Do GLP-1s Reduce Drinking? 10 Studies on the Addiction Question

The first randomized trial of semaglutide in alcohol use disorder: 48 adults, not seeking treatment, on a low dose for nine weeks. It cut how much people drank in a monitored laboratory session, how much they drank on days they drank, and how much they craved alcohol — and reduced cigarettes among the smokers in the sample. ★ Note what did not move: the number of drinking days and drinks per calendar day were unchanged. Forty-eight people, nine weeks, below the weight-loss dose. It is the strongest single result in the field, which says as much about the field as about the trial.

PMID 39937469 ↗DOI 10.1001/jamapsychiatry.2024.4789 ↗

The larger and earlier randomized trial, and formally a negative one: 127 adults with alcohol use disorder on weekly exenatide or placebo for 26 weeks alongside therapy. The main endpoint, heavy-drinking days, did not improve. ★ A pre-specified subgroup with BMI above 30 did — substantially — and brain imaging in that group showed reduced reactivity to alcohol cues. ⚠ Subgroup findings from a failed trial are the classic way to be misled. But the specific subgroup that responded is the one the whole obesity-drug hypothesis predicts, and it shaped every trial that followed.

PMID 36066977 ↗DOI 10.1172/jci.insight.159863 ↗

Cannabis use disorder, across health-record cohorts of people with obesity or diabetes, showing 32 to 56% lower rates of new diagnoses on semaglutide and similar reductions in recurrence. ⚠ Two problems compound here: the usual confounding, plus the fact that cannabis use is captured in medical records only when someone chose to record it. What is being measured is diagnosis, not use — and those diverge in exactly the population most likely to be prescribed a weight drug.

PMID 38486046 ↗DOI 10.1038/s41380-024-02498-5 ↗

Tobacco use disorder, using a design that simulates a randomized trial inside the records and compares against five different drug classes rather than one. Semaglutide was associated with fewer tobacco-related medical visits and fewer cessation prescriptions, with the largest gaps against insulin. ★ Consistency across every comparator is a genuine strength. The unfixable gap is motivation: wanting to quit is invisible in a claims database, and people who pursue a weight drug may differ on precisely that.

PMID 39074369 ↗DOI 10.7326/M23-2718 ↗

The most serious endpoint in the series — actual opioid overdose, in patients with diabetes and existing opioid use disorder — where semaglutide was associated with 42 to 68% lower risk across seven comparators. ⛔ The confounding here is severe and obvious: people who stay engaged with medical care both fill prescriptions and overdose less. The magnitude and consistency are what have driven calls for a randomized trial, and no such trial has reported.

PMID 39320894 ↗DOI 10.1001/jamanetworkopen.2024.35247 ↗

The most methodologically interesting entry: Finnish national registries covering 227,866 people with alcohol use disorder, analyzed so that each person serves as their own control across periods on and off a drug. Alcohol-related hospitalization fell 36% during semaglutide use and 28% during liraglutide use — against 14% for naltrexone, the established treatment. ★ Using people as their own controls removes everything stable about them, which is most of what wrecks the studies above. ⚠ It cannot fix reverse causation: people may start a weight drug during a period when they are already drinking less.

PMID 39535805 ↗DOI 10.1001/jamapsychiatry.2024.3599 ↗

A survey of 153 adults who had started semaglutide or tirzepatide for weight loss, about two-thirds of whom reported drinking less, with fewer binge episodes than a comparison group. ⛔ Self-reported, retrospective, unrandomized, and alcohol consumption is the classic thing people under-report. It earns its place for being early and for covering tirzepatide, and it should carry the least weight on this page.

PMID 38017205 ↗DOI 10.1038/s41598-023-47934-8 ↗

A claims analysis that does the one thing most of this literature does not: compares against DPP-4 inhibitors, a drug class prescribed to nearly identical patients for the same condition but with no action on brain reward pathways. Alcohol-related healthcare use was lower on the GLP-1 drugs. ★ That comparator choice matters more than the effect size — it strips out most of the healthy-user bias that makes comparisons against insulin or metformin unreliable.

PMID 40048376 ↗DOI 10.1172/JCI188314 ↗

A systematic review of the randomized evidence through mid-2024, which turns out to be the exenatide trial above plus a handful of small studies in smoking and cocaine use. Its conclusion is that the signals are heterogeneous, driven by subgroups — obesity in particular — and insufficient to support prescribing. ★ Its value is as an inventory: it is the cleanest accounting of which randomized trials have actually finished, which is a much shorter list than the volume of coverage implies.

PMID 39288591 ↗DOI 10.1016/j.drugalcdep.2024.112424 ↗

The best framing of the whole field, written by the NIH group that produced much of the underlying science and by the lead authors of two papers on this page. It lays out why the preclinical and observational case is compelling, why the randomized evidence does not yet support use, and what trials would have to show to change practice. ★ If a patient asks whether they should take a GLP-1 to stop drinking, this is the paper that answers them — and its answer is not yet.

PMID 38001271 ↗DOI 10.1038/s41591-023-02634-8 ↗