228,000 People, 17 Years of Data: The Swedish Study That Put GLP-1 and Alcohol on the Map

In November 2024, a research team led by Markku Lähteenvuo at the University of Eastern Finland published a study so large it was hard to argue with. Using Swedish nationwide registries spanning 2006 to 2023, they tracked every resident aged 16-64 who had a diagnosis of alcohol use disorder — 227,886 people — and asked a simple question: were they hospitalized less often during periods when they were taking GLP-1 medications?

The answer: yes, substantially.

Why This Study Design Matters

Most addiction-related GLP-1 studies so far have been between-person comparisons — comparing semaglutide patients against patients on other medications. That design is vulnerable to confounding. Maybe semaglutide patients are healthier, wealthier, more motivated, or different in ways the data can't capture.

The Lähteenvuo study sidestepped this problem with a within-individual design. Each person served as their own control. The researchers compared each patient's hospitalization risk during periods they were on a GLP-1 medication against periods they were not. Same person, different time windows. This design eliminates most time-invariant confounding — because genetics, socioeconomic background, personality, and chronic comorbidities don't change between the comparison periods.

The Numbers

Of the 227,886 individuals with AUD in the cohort, 4,321 used semaglutide and 2,509 used liraglutide at some point during the follow-up period. The median follow-up was 8.8 years.

The researchers also compared GLP-1 agonists head-to-head against the three FDA-approved AUD medications: naltrexone, acamprosate, and disulfiram. The comparison was striking — semaglutide performed comparably to or better than these approved treatments on the hospitalization outcome. While this doesn't mean semaglutide is an AUD treatment (different indication, different patient selection), it puts the effect size in context.

What Made This Study Credible

Several features of the study design gave it unusual credibility in the observational literature:

Scale. Nearly 228,000 people is enormous for an addiction study. Most GLP-1/addiction research involves hundreds or thousands — this involved hundreds of thousands.

Registry data. Sweden's national health registries capture virtually all healthcare encounters, prescriptions, and hospitalizations for every resident. There's no selection bias from who enrolls or who gets lost to follow-up. If you live in Sweden and get hospitalized, it's in the data.

The PRE2DUP method. The research team used their own validated method (PRE2DUP) for defining drug exposure periods from prescription registry data. This method accounts for varying dosing, stockpiling, and gaps in treatment — reducing the risk of misclassifying exposure status.

Long follow-up. A median of 8.8 years of follow-up per person means this isn't capturing a short-term novelty effect. The reduced hospitalization risk persisted over years of exposure.

Limitations and What It Doesn't Prove

Other limitations: the study used hospitalization as the outcome, which captures severe AUD events but misses outpatient drinking patterns, moderate consumption changes, and quality-of-life impacts. Someone could drink significantly less without ever being hospitalized — or continue drinking but avoid the ER for other reasons.

The indication bias also deserves mention. People prescribed GLP-1 medications have diabetes or obesity — conditions that already put them in regular medical contact. This ongoing medical supervision might independently reduce hospitalization risk through better overall health management.

Where This Fits in the Evidence Base

The Lähteenvuo study sits between two other landmark papers:

Below it in evidence strength: the Wang/Volkow Nature Communications studies using U.S. EHR data, which compared semaglutide patients against patients on other medications (between-person design, more vulnerable to confounding).

Above it: the Hendershot JAMA Psychiatry RCT, which randomized 48 participants to semaglutide or placebo and showed direct reductions in drinking behavior under controlled conditions.

Together, the three studies form a compelling evidence chain: large-scale observational signal → sophisticated within-individual registry confirmation → randomized experimental validation. Each study's strengths compensate for the others' limitations.