictions across all people.” The population was older, predominantly male US veterans with diabetes — not representative of most people with substance use disorders. Generalizing a veterans EHR cohort to the broader US addiction population is not supported by this data. Different GLP-1 agents were pooled; substance-specific and drug-specific effects have not been disentangled. The cross-substance signal is suggestive of a shared mechanism, but it does not confirm one.
The honest read is more interesting than any of these distortions: a large, carefully designed observational study has added meaningful statistical weight to a biologically plausible hypothesis. The next step is a randomized trial. That framing doesn’t make for a clean headline, but it’s what the evidence actually supports.
💬 Join the Conversation
Have a question about how this applies to your situation? Ask Dr. Caplan →
Want to discuss this topic with other patients and caregivers? Join the forum discussion →
Have thoughts on this? Share it:
Cai M, Choi T, Xie Y, Al-Aly Z. Glucagon-like peptide-1 receptor agonists and risk of substance use disorders among US veterans with type 2 diabetes: cohort study. BMJ. 2026 Mar 4;392:e086886. doi: 10.1136/bmj-2025-086886. PMID: 41781010; PMCID: PMC12958796.
Frequently Asked Questions
Did this study prove that GLP-1 drugs prevent cannabis use disorder?
No. This was an observational cohort study, not a randomized controlled trial. The study found an association between GLP-1 receptor agonist use and lower rates of cannabis use disorder diagnosis over three years, but association is not causation. The researchers themselves call for clinical trials before applying these findings clinically. Target trial emulation methodology reduces some sources of bias but cannot eliminate confounding from unmeasured variables.
Can I ask my doctor to prescribe Ozempic to treat my cannabis use disorder?
GLP-1 medications have no FDA-approved indication for cannabis use disorder or any addiction. Prescribing them off-label for this purpose would be outside current clinical standards and is not supported by this observational study. If you have metabolic reasons (type 2 diabetes, obesity) that independently justify GLP-1 therapy, discussing the full picture with your physician makes sense. But the evidence base for using these drugs specifically for addiction treatment is not yet at the clinical application threshold.
Why did GLP-1 drugs show reduced risk across so many different substances at once?
The cross-substance consistency is what makes this study biologically interesting. GLP-1 receptors are present in brain reward circuits, including the ventral tegmental area and nucleus accumbens — regions central to dopamine-mediated craving across all addictive substances. The leading hypothesis is that GLP-1 agonists dampen a shared craving signal rather than targeting any substance-specific pathway. This would explain why alcohol, cannabis, cocaine, nicotine, and opioids all show similar signals. However, this mechanism remains a hypothesis; it has not been confirmed in human neuroimaging or pharmacological studies.
What is the connection between GLP-1 receptors and the endocannabinoid system?
Both GLP-1 receptors and CB1 endocannabinoid receptors are expressed in overlapping regions of the mesolimbic dopamine system, which governs reward, motivation, and craving. Emerging preclinical research suggests these systems interact, though the nature and clinical significance of that interaction has not been fully characterized in humans. In cannabis use disorder specifically, CB1 receptor downregulation in reward circuits is a key part of how dependence develops. Whether GLP-1 drugs modulate this process directly is an open and clinically important question.
Who were the people studied, and does that affect how widely the findings apply?
The 606,434 participants were US veterans with type 2 diabetes — a population that is predominantly older, predominantly male, with high rates of comorbid conditions including PTSD, chronic pain, and cardiovascular disease, and with access to VA health care. This population differs substantially from the typical medical cannabis patient, younger adult recreational user, or otherwise healthy person with cannabis use disorder. Generalizing the findings beyond veterans with diabetes requires caution and awaits studies in more representative populations.
If I’m already on a GLP-1 drug, does this study change anything for me?
It doesn’t change your prescription or your treatment plan, but it does add context worth discussing with your physician. If you use cannabis and have been concerned about developing a problematic use pattern, your GLP-1 medication may carry a modest protective signal on top of its metabolic benefits. If you already have cannabis use disorder, the study’s second protocol suggests GLP-1 users experienced fewer serious outcomes. Neither finding is a clinical prescription, but both are relevant to an informed conversation with your care team.
What were the key limitations of this study?
Several limitations are important to understand. First, the comparator was SGLT-2 inhibitors, not placebo — the relative benefit of GLP-1 drugs depends partly on what SGLT-2 drugs do to addiction risk, which is unknown. Second, substance use disorder diagnoses in electronic health records undercount actual substance use due to non-disclosure and variable coding. Third, unmeasured confounders (health behaviors, engagement with care, social support) may explain part of the observed difference. Fourth, the three-year follow-up may not capture long-term effects. Fifth, all participants had type 2 diabetes, limiting generalizability.
How does this study compare to earlier research on semaglutide and cannabis?
A 2024 study in Molecular Psychiatry by Wang and colleagues found that semaglutide specifically was associated with reduced incidence and relapse of cannabis use disorder in a real-world retrospective cohort. That earlier study was smaller and more narrowly focused on cannabis; the current BMJ cohort is larger by orders of magnitude and examines six substance types simultaneously in a more methodologically rigorous framework. The direction of the finding is consistent across both studies, which adds credibility even while causation remains unestablished.
Does this mean cannabis itself becomes less effective or less enjoyable when someone is on a GLP-1 drug?
This study did not examine the acute effects of cannabis in people taking GLP-1 drugs. The study measured cannabis use disorder incidence over time in a population not specifically seeking cannabis effects. Some patients taking GLP-1 drugs anecdotally report reduced reward salience across multiple pleasurable behaviors, which aligns with the reward-dampening hypothesis. Whether this affects therapeutic cannabis efficacy in any meaningful way for medical patients is not something this study can answer, and would require dedicated pharmacological research.
What would need to happen for GLP-1 drugs to become a recognized treatment for cannabis use disorder?
At minimum, one or more adequately powered randomized controlled trials would be needed, testing a GLP-1 receptor agonist against placebo in participants with cannabis use disorder — not limited to people with diabetes. The trial would need to demonstrate reduced cannabis use, reduced craving, and/or improved DSM-5 diagnostic outcomes over a clinically meaningful time horizon, with acceptable safety and tolerability. Regulatory review and FDA approval would follow successful RCT evidence. The National Institute on Drug Abuse has reportedly been evaluating tirzepatide for cannabis use disorder in clinical trials, which could produce relevant data in coming years.
