What should we do with a drug that appears to treat addiction without being designed to?

That question sharpened for me when I read the new Lancet Psychiatry Swedish register-based study on GLP-1 receptor agonists and substance use disorder hospitalizations. The researchers used a within-individual design, following the same patients across periods of GLP-1 treatment and non-treatment, which is exactly the right method to strip out confounding. The signal was unambiguous: GLP-1 treatment was associated with meaningfully lower rates of alcohol use disorder and substance use disorder-related hospital admissions. And then came the detail that should rearrange how the field thinks about this class of drugs. After discontinuation, the protective association persisted for alcohol use disorder through the first 182 days. For substance use disorder broadly, it did not.

A Mechanism the Reward Circuit Can’t Ignore

The pharmacology here is not mysterious, even if the clinical implications remain unsettled. GLP-1-producing neurons in the nucleus of the solitary tract project directly to the ventral tegmental area and the nucleus accumbens, the core dopamine nodes of the reward circuit. This means GLP-1 receptor agonists are not arriving at addiction pathways through some indirect metabolic detour, they are landing on the same circuitry that drives compulsive substance use. A 2023 eBioMedicine study by Aranäs and colleagues showed that semaglutide significantly reduced alcohol intake and relapse-like drinking in rodents, and attenuated alcohol’s capacity to elevate dopamine in the nucleus accumbens. That finding is mechanistically coherent with what the Swedish data are showing in human populations at scale.

The discontinuation asymmetry is the piece I keep turning over clinically. The fact that the AUD protective association persists for up to six months after stopping the medication, while the broader SUD benefit does not, suggests we may be looking at two different mechanistic processes. Alcohol use disorder has a slower-cycling relapse biology, the neuroadaptive changes driving craving and compulsion build over longer time horizons than, say, stimulant or opioid use disorder. It is plausible that sustained GLP-1 exposure produces some durable neuroadaptive correction in the mesolimbic system specifically for alcohol that doesn’t generalize across substances. Or the Swedish registry is capturing something more prosaic: that AUD patients who benefit metabolically from GLP-1 treatment sustain behavioral changes longer because the comorbid conditions driving drinking (metabolic stress, sleep disruption, chronic pain) are still partially controlled post-discontinuation. The data cannot distinguish between those explanations yet.

The comorbidity overlap here is substantial and often underestimated. A 2023 nationally representative cross-sectional study found that an estimated 3.8% of U.S. adults carry both alcohol use disorder and obesity simultaneously, representing millions of patients who are already candidates for GLP-1 therapy on metabolic grounds alone, and who are walking into addiction medicine clinics without anyone connecting those two treatment tracks.

What Trial Designers Are Getting Wrong

The registry methodology in this study deserves a candid note. A systematic review of the Swedish National Patient Register found a median positive predictive value of 84% for diagnostic codes against patient chart data, solid for population-level inference, but not granular enough to tell us anything about drinking quantity, craving scores, or the functional severity of relapse events. Hospitalizations are the iceberg’s tip. In my practice running trials across addiction and mood disorder populations, the patients who matter most clinically, the ones who relapse but don’t hit an ED, the ones who cut back but don’t abstain, the ones managing with telehealth support across distance, are invisible in register data. That is not a criticism of the Swedish team’s design, which is rigorous within its constraints. It is a gap the next generation of trials must close.

The discontinuation finding should be driving a specific protocol conversation right now. The precedent from alcohol pharmacotherapy is instructive: a meta-analysis of 64 randomized trials found that acamprosate’s benefits were most pronounced during active treatment and attenuated after discontinuation, which is why continuous treatment duration became a design variable in later AUD trials. GLP-1 trials examining addiction outcomes need the same rigor around cessation windows, pre-specified follow-up at 90, 182, and 365 days post-discontinuation, with drinking quantity and craving endpoints, not just hospitalization rates. And they need to enroll patients who carry both metabolic and substance use diagnoses, not exclude them as comorbid noise. The dual-diagnosis population is the signal, not the confounder.

Having treated patients at the intersection of obesity and alcohol use disorder through both in-person and remote telehealth settings, I can tell you the clinical picture is more entangled than any single-indication trial captures. A patient whose GLP-1 prescription lapses because their insurer stops covering it is facing metabolic destabilization and, potentially, a six-month window of elevated addiction risk simultaneously. That is not a pharmacology question anymore, it is a systems design question, and somebody needs to study it prospectively.

The Forward Signal

I’ll be watching whether the FDA’s expanding semaglutide indications, cardiovascular risk reduction approved March 2024, sleep apnea, kidney disease following, create enough prescribing volume in high-comorbidity populations to generate the naturalistic data that could power a properly designed addiction trial. The Swedish study is the hypothesis. The randomized evidence is still owed.

References

  1. Lancet Psychiatry, “Association of GLP-1 receptor agonists with alcohol use disorder-related and substance use disorder-related hospital admissions during treatment and after discontinuation: a Swedish register-based within-individual observational study”
  2. Aranäs et al., eBioMedicine (2023), “Semaglutide reduces alcohol intake and modulates dopamine in the nucleus accumbens in rodents”
  3. Endocrinology (2012), “GLP-1 Neurons in the Nucleus of the Solitary Tract Project Directly to the Ventral Tegmental Area and Nucleus Accumbens to Control for Food Intake”
  4. European Journal of Epidemiology, “Validation of diagnostic codes in the Swedish National Patient Register: systematic review”
  5. Maisel et al., “Meta-analysis of naltrexone and acamprosate for treating alcohol use disorders”
  6. Cross-sectional study (2023), “Comorbidity of heavy drinking, alcohol use disorder, and obesity among US adults”
  7. Pharmacy Times, “FDA Approves Semaglutide for Cardiovascular Disease Indication (March 2024)”
+ posts

Dr. Leonardo Vando, MD is Principal Investigator of Psychiatry at Vando Medical Services.