Interesting! I feel like I slightly disagree with the general conclusion that this study provides evidence against semaglutides being good (or miraculous) for alcohol use disorder for a couple reasons:
-The placebo was SO good (reducing alcohol use by >1/2 over a few weeks) that it's hard to imagine that even if GLP-1 was working optimally it could perform any better! I think the takeaway here is less that we can learn anything about semaglutide's utility and more that bringing alcoholics into a lab and injecting them once a week is a miraculously good intervention against alcohol use disorder. But it's possible that this could fail to generalize over long time horizons, and obviously seems impractical, whereas semaglutides could continue to work as well as the placebo is working in this study (we have no evidence either way to this).
-I basically think that outcomes 2 and 3 from the study are epistemically near-useless in determining whether semaglutides are effective. Like, the delayed drinking experiment is obviously so silly and I don't think would provide any evidence in any direction regardless of the outcome. Like, if you told me that every person in the control group drank the alcohol immediately and every person in the experimental group waited the entire 50 minutes, I still probably wouldn't index on it at all, just because of how meaningless it is as a proxy for alcoholism, and I would assume the drugs or the lab setting were just having some second-order effect on people's patience.
I agree that the strong reductions in the placebo group reduce the evidence that this provides against semaglutide working. In a way it's more than just placebo—in all studies it seems like any kind of regular contact with doctors helps. Or it could be that just the act of *recording* how much you drink causes reductions.
That said, I'd still have hoped that semaglutide would have provided some benefit on top of that placebo (or whatever) effect. It would be a somewhat strange mechanism for semaglutide to be helpful by itself, but have those benefits totally masked by placebo etc.
I mean, reducing alcohol consumption to about 1 drink per day across two dozen participants might just be a practical limit to the efficacy of basically any intervention (with this experimental setup and recruitment, at least). And to be fair, it's not the experimental group's fault that they started with a lower baseline; they did end up at a lower final value in the end, and perhaps % reduction is a less relevant figure of merit than final consumption, since you might get diminishing returns below a certain threshold, due to some fraction of the sample size just not showing any benefits, and another (perhaps very significant, due to the DSM classification) fraction not aiming to cut their alcohol consumption below, say, 1 drink a day.
Also, you could imagine a world, though it's slightly far-fetched, where:
>Semaglutide is very effective at reducing "cravings" or something, and thus is effective at treating alcohol use disorder.
>The placebo (contact w doctor, study participants trying to "get serious" about dealing with alcohol use, etc) is very effective.
>Semaglutide also reduces the effectiveness of these other placebo considerations for some yet-to-be-understood reason (makes patients' "cravings" to succeed at the initial goal of reducing alcohol use go down, while still diminishing their "cravings" for alcohol, idk lol)
>It appears that semaglutide is no better than or even WORSE than the placebo at treating alcohol use disorder, despite it actually being efficacious.
Yup, that's totally possible! It's a judgement call, but I personally would rate that plausible. To put it another way, if there was no decline in either group, then the results of this paper would be *much* worse for the prospects of semaglutide being helpful for alcoholism. Future studies should try to avoid this somehow.
That being said, I guess I feel like there are a lot of moving parts in this story and I wouldn't have predicted it in advance. It's something that I rate as plausible mostly because it's the easiest way to reconcile the experiment with the (pretty strong) anecdotal evidence. If I start from a baseline of already having that anecdotal evidence, I still tend to see this as bad news.
Maybe, although I'd think most people would join AA or something? Also, the average person was only averaging around 3 drinks/day at the start. That's probably not super healthy, but I also know a *lot* of people who do that for years and still live productive lives. (I'd also have thought there would be a significant delay between recruitment and the study actually starting, although from reading the paper I can't actually tell if that's true.) My guess is that the decline in the placebo group is more due to the placebo effect or contact with doctors, etc.
