What UK Biobank Data Suggests About Cannabis, BMI, Diabetes, and Hypertension

Researchers followed 91,002 UK Biobank participants who were free of metabolic disease at baseline and compared incident hypertension, type 2 diabetes, obesity, hyperlipidemia, and non-alcoholic fatty liver disease by cannabis use. Published in Nutrition, Metabolism and Cardiovascular Diseases on April 17, 2026, the analysis found an adjusted hazard ratio of 0.82 (95% CI 0.69 to 0.99) for type 2 diabetes.

The more useful finding is not the headline number. It is that the direction of the association changed with body mass index, which means this is a study about patient selection rather than a study about cannabis.

Patients with rising A1c values, a new hypertension diagnosis, or a weight trajectory they dislike ask directly whether cannabis is helping or hurting. Until recently the honest answer was that the population data were thin and inconsistent. This cohort is large enough to give a partial answer.

The partial answer is uncomfortable for anyone who wants a single sentence. Cannabis use tracked with lower metabolic risk in leaner participants and with higher risk in participants with obesity. That pattern argues against any general statement about cannabis and metabolic health, and in favor of a conversation that starts with the individual patient’s body.

The study, Association between cannabis use and risk of metabolic disease in UK biobank, was conducted by Shaowen Zhou, Juhui He, Xiaolu Chen, Chee Shin Lee, Lixia Zhang, Ying Hu, and Zhaoxia Liang at Women’s Hospital School of Medicine, Zhejiang University, and Taizhou First People’s Hospital. It appeared in Nutrition, Metabolism and Cardiovascular Diseases, volume 36, issue 9, article 104762, published April 17, 2026.

The analytic sample was 91,002 UK Biobank participants with no prevalent metabolic disease at baseline. Multivariable Cox proportional hazards models produced these adjusted hazard ratios for cannabis use: overall metabolic disease 0.94 (95% CI 0.89 to 0.99), hypertension 0.93 (95% CI 0.87 to 1.00), type 2 diabetes 0.82 (95% CI 0.69 to 0.99), and obesity among heavy users 0.57 (95% CI 0.41 to 0.78).

Two of those intervals deserve attention. The hypertension interval reaches 1.00, so that result does not exclude no effect. The type 2 diabetes interval stops at 0.99, which clears the threshold but leaves very little margin. Neither hyperlipidemia nor non-alcoholic fatty liver disease showed a significant association in either direction.

The authors tested whether body mass index modified the association and found significant interactions at P less than 0.01. Below a BMI of 25, the inverse associations were most evident. Above a BMI of 30, the pattern inverted: moderate cannabis use was associated with an adjusted hazard ratio of 1.26 for overall metabolic disease and 1.40 for hypertension.

An interaction of that size is not a footnote. It means the pooled hazard ratios in the abstract are averages across two populations that behaved differently, and that quoting the pooled number to a patient with obesity misrepresents what the data say about that patient.

It is worth being precise about the subgroup labels. The elevated risk above a BMI of 30 was reported for moderate use, and the 43 percent lower obesity risk was reported for heavy users. Those are different exposure categories, and collapsing them into a single statement about cannabis users is one of the ways this paper has been misreported.

There is a coherent endocannabinoid account of why body weight might change the direction of this association. CB1 receptors are expressed in adipose tissue, liver, skeletal muscle, and hypothalamus. In obesity, peripheral endocannabinoid tone is characteristically elevated. Adding an exogenous CB1 agonist to a system already running high is not pharmacologically equivalent to adding one to a system with normal tone.

That story fits the data. Fitting the data is not the same as explaining them. This cohort measured self-reported cannabis exposure and recorded incident diagnoses. It did not measure endocannabinoid tone, receptor density, cannabinoid dose, or product composition, so the mechanism remains a hypothesis that this study is consistent with rather than one it tests.

Readers who want mechanistic human evidence should look to controlled physiology rather than cohort epidemiology. Cohort data can tell you where to point a trial. It cannot substitute for one.

The exposure here is self-reported cannabis use recorded in a UK cohort recruited between 2006 and 2010, during a period when cannabis was illegal in the United Kingdom. Under-reporting in that setting is expected, and it is not random: people who under-report may differ systematically from people who do not.

