Semaglutide and Alcohol Use Disorder: A Phase 2 RCT

The 48-participant randomised trial showed that nine weeks of once-weekly semaglutide reduced how much alcohol non-treatment-seeking adults with alcohol use disorder drank in a laboratory self-administration session, relative to placebo, with a medium-to-large standardised effect on grams consumed (β −0.48; 95% CI −0.85 to −0.11; P = .01) and on peak breath alcohol concentration (β −0.46; 95% CI −0.87 to −0.06; P = .03) 1. It also showed that average drinks per calendar day and the number of drinking days did not change. Weekly craving and drinks per drinking day fell. That combination, less alcohol per occasion and less reported craving, with no change in how often participants drank, is the finding. It is a mechanistic signal from a phase 2 trial, not evidence of clinical efficacy, and no regulator has approved the drug for this use.

This site's other centrally acting peptide with behavioural endpoints is bremelanotide, which acts at melanocortin receptors. This article covers a different receptor system and a different kind of endpoint. Its subject is endpoint validity: why a laboratory paradigm is a mechanistic readout rather than a clinical result, and what the tiers of evidence beneath this trial can and cannot carry. Each section states its tier.

The mesolimbic rationale

Evidence tier: rodent models and in vitro electrophysiology. GLP-1 receptors are expressed in brain regions that process reward, and the hypothesis is that receptor activation dampens the dopaminergic response to alcohol in the nucleus accumbens. In rats given intermittent access to alcohol, acute and repeated semaglutide reduced intake and prevented relapse-like drinking after abstinence in both sexes. Fluorescently labelled semaglutide was detected in the nucleus accumbens of alcohol-drinking rats. In mice, semaglutide attenuated alcohol-induced locomotor stimulation and the rise in accumbal dopamine, and raised dopamine metabolite levels when alcohol was present 4.

A second study tested semaglutide in a binge-like drinking-in-the-dark procedure in mice and in binge-like and dependence-induced drinking in rats. It reduced drinking in every model, and also reduced intake of other caloric and non-caloric solutions, which is a reminder that the effect is not alcohol-specific. In brain slices from alcohol-naive rats it increased the frequency of spontaneous inhibitory postsynaptic currents in the central amygdala and infralimbic cortex, consistent with enhanced GABA release; in alcohol-dependent rats it had no overall effect on GABA transmission 3. The species gap is the obvious limit. These are rodents, drinking under laboratory schedules, at exposures chosen by the experimenter.

Observational signals that preceded the trial

Evidence tier: retrospective cohorts. In a retrospective study of electronic health records covering 83,825 patients with obesity, semaglutide compared with other anti-obesity medications was associated with a 50% to 56% lower risk of both incident and recurrent alcohol use disorder over 12 months; the pattern was replicated in 598,803 patients with type 2 diabetes 5. In a Swedish nationwide register of 227,866 people with alcohol use disorder followed for a median of 8.8 years, periods of semaglutide use were associated with a lower risk of hospitalisation for the disorder than periods of non-use in the same individual, with an adjusted hazard ratio of 0.64 (95% CI 0.50 to 0.83) 6.

Both studies carry the limits of their design. People prescribed semaglutide differ from people who are not, and a within-individual design controls for stable traits but not for the circumstances that lead a person to start or stop a drug. Both sets of authors said randomised trials were needed 56. The 48-person trial is the first to appear.

Trial design: non-treatment-seeking participants, nine weeks, blinded

Evidence tier: human randomised trial, phase 2. The trial was double-blind, randomised and parallel-arm, with nine weeks of outpatient treatment at a single US academic medical centre, enrolling from September 2022 to February 2024. It was registered in August 2022 as a human laboratory screening study, with grams of alcohol consumed and peak breath alcohol concentration during a self-administration procedure as the two primary outcome measures 2. Of 504 people assessed, 48 were randomised, 24 to each arm. Thirty-four (71%) were female; mean age was 39.9 years 1.

Participants were not seeking treatment. They met past-year criteria for alcohol use disorder, reported past-month drinking above 7 standard drinks per week for women or 14 for men, and had at least two heavy drinking episodes in the month. At baseline the sample averaged 2.9 drinks per calendar day, 4.2 drinks per drinking day, 20.2 drinking days and 9.1 heavy drinking days in the preceding 28 days; the authors characterise the severity as moderate 1. The semaglutide arm escalated through three steps at weekly clinic visits, spending four weeks at the first, four at the second and one at the third, with the final step contingent on tolerability. The post-treatment laboratory session took place at the second step, so the primary outcome reflects an exposure below those used in the drug's approved indications 12.

