GLP-1 and Addiction: What the March 2026 BMJ Study Actually Shows

The Ozempic-quiets-cravings story finally has a real paper behind it

On March 4, 2026, BMJ published the first large-scale, clinical-grade look at something GLP-1 users had been whispering about for three years. The shot makes more than food go quiet. Wine goes quiet. Cigarettes go quiet. Pills go quiet. The whole reward system dims.

The study comes out of Washington University in St. Louis (WashU Medicine) and pulls more than 600,000 US veterans with type 2 diabetes from Veterans Health Administration records. Retrospective and observational — not a randomized trial — but the scale is the draw. Nobody is about to run a 600,000-person RCT on addiction outcomes.

Two numbers carried most of the coverage: opioid use disorder fell 25% in the GLP-1 group versus the SGLT2 group, and drug-related deaths among patients who already had a substance use disorder dropped 50% over three years of follow-up. Sit with those. They are also the numbers most likely to get misread.

What the study actually measured

Design outranks headline here. WashU's team took US veterans with type 2 diabetes and split them into two exposure arms: patients starting a GLP-1 receptor agonist (mostly semaglutide, liraglutide, or dulaglutide) versus patients starting an SGLT2 inhibitor. The SGLT2 arm is the clever choice. Both classes are modern, cardiovascular-friendly diabetes drugs prescribed to similar populations, which defuses the "GLP-1 users are just healthier" confounder more than a placebo comparison ever could.

Follow-up ran up to 3 years. Two parallel analyses on the same cohort:

• Analysis A — patients without a pre-existing substance use disorder, tracking whether they developed one.
• Analysis B — patients with a pre-existing substance use disorder, tracking emergency visits, hospitalizations, overdoses, and deaths.

One analysis asks: does this drug keep addiction from starting? The other asks: does it reduce harm in people already fighting it? Different questions. Different answers. Both matter.

One catch up front, worth flagging before the numbers land: the cohort is >90% male, older, and all diabetic. Veterans are not a random slice of America. Hold that while you read the tables.

The new-onset numbers: across the board, not just alcohol

Analysis A tracked patients who started with a clean substance-use history. Lower hazard ratios mean the GLP-1 group developed the condition less often than the SGLT2 group over the follow-up window.

The thing that jumps out: this isn't substance-specific. Alcohol, nicotine, cannabis, cocaine, opioids — all down, and by numbers that do not look like noise. If GLP-1s only quieted one pathway, you'd expect one column to move while the rest sat flat. Every column moved.

That fits what preclinical researchers had been watching for a decade. Rats on semaglutide cut their alcohol self-administration. Mice reduced nicotine intake. The animal work pointed at a broad reward-circuit effect, not an alcohol-only one. The BMJ numbers are the first human-scale signal that rhymes with that story.

The harm-reduction numbers hit harder

Analysis B shifted the tone of the coverage. These are patients who already carried a substance use disorder diagnosis at baseline. Over 3 years of follow-up:

A 50% reduction in drug-related deaths is the kind of number that, if a randomized trial backs it up, rewrites the treatment landscape. The US is still losing more than 100,000 people a year to opioid overdose. Naloxone access expanded from 2023–2026 and saved lives, but the curve has not bent the way public health officials hoped.

The harm-reduction signal is larger than the prevention signal. That's unusual, and that's the part worth watching.

One caveat that gets lost: patients who keep showing up for diabetes appointments and refilling a GLP-1 are, by definition, engaged with healthcare. Engagement itself is protective. The authors adjusted for it. Adjustment is not the same thing as randomization.

Why this might work at all

The cleanest mechanistic story lives in two brain regions most people never learn the names of: the nucleus accumbens and the ventral tegmental area. These are the reward-processing hubs — the parts that light up when you eat, drink, smoke, or scroll. GLP-1 receptors sit in both.

When semaglutide or liraglutide crosses into the brain, it appears to blunt dopaminergic reward signaling broadly rather than substance by substance. The whole reward-prediction circuit powers down. That's almost certainly the same neurobiology GLP-1 patients describe as "food noise" going silent — the endless, pre-meal, is-there-chocolate-in-the-cupboard chatter that switches off around week 3 or 4.

The BMJ authors and several addiction researchers have floated a blunt reframe: if the mechanism quiets food noise, it probably quiets drug noise too, because the brain doesn't file craving by category. A craving is a craving. The circuitry behind it doesn't distinguish ice cream from bourbon from nicotine from oxycodone. It only knows want.

