Deep divefrom Episode 12Fat loss11 November 202617 min

Can Weight-Loss Jabs Curb Drinking and Smoking?

A deep dive into one paper: what was known before, what the researchers did, what the numbers say and what they don't.

The paper: GLP-1 Receptor Agonists for treatment of tobacco and alcohol dependence: a systematic review and meta-analysis of randomised controlled trials (medRxiv preprint, 2026)

MP3

Chapters

Transcript

The script the two AI voices read from. The times are worked out from the script, not heard, so a line can start a few seconds early or late.

Read the whole episode

Cold open

Host 1You heard the trailer — this is the whole story. Welcome to the Hastrup Review of Science, where artificial intelligence reads real science, so you don't have to pretend you understood the paper.

Host 2Today is one paper: a review that gathers every randomised trial testing the new weight-loss and diabetes injections against smoking and drinking. It comes from Gemma Taylor and nine colleagues at the Universities of Bristol and Bath, and it was posted on the preprint server medRxiv on October twenty-third, so it has not yet been through peer review.

Host 1So the question is whether drugs like the ones people take to lose weight also make people smoke or drink less, and this team pulled together every proper experiment on that?

Host 2Yes, every randomised trial they could find, fourteen of them with one thousand one hundred and forty-nine people in total. The short answer is a modest, fairly consistent signal for less drinking, mostly from one drug called semaglutide, and no clear answer at all yet for quitting smoking.

Why this question exists

Host 1Before we get to the drugs, why do smoking and drinking get their own research field like this?

Host 2Because of the scale of harm. The paper cites more than seven point seven million deaths a year from smoking and about two point six million from alcohol, with global costs estimated at one point four and two point six trillion dollars.

Host 1And the medicines we already have for quitting, how well do those work?

Host 2The authors call them limited. For alcohol, the three classic drugs, acamprosate, naltrexone and disulfiram, have modest effects, few people take them, people often stop, and relapse is high.

Host 1What about smoking, is there a specific problem there that a new kind of drug could help with?

Host 2One the paper stresses is weight gain after quitting: fear of it puts people off trying, and no current treatment reliably prevents it. Some existing quit-smoking medicines also come with cautions for people with type two diabetes or mental health conditions.

Host 1So where do these weight-loss drugs come in? What does the name GLP one actually refer to?

Host 2GLP one is a gut hormone your body releases after a meal; it helps control blood sugar and tells the brain you have had enough. The drugs are called GLP one receptor agonists, which simply means they switch on the same receptor the hormone uses, but last much longer.

Host 1How long have these drugs been around?

Host 2Looking at the wider history of the field, the first one, exenatide, was approved for diabetes in the mid two thousands and was based on a molecule found in the saliva of the Gila monster lizard. Later came liraglutide, dulaglutide and semaglutide, and semaglutide in particular became famous for weight loss.

What was known before

Host 1And why would anyone think a drug for blood sugar might change smoking or drinking?

Host 2Because the receptor is not only in the gut and pancreas. The paper notes that these drugs act on the brain's reward circuitry, the dopamine pathways involved in addiction, and in animals they reduce nicotine self-administration, withdrawal, alcohol drinking and relapse-like behaviour.

Host 1When you say self-administration in animals, does that mean the rat can press a lever to get the drug itself?

Host 2Exactly that. If a drug makes an animal press the lever less often for nicotine or alcohol, that suggests the reward feels smaller, though it is a long way from a person at a bar.

Host 1Was there anything in actual people before this review?

Host 2Yes, two kinds of evidence. Large observational studies found that people prescribed these drugs drank less and engaged more with stop-smoking services, and more broadly in the field there have been many patient reports of simply losing interest in a drink.

Host 1Why isn't that enough? If lots of people on the drug drink less, isn't that already the answer?

Host 2No, because people who get prescribed these drugs differ from people who do not, in health, in contact with doctors, in motivation. The paper says those findings remain vulnerable to confounding and selection bias, which is why randomised trials matter.

Host 1And the randomised trials that did exist before, what did they show?

Host 2A mixed picture. In one trial, dulaglutide added to the quit-smoking drug varenicline did not improve quitting but did reduce weight gain afterwards, while exenatide added to nicotine patches increased quit rates and reduced weight gain.

The open question

Host 1So with all those small studies pointing in different directions, what exactly was still unanswered?

Host 2Whether, taken together, the trials show these drugs change smoking and drinking at all. Earlier reviews focused mostly on alcohol or on observational data, treated tobacco only in words rather than numbers, missed some eligible trials, and in one case seemed to pool starting values instead of outcomes.

What the team did

Host 1How did the team actually go about finding all the trials?

Host 2They registered their plan in advance on a public register called PROSPERO and followed the standard reporting guidance for reviews. Then they searched six large databases plus trial registries from the very beginning of each up to August eighteenth, twenty twenty-six.