The dataset also carries no product-level detail. There is no record of THC to CBD ratio, route of administration, dose, or frequency beyond broad use categories. A daily oral CBD-dominant preparation and occasional high-THC inhalation are the same exposure in this analysis, which is a substantial limitation for any clinical translation.

Finally, the UK Biobank is not representative of the UK population. Participants are healthier, wealthier, and less deprived than the general population, which limits how far these hazard ratios travel.

The practical output of this paper is a question, not a recommendation. The question is whether a given patient sits in the part of the distribution where the association was favorable or the part where it was not, and the first input to that question is body mass index.

For a patient with a BMI under 25 who uses cannabis and is worried that it is damaging their metabolic health, this cohort is reassuring at the population level. It is not permission, and it is not evidence of benefit, but it argues against reflexive alarm.

For a patient with a BMI over 30, the honest statement is that the population signal in this dataset ran the other way, that the reason is unknown, and that appetite effects, product choice, and sleep disruption are all plausible contributors worth examining individually. That is a more useful conversation than a verdict.

This is a well-powered observational cohort, which places it above cross-sectional surveys and below randomized trials. The UK Biobank supports long follow-up and adjusted modeling, and 91,002 participants is enough to detect the modest effect sizes reported here. For the question of whether a population-level association exists, the design is adequate.

For the question of whether cannabis changes metabolic risk, the design cannot deliver. Cannabis users in this cohort differ from non-users in ways no model fully captures, including diet, activity, alcohol, sleep, and health-seeking behavior. The authors themselves conclude that the findings should be interpreted cautiously given the observational design.

The hypertension confidence interval reaches 1.00. A result whose interval includes the null should not be described as a reduction in hypertension risk, and several summaries of this paper have done exactly that. The type 2 diabetes interval stops at 0.99, which is statistically significant by convention but fragile enough that a modest change in modeling choices could move it.

Self-reported cannabis use collected in a jurisdiction where cannabis was illegal invites differential misclassification. The heavy-use obesity result, an adjusted hazard ratio of 0.57, is the largest effect in the paper and also the one most exposed to reverse causation, since body weight influences who becomes and stays a heavy user as readily as the reverse.

This cohort does not show that cannabis prevents type 2 diabetes, lowers blood pressure, or treats metabolic disease. It reports associations between a self-reported exposure and incident diagnoses in a non-representative volunteer cohort.

It also identifies no product, dose, ratio, route, or timing. Nothing in this dataset supports recommending a cannabis preparation for metabolic indications, and nothing in it establishes which patients would benefit if such an effect existed.

The result sits alongside a consistent epidemiological pattern in which cannabis users show lower average body mass index and lower diabetes prevalence than non-users, despite reporting higher caloric intake. That pattern has been reported often enough to be taken seriously as an observation and has never been explained well enough to be taken seriously as a mechanism.

What this cohort adds is the effect modification by body weight. Previous work treated cannabis users as one group. Splitting them by BMI produced opposite signs, which suggests the field has been averaging across populations that do not belong in the same estimate.

I read this paper twice, and the second reading was more useful than the first. The headline everyone repeated was that cannabis lowers diabetes and hypertension risk. The hypertension interval touches 1.00. That is not a finding, and repeating it as one does patients a disservice.

What I do take from it is the BMI interaction, because it matches something I see. In leaner patients, cannabis rarely disturbs metabolic numbers in either direction. In patients carrying significant weight, the variable that most often matters is not the cannabinoid at all. It is what happens two hours after a high-THC evening dose, and whether the patient is eating through it. That is a modifiable problem, and it is the conversation this paper should be starting.

I would not change a single patient’s plan on the basis of a hazard ratio of 0.82 from a volunteer cohort. I would use this to justify asking better questions at the visit.

In 91,002 UK Biobank adults, cannabis use was associated with modestly lower incident type 2 diabetes and overall metabolic disease, with the association concentrated below a BMI of 25 and reversed above a BMI of 30. Treat the direction as hypothesis-generating, treat body mass index as the variable that changes the conversation, and do not present any of it to a patient as evidence of benefit.

Carry forward the interaction, not the pooled estimate. The useful sentence is that the relationship between cannabis use and metabolic risk in this cohort depended on body weight. The sentence to leave behind is that cannabis lowers hypertension risk, because the confidence interval for that outcome does not support the claim.