That's a hypothesis, not a conclusion. But now it carries a 600,000-person observational signal on top of a decade of rat and mouse work. That's why clinicians raised their eyebrows when the paper dropped.

Why this isn't a prescription yet

This is exactly the kind of finding that gets flattened into "Ozempic cures addiction" by the time it reaches TikTok, so the honest reading has to go here.

1. Observational, not randomized. The study shows association, not causation. The GLP-1 and SGLT2 groups matched well on paper, but there is always a chance that the type of patient who gets and stays on a GLP-1 differs from the type who gets and stays on an SGLT2 in ways administrative records never see.

2. The population is narrow. US veterans with type 2 diabetes skew >90% male, skew older, and all carry a diabetes diagnosis. The people most affected by alcohol use disorder and opioid use disorder in the general population are often younger, often female, often non-diabetic. You cannot assume the 25% opioid-risk reduction transfers cleanly to a 28-year-old woman with alcohol use disorder and no metabolic disease.

3. Unmeasured confounders. Income, housing stability, social support, severity of diabetes, severity of mental-health comorbidity — the records don't capture all of it. Propensity-score matching is a real statistical tool. It cannot pretend to be randomization.

4. Nowhere is it approved for addiction. Not the FDA. Not the EMA. Not Japan's PMDA, Korea's MFDS, China's NMPA, Saudi Arabia's SFDA, the UAE's MOHAP, Taiwan's TFDA, or the Hong Kong Drug Office. Every on-label use you can name is for type 2 diabetes, obesity, or cardiovascular risk reduction. Addiction is off-label everywhere.

5. No guideline supports self-medicating. The CDC has not recommended it for opioid use disorder. ASAM (the American Society of Addiction Medicine) has not recommended it for alcohol use disorder. Ordering compounded semaglutide from a telehealth site because you want to drink less is not an evidence-based treatment plan.

6. Prospective RCTs haven't read out. WashU is running a randomized trial of semaglutide for alcohol use disorder with NIAAA support. Tobacco cessation trials are underway. Opioid trials are in planning. None of them have published. Until they do, the BMJ paper is the strongest human evidence we have — and it is still a retrospective cohort.

Observational evidence this strong usually means a randomized trial is worth running. It doesn't mean the randomized trial will agree.

If you're already on a GLP-1, here's how to read this news

The most common question flooding clinician inboxes the week the paper dropped went roughly: "I'm on Wegovy. I've also been drinking less since I started. Is that the drug?"

Maybe. Partly. Also partly because most people on a GLP-1 are losing weight, sleeping better, eating out less, and living in a different relationship with their body. Isolating one variable in that stack is hard.

A few practical framings:

• If you've been on a GLP-1 for 3–6 months and your alcohol intake slid down on its own — that's consistent with the mechanism. It's real. Take it.
• If you have a history of alcohol use disorder or opioid use disorder and you're starting a GLP-1 for obesity or diabetes, tell your prescriber. Not because they will switch you to a different drug — because the context matters to any clinician treating you.
• If you're coming off a GLP-1 for any reason — coverage change, side effects, switching to something else — the drug noise may return the way food noise does. Worth expecting rather than being blindsided by.
• Do not stop naltrexone, acamprosate, buprenorphine, methadone, or any other prescribed addiction medication because your GLP-1 "seems to be working." That is not what the BMJ paper says, and walking away from gold-standard care is how relapses happen.

Questions worth bringing to your doctor

If you're in the US and this news made you want a real conversation — with your PCP, your endocrinologist, or your addiction medicine clinician — these are the prompts that move the discussion forward:

• I've noticed X since starting the GLP-1 (less drinking, fewer cigarette cravings, whatever it is). Is that worth tracking?
• I have a personal or family history of substance use disorder. Does that change how you're thinking about this prescription?
• I'm already on naltrexone, buprenorphine, or bupropion — any interaction risk with adding or staying on semaglutide or tirzepatide?
• If I come off the GLP-1, what should I watch for behaviorally, not just weight-wise?
• Is there a GLP-1-plus-behavioral-program you'd recommend — or that a research center in our region is running?

Notice what's not on the list: "Can you prescribe Ozempic for my drinking?" No reasonable clinician will say yes to that in 2026, and it's not the right frame anyway.

What this paper does not replace

Secondary coverage tends to skip this part, so it goes here in bold: evidence-based addiction treatment remains first-line, without exception.