Host 1Why does it matter that they registered the plan first?

Host 2It stops a team from quietly changing which outcomes or analyses count after seeing the data. They also report where they deviated from that plan, in their supplement.

Host 1How many papers did they have to go through to end up with fourteen trials?

Host 2They found twelve thousand five hundred and eighty-three records, screened four thousand three hundred and thirty-three titles and abstracts, and read two hundred and ninety-eight full reports. Two reviewers did the screening, data extraction and bias assessment independently, with a third settling disagreements.

Host 1What counted as a trial they would include?

Host 2A randomised trial comparing one of these drugs with a comparator in people who smoked or drank, measuring quitting, consumption, craving, dependence or risk. In practice every comparison group got a placebo, and the follow-up had to be at least twenty-eight days.

Host 1Who were the people in these trials? Were they all trying to quit?

Host 2Mostly not, and that matters. Only six of the fourteen trials recruited people actually seeking treatment, three for tobacco and three for alcohol; the other eight recruited people who were not trying to quit, often people with diabetes, obesity or schizophrenia.

Host 1Where were these trials done, and which drugs were tested?

Host 2Five in Denmark, six in the United States, two in the United Kingdom and one in Switzerland. Five tested liraglutide, six semaglutide, two exenatide and one dulaglutide, for six to fifty-two weeks, nearly all as injections.

Host 1If the trials measured different things in different ways, how do you add them together?

Host 2They convert each result into a standardised mean difference: the gap between drug and placebo, divided by how much people normally vary. That puts drinks per day, grams of alcohol and craving scores on one common scale.

Host 1Can you give me a feel for that scale? What counts as a big or small number?

Host 2As a rough rule of thumb in the wider field, around zero point two is called small, around zero point five medium and around zero point eight large. The authors warn that these units cannot be read directly as drinks or cigarettes avoided.

Host 1And when they combine the trials, do they just average them?

Host 2A weighted average, using what is called a random-effects model, which assumes each trial measures a slightly different true effect because the people and doses differ. For yes-or-no outcomes like quitting, they used odds ratios instead, where one means no difference.

Host 1How did they judge whether each trial could be trusted?

Host 2They used the Cochrane risk-of-bias tool on every result, and then rated how certain the overall evidence was for each outcome with a system called GRADE, from high down to very low.

The results

Host 1So let's start with the headline people care about: did these drugs help people quit smoking?

Host 2Not clearly. Three trials in people seeking to quit gave an odds ratio of one point three nine, pointing towards benefit, but the range of plausible values ran from zero point seven seven to two point five three, so it is compatible with both helping and harming.

Host 1So if I have this right, the arrow points the right way, but the result could still easily be zero?

Host 2That is right, and it holds for all four smoking outcomes: quitting, cigarettes per day, nicotine dependence and weight gain after quitting. Each one leaned towards the drug, and each one included no effect.

Host 1Wasn't there something about semaglutide and the number of cigarettes?

Host 2Yes. Restricted to three semaglutide trials, cigarettes per day dropped with a standardised difference of minus zero point one, with a tight range of minus zero point one six to minus zero point zero four.

Host 1Is minus zero point one a meaningful drop in cigarettes for someone who smokes?

Host 2It is small, below even the usual threshold for a small effect. And the authors point out that one trial contributed almost all the evidence for it, and those cigarette results were at high risk of bias.

Host 1What about the weight gain after quitting, which you said was a big reason people don't try?

Host 2Two trials, a standardised difference of minus zero point two seven, again leaning to the drug but crossing zero. Interestingly, this was the one outcome rated moderate certainty, because those two trials were relatively well conducted.

Host 1Now alcohol. Is the picture there stronger than for smoking?

Host 2Somewhat. Across all drugs, the one firm result was more days without alcohol, a standardised difference of zero point three five, with a range from zero point one two to zero point five nine, from three trials and two hundred and seventy-seven people.

Host 1And when they look only at semaglutide, does it get clearer?

Host 2Yes. Semaglutide alone reduced grams of alcohol per day by minus zero point five six, craving by minus zero point six, and drinks on drinking days by minus zero point five nine, all with ranges that stay clear of zero.

Host 1Those are around the medium size you mentioned. How many people are behind those semaglutide numbers?

Host 2Very few: two or three trials each, roughly one hundred to two hundred people. Heavy drinking days fell by minus zero point four five and drinks per calendar day by minus zero point four eight, but the latter rests on only ninety-eight people.

Host 1Was there anything where the drugs did nothing at all for alcohol?

Host 2Yes. The standard alcohol screening questionnaire, AUDIT, barely moved across five trials, at minus zero point zero three. The authors note that some of its questions ask about the past year, so it may not respond to a few months of change.

Host 1And drinks per week in the trials that weren't about drinking?