The laboratory self-administration endpoint and what it measures

The paradigm has two phases. In the first, participants could choose to delay the start of drinking for up to 50 minutes in exchange for monetary reinforcement. In the second, they had access to their preferred alcoholic beverage for 120 minutes and drank at their own pace toward their preferred effect. The quantity available was set from anthropometric formulas against a prespecified ceiling on breath alcohol, which was measured every 30 minutes after drinking began. The two co-primary measures were estimated grams of ethanol consumed and peak breath alcohol concentration. One session ran before the first dose and one between weeks 8 and 9 1.

What the paradigm measures is consumption under controlled conditions when alcohol is freely available and an alternative reinforcer has just been offered. It is sensitive, repeatable and blinded, which is why it is used to screen medications before large trials. It is not a measure of drinking in daily life, and a change in it is not a clinical outcome. The registration describes the trial as a human laboratory screening study, which is the correct description 2.

Which outcomes moved and which did not

The pattern is internally consistent. Participants on semaglutide drank on as many days as those on placebo but drank less on each of those days, both in the laboratory and by self-report, and reported less craving. Effect sizes were small in weeks 1 to 4 and grew in weeks 5 to 8, in step with the escalation 1. Six of 48 participants did not complete the outpatient visits. Adverse effects on semaglutide were the expected ones and mostly mild; there were no serious adverse events, no adverse interactions with alcohol and no treatment-related discontinuations. Mean weight changed by −5.05% on semaglutide and +0.18% on placebo 1.

Craving scales: self-report instruments and their weaknesses

Craving was measured weekly with the Penn Alcohol Craving Scale, a five-item self-report instrument scored 0 to 30 1. Self-reported craving is the outcome most often quoted from this trial and the one least able to bear the weight. It is subjective, retrospective over the preceding week, and susceptible to expectancy. In a blinded trial expectancy should be balanced, but a drug that produces nausea and reduces appetite may partially unblind participants, and the same gastrointestinal effects that reduce the desire to eat may reduce the desire to drink without acting on reward circuitry at all. The trial cannot separate those explanations. The laboratory consumption measures are more robust, because they record behaviour rather than report, and they moved in the same direction.

Why a 48-person trial cannot support an efficacy claim

A phase 2 screening trial is designed to detect whether a signal exists, not to estimate its size reliably or to show that it matters. Three features of this one limit inference. The sample is small, so the confidence intervals are wide: the primary effect on grams consumed is compatible with a standardised effect anywhere from −0.11 to −0.85. The duration is nine weeks, and alcohol use disorder relapses over years. The population was non-treatment-seeking with moderate severity and lower consumption than most treatment-seeking samples, so the result may not transfer to the people a treatment would be for. The authors list all three and call for replication in heavier drinkers and larger trials 1.

What a confirmatory trial in treatment-seeking populations would need

• Treatment-seeking participants with moderate to severe alcohol use disorder, since that is the population any indication would serve.
• A clinical primary endpoint accepted by regulators, such as the percentage of heavy drinking days or the proportion with no heavy drinking days over at least 12 to 24 weeks, rather than a laboratory session.
• Exposure at or above the level the screening trial reached, since the authors note that the low exposure may have limited detection.
• Pre-registration of the testing hierarchy, so that craving and per-occasion measures are not promoted after the fact over calendar-day measures that did not move.
• Measurement of gastrointestinal adverse effects and appetite alongside drinking, to test whether the effect is specific to alcohol or a general reduction in consumption, as the rodent data suggest.
• A sample in the hundreds, powered for the clinical endpoint, across multiple sites.

Where the evidence stops

In rodents, semaglutide reduces alcohol intake across models and blunts alcohol's dopaminergic effects, and also reduces intake of other solutions 34. In retrospective cohorts, people prescribed semaglutide have fewer alcohol-related diagnoses and hospitalisations, with the confounding those designs carry 56. In one randomised trial of 48 non-treatment-seeking adults, nine weeks of semaglutide reduced laboratory consumption, per-occasion drinking and reported craving, and left drinking frequency unchanged 1.

Those three tiers are consistent with each other. Consistency across tiers is encouraging; it is not the same as an efficacy result at the top tier, which does not yet exist. The correct description of the state of the evidence in October 2026 is that a mechanistic human signal has been obtained and a clinical trial has been justified. That is what the authors claimed, and it is all that can be claimed.