Thank you for this! It's very thought-provoking for me. I have been following the research on this as closely as I can. I am on tirzepatide, not semiglutide. I have been obese for most of my life and definitely fall into the "alcohol use disorder" level of consumption. My experience has been fairly dramatic weight loss, and that alcohol is less appealing. My alcohol consumption has dropped some, but not as much as I hoped it effortlessly would. Basically, fried food and sugar do nothing for me now, and I get full fast when I eat, but I still like my drinks, and they still go down easy.
Glad to hear it is at somewhat helpful for you! The possible mechanisms here are quite mysterious. For example, I've noticed that when I'm jetlagged, I find sugar (and everything else) much harder to resist. The brain is a strange thing.
Does citing a DSM-V diagnostic help you pass IRB? Is there any other reason to open the DSM these days? It feels like citing a dictionary definition at the start of a term paper.
Anyway I've been on this family of drugs for a couple years now and lost 20% of my starting body mass, which is REALLY nice, but that is not what excites me most about GLP1 family drugs. The biggest difference to me is learning about my body without the shame of "failing" a "diet". Incidentally I DETEST the term 'losing weight' - what you should mean is 'improve body composition'. I can lose weight by chopping off a leg after all. Or put in more realistic terms, everyone's weight can fluctuate by 2-5% weekly just by chance. If you are thinking in terms of losing weight, the scale will only confuse you.
What makes me really excited is the idea that researchers and clinicians might start taking human perceptions and sensations seriously.
If I won the lottery I would fund studies into perceptions associated with hunger and satiation. It sounds very simple - hungry is hungry, right? But there are a bundle of hormones and neurotransmitters mediating the feeding response. I'm most curious as to whether there are sensations that can be differentiated to the different hormone states.
See 'Hungry Brain' by Guyenet
That is, can people learn to feel or notice:
* High and low blood sugar (somewhat yes, though the feeling of hypoglycemia is more associated with blood sugar decreasing rapidly)
* increased hunger over different time periods after consuming alcohol
* fat metabolism mode (higher CO2 in blood?)
* ghrelin or leptin more directly
* the increase in hunger while eating a meal and being shown a new food item like desert
I think using the DSM criteria is most helpful for not giving reviewers any excuse to complain about how you picked the patient population! I agree with you that hunger is very complicated. Most of all, I learned this from trying the potato diet: https://dynomight.net/potato-diet/ (I also recommend "The Hungry Brain")
"(People think the problem with science is that it’s too woke. While I don’t really disagree, I still think the bigger problem is screwed up incentives that force everyone oversell everything, because that’s what you have to do to survive. But that’s a story for another time.)"
This seems like the kind of research where have some medium/high amount of regulation is more reasonable. ("Let's give synthetic hormones and alcohol to people with alcohol use disorders!") But still, it does seem like the regulatory burden is quite enormous! I'd love to hear from people doing this kind of research if you couldn't have a 50 page application that still provided the same level of protection.
Also, it seems like that application preregisters a lot of the statistical analysis. I think preregistering statistical analysis is GREAT, but I'm not sure why that needs to be part of the same process...
There's a lot of research like this - a lot - and so much of it gets cited as peer-reviewed expert opinions. The underlying problem with this paper in particular is how small the study sample is.
Even if it were randomized, you still wouldn't be statistically balanced anyway and there could be some unobservable characteristic that could still mess with the results. In this case we were lucky to have an observable characteristic to invalidate the finding. At minimum you would have thought they would control for these covariates when presenting the findings.
I wish at the top of every research paper they would say:
randomized/not-randomized
sufficiently powered/not sufficiently powered
how people were recruited into the study
how were outcomes collected
i think some sort of clear ratings system based on the above would help average joe navigate good empirical results from maybe good empirical results.
Honestly, despite my quibbles about the *writing*, I feel the study itself was quite well done. They did about as good a job as you could ask of balancing the two groups. They give stats on sex/age/race/alcohol/cigarettes/BMI/blood pressure and it all looks pretty good. It's a small-ish sample, sure, but it makes sense for