Host 2Three trials of liraglutide and dulaglutide in people not selected for alcohol problems showed almost nothing, minus zero point one two and crossing zero. Leaving out the dulaglutide trial, the estimate even tipped slightly toward placebo.

Host 1You mentioned a lot of these trials rated poorly on bias. How poorly?

Host 2Of forty-one results checked, only two were at low risk of bias, thirty-one raised some concerns, and eight were at high risk. Certainty was moderate for one outcome, low for four and very low for eight.

Host 1What was the main thing going wrong in those trials?

Host 2Missing data. Many participants dropped out, and dropping out was likely connected to their drinking or smoking, which can tilt the results either way.

A word from our sponsor

Host 2This message is a product recall from the makers of the Pooled Umbrella, the umbrella that combines fourteen small umbrellas into one.

Host 1I own one. On a good day it covers one thousand one hundred and forty-nine people, mostly strangers.

Host 2Owners report that the confidence interval around the canopy is wider than the canopy, and includes being completely wet.

Host 1Mine also showed substantial heterogeneity. Over my left shoulder it rained upwards.

Host 2If your umbrella has begun keeping only the days without rain, stop using it at once. Returned umbrellas will be standardised, and the weather they owe you will be posted back.

What it means

Host 1So what is the main lesson here? Do these drugs work for alcohol or not?

Host 2They may reduce how much people drink, with the clearest evidence for semaglutide. The authors' own words are that the evidence indicates therapeutic potential but does not establish a role in addiction treatment.

Host 1If someone listening is already on semaglutide for weight or diabetes, what does this tell them?

Host 2That if they find they drink a bit less, that fits with what these trials suggest. It does not tell anyone to start the drug for drinking, and nothing here is about doses or safety for that purpose.

Host 1Does drinking less but not stopping actually count as success?

Host 2It can, and the authors make that point: consumption may fall without abstinence, especially when stopping was never the goal. Whether it matters clinically depends on whether it actually reduces harm and fits what the patient wants.

Host 1Why does semaglutide look better than the others? Is it simply the stronger drug?

Host 2Maybe, but this review cannot say. The semaglutide trials also differed in their populations, outcomes and design, and nobody compared the drugs head to head, so the authors explicitly say superiority cannot be established.

Host 1And do we now know the reason it works, the brain reward idea?

Host 2No. Less drinking is consistent with dampened reward, as in the animal studies, but the authors say human trials cannot establish that mechanism.

Honest caveats

Host 1What are the biggest reasons to be cautious about all this?

Host 2Size first: most analyses had two to four trials and did not reach the number of people needed for a confident answer. Second, most alcohol outcomes were self-reported, and only one trial checked drinking with a blood marker.

Host 1Does it matter that many participants weren't trying to quit in the first place?

Host 2Yes, a lot. Only three of eight smoking trials and three of eleven alcohol trials recruited people seeking treatment, so many results describe people who never meant to change. The authors' checks within treatment-seeking groups were exploratory and often imprecise.

Host 1Who paid for these trials? Was the drug industry involved?

Host 2Six of the fourteen trials had industry funding or support, and industry-related conflicts among trial authors were common. The review notes that this can be associated with more favourable findings.

Host 1Could there be negative trials that never got published, making the whole thing look better?

Host 2It is possible. With only two to five trials per outcome, the usual tests for that are not reliable, so the authors say they cannot rule publication bias in or out.

Host 1Anything else limiting who these results apply to?

Host 2All the trials were in high-income countries, mostly in Denmark and the United States. And one cigarette result and one dependence result in the paper's text look like they have their sign flipped, a reminder that this is a preprint still awaiting review.

What to watch next

Host 1So what should we be looking out for over the next few years?

Host 2The review lists thirty trials that are ongoing or completed but not yet reported, so this evidence base could roughly triple. Those results will decide whether today's modest signal holds up.

Host 1What would a really convincing trial look like, according to the authors?

Host 2One in people actually seeking treatment, measuring consumption, craving, dependence and lasting abstinence over longer follow-up, reporting real drinks and cigarettes rather than standardised units, and verifying with biological tests where possible. They also want direct comparisons with the established treatments.

Host 1Are there other drugs coming along in the same family?

Host 2The authors point to tirzepatide, which acts on both the GLP one receptor and a second gut-hormone receptor; early animal and observational evidence suggests it may reduce drinking, and they call for trials in people with alcohol use disorder.

Host 1So, in a sentence, where does that leave us?

Host 2A modest, mostly semaglutide signal for drinking less, an unanswered question for smoking, and evidence the authors themselves rate as low to very low certainty. Worth watching closely, not yet worth acting on.

Sign-off

Host 1That was the Hastrup Review of Science deep dive. Every weekend this same feed carries the roundup, the week's new papers in one short episode.