For alcohol use disorder:

• Naltrexone (oral or Vivitrol monthly injection)
• Acamprosate
• Cognitive behavioral therapy (CBT) and motivational interviewing (MI)
• Mutual-help groups — AA, SMART Recovery, Refuge Recovery

For opioid use disorder:

• Buprenorphine (Suboxone, Sublocade)
• Methadone (through OTPs — opioid treatment programs)
• Naltrexone (Vivitrol)
• Harm reduction — naloxone at home, fentanyl test strips, syringe services where available

For nicotine dependence:

• Varenicline (Chantix)
• Bupropion (Zyban / Wellbutrin SR off-label)
• Nicotine replacement — patch, gum, lozenge, inhaler
• Quit counseling, text-based programs (SmokefreeTXT), 1-800-QUIT-NOW

None of this changes because of the BMJ paper. If anything, the right read is that these treatments might one day be augmented by a GLP-1, not replaced by one. A patient on buprenorphine plus semaglutide, with a prescriber coordinating both, is not the same clinical picture as a patient self-treating opioid cravings with compounded semaglutide from a telehealth clinic.

How this reads for readers in the US right now

US context is what makes this paper land harder here than almost anywhere else. A few framing points:

• Opioid overdose deaths remain above 100,000 per year. The 2023–2026 expansion of naloxone access — over-the-counter Narcan in pharmacies from 2023, standing orders across all 50 states — saved lives, but the curve didn't bend the way policymakers hoped. Anything that drops overdose risk by 40% in a follow-up study will get institutional attention fast.
• The US is where GLP-1 access is already widest. Wegovy for obesity, Ozempic for type 2 diabetes, Rybelsus and the 25 mg oral Wegovy that launched in 2026 for patients who don't want the shot, Zepbound for obesity plus obstructive sleep apnea, Mounjaro for diabetes, Saxenda for liraglutide-based obesity care. If any country is positioned to accidentally run a real-world GLP-1-and-addiction experiment, it's this one.
• Coverage is still the choke point. Commercial insurance covers GLP-1s for obesity unevenly. Medicare Part D covers Wegovy for cardiovascular risk reduction (post-2024 SELECT label) but not for weight loss alone. Medicaid varies wildly state by state. Whether the BMJ signal ever becomes "my plan covers Ozempic because I have alcohol use disorder" is a long, political road — not a 2026 road.
• The CDC has issued no guidance on GLP-1s for substance use disorder. ASAM has not revised its practice guidelines. The FDA has not opened an addiction indication pathway for semaglutide. As of April 2026, the clinical establishment is in "interesting signal, let's run the trial" mode, not "write the prescription" mode.

The research-to-prescription pipeline for anything involving addiction in the US is slow by design. It has to survive RCTs, label updates, payer review, and DEA scrutiny of adjacent questions. Realistic timelines for any GLP-1 to carry an addiction indication are years out — if any ever gets there. None of that is a reason to dismiss the BMJ paper. It's a reason to read it for what it is: the first large human-scale evidence that a hypothesis clinicians were half-joking about in 2023 might hold.

The last thing worth saying

Most drug stories in 2026 run either too hyped or too cynical. This one lives somewhere in between. The signal is real. The mechanism is plausible. The limitations are real too. A retrospective cohort of diabetic US veterans is not a randomized trial of the people most affected by the conditions under discussion.

If you're on a GLP-1 and your drinking has quietly slid down, you're not imagining it. If you or someone you love is fighting an addiction, the best available care is still the care that has been proven for decades. The BMJ paper hints that those two sentences may be more connected than anyone thought. It doesn't yet tell us how to prescribe around that.

The honest framing in April 2026: a drug class we already prescribe to tens of millions of people may do something meaningful we didn't design it to do. Figuring out what to do with that — carefully, in trials, with the people who need it most — is the next few years of work.

Bring the paper to your next appointment if it changes how you think about your own treatment. That is the conversation worth having. Everything else can wait for the randomized data.

Sources

• BMJ, March 4, 2026 — GLP-1 receptor agonists and substance use disorders in US veterans
• WashU Medicine — press release on the cohort study
• NIAAA — semaglutide for alcohol use disorder, ongoing trial
• CDC — opioid overdose data, 2023–2026
• ASAM — current practice guidelines for alcohol and opioid use disorders

Start managing your GLP-1 with Blueshot

AI coaching, injection scheduling, and weight tracking in one app