Ozempic for Addiction: What the VA's 600,000-Person Study Actually Found

The headline writers ran with the obvious frame: “Ozempic could cure addiction.” The study, published in The BMJ and released June 3, does not say that. What it says is more carefully qualified — and still significant enough that every clinician working in addiction medicine should read the actual paper, not the press release.

The study is observational, which matters enormously for how you interpret it. Researchers at the Veterans Affairs healthcare system analyzed a cohort of more than 600,000 veterans who had been prescribed either a GLP-1 receptor agonist (semaglutide, liraglutide, or dulaglutide) or a comparator diabetes medication between 2020 and 2024. They then tracked both groups for rates of new substance use disorder diagnoses, overdose events, hospitalizations, and death over the follow-up period.

The results across six substance categories — alcohol, opioids, cocaine, cannabis, nicotine, and stimulants — were consistent, large, and in the same direction.

Veterans taking GLP-1 medications were 14% less likely than those on comparator drugs to develop a new substance use disorder. Among veterans who already had a substance use disorder at baseline, the GLP-1 group experienced 39% fewer overdose events and 50% fewer substance-related deaths compared to the comparator group.

Fifty percent fewer deaths. Across every major addiction category. In the world’s largest healthcare system. That is not a subtle signal in a noisy dataset.

Why GLP-1 drugs might affect addiction

GLP-1 receptors — glucagon-like peptide-1 receptors — are present throughout the brain’s reward circuitry, including the nucleus accumbens and the ventral tegmental area. These are the same structures central to the neuroscience of addiction: the nucleus accumbens is where dopamine floods when a person uses cocaine, takes an opioid, or drinks alcohol; the VTA is where that dopamine is produced. When GLP-1 agonists bind to receptors in these regions, they appear to modulate the dopaminergic response to rewarding stimuli — reducing the intensity of the signal, not eliminating it.

The mechanism is functionally similar to what GLP-1 drugs do for food intake: they reduce the hedonic reward of eating without creating anhedonia. If the same thing is happening with drugs of abuse, GLP-1 medications may be reducing the intensity of the craving signal and the reward itself — taking the edge off in a way that changes the cost-benefit calculation the addicted brain runs continuously.

Preclinical research supports this. Animal studies have shown semaglutide reduces alcohol consumption and cocaine self-administration. A Phase 2 human trial of semaglutide for alcohol use disorder, published in Addiction in 2025, found significant reductions in both craving and drinks per drinking day compared to placebo. Keith Humphreys of Stanford, commenting on the BMJ study in Science, called GLP-1 drugs “the most credible pharmacological advance in addiction medicine since naltrexone.” That is a considered statement from someone who has evaluated a lot of overhyped claims in this field.

What the study does not prove

The observational design limits causal inference in ways that matter. People prescribed GLP-1 medications versus comparator diabetes drugs are not equivalent populations. GLP-1 drugs are newer, more expensive, and more likely to be prescribed to patients who are better clinically managed — patients with more frequent healthcare contact, better insurance, stronger social support networks, and potentially lower baseline substance use risk.

The researchers applied propensity score matching to control for observed confounders (age, comorbidities, baseline substance use, prescribing patterns). Unobserved confounders — differences between groups that weren’t measured — cannot be controlled for. The 14% reduction in new SUD diagnoses is where this concern is sharpest; the 39% reduction in overdoses among existing SUD patients is more credible, partly because the baseline populations are more similar once you condition on having an existing diagnosis.

This is not a reason to dismiss the study. It is a reason to treat it as a strong hypothesis generator and wait for the randomized controlled trials.

It is a reason to treat it as a strong hypothesis generator and wait for the randomized controlled trials.

What the RCTs will tell us

Several trials are underway. A VA-funded RCT of semaglutide for alcohol use disorder is enrolling. NIH HEAL Initiative has funded a trial for opioid use disorder. A tobacco cessation trial using a GLP-1 agonist is in Phase 2. These results are expected between 2027 and 2029.

If those trials reproduce the magnitude of benefit seen in the VA observational study — which is a very large “if,” since RCTs frequently show attenuated effects compared to observational studies — GLP-1 drugs would represent the most significant advance in addiction pharmacotherapy in a generation. If the trials show smaller effects, the clinical signal is still worth characterizing precisely.

The access problem no one is talking about

There is a conversation the field hasn’t quite started yet, and it is the one that will matter most if the RCT data holds. GLP-1 medications cost approximately $900–$1,300 per month without insurance. Medicare Part D covers semaglutide for diabetes but not for obesity or SUD; Medicaid coverage varies widely by state. Arizona’s AHCCCS covers GLP-1 drugs for diabetes.

If GLP-1 drugs are eventually approved for SUD, the coverage question will immediately become the central question — as it always does in addiction medicine, where the most effective treatments (buprenorphine, methadone) have historically been the hardest for the people who need them most to access. The VA study was conducted in a fully integrated healthcare system where cost is not a barrier to prescription. The community population most at risk for overdose does not live in that system.

Getting ahead of this access question — in parallel with the RCT data, not after it — is the work that needs to happen now.