Among people with type 2 diabetes who started a GLP-1 drug, the ones who also used cannabis died more often. Over a median of about two and a half years, cannabis users were 64 percent more likely to die from any cause than matched patients who did not use it, and the risk climbed the heavier the use.
That comes from a propensity-matched cohort study ↗ published July 29 in Drug and Alcohol Dependence, led by Aseel Salameh and colleagues at the Tri-Service General Hospital and National Defense Medical University in Taipei, with co-authors at Beth Israel Deaconess Medical Center in Boston. They used TriNetX, a network of de-identified electronic health records, to pull adults with type 2 diabetes who started a GLP-1 receptor agonist ↗ between 2017 and 2025.
GLP-1 receptor agonists, the class that includes semaglutide ↗ (Ozempic, Wegovy) and tirzepatide ↗ (Mounjaro, Zepbound), lower blood sugar and, in large trials, cut heart and kidney complications. The question here was whether cannabis changes that payoff. Out of 614,333 eligible patients, the authors found 4,117 with a documented cannabis-use diagnosis and matched each to a non-user on more than 40 baseline characteristics, then tracked who died, who had a major heart event (heart attack, stroke, and the like), and who had a major kidney event.
Every outcome was worse in the cannabis group. All-cause mortality carried a hazard ratio of 1.64 (95 percent confidence interval 1.38 to 1.96), meaning about a 64 percent higher death rate. Major heart events ran 42 percent higher, kidney events 55 percent higher. And the pattern scaled with severity. Patients coded with cannabis use disorder had more than double the mortality risk (hazard ratio 2.20), against 1.73 for use without a disorder diagnosis. A dose-response like that is one of the things that makes an association harder to dismiss.
Harder, not impossible. This is a records-database study, and its currency is association, not cause. A cannabis diagnosis in a health record flags the people whose use was heavy or troubled enough for a clinician to write it down, and those patients differ from non-users in ways no matching table fully captures: whether they take the GLP-1 as prescribed, tobacco, other substances, income, how often they show up for care. The authors report E-values of 2.21 to 2.63, their own estimate of how strong a hidden confounder would have to be to erase the finding. That is a moderate bar, and the lifestyle cluster that travels with heavy cannabis use in diabetes could plausibly clear it. Nothing here shows that cannabis chemically blunts a GLP-1 drug. It shows that, among people on one, cannabis use marks a group that does worse.
That distinction matters, because a separate line of research points the other way on a different question. Several TriNetX and veteran-cohort analyses have suggested GLP-1 drugs may lower the risk of developing substance-use problems in the first place, cannabis use disorder among them. This study does not contradict that. It looks at outcomes among people who already have both, and finds the combination tracks with higher risk.
One exploratory thread is worth flagging with the same caution. In the cannabis group, patients also taking an SGLT-2 inhibitor (a separate class of diabetes pill, not a peptide, so no card on this platform) had lower mortality (hazard ratio 0.62) and fewer kidney events. That is a hypothesis thrown up by the same confounded data, not a treatment recommendation.
The practical takeaway the authors land on is small and reasonable. Ask. A prescriber starting a patient on a GLP-1 drug rarely asks about cannabis, and this is a reason to. Whether the drug works differently in cannabis users, or whether cannabis users are simply a higher-risk group who need closer follow-up, the records cannot separate. The next study will have to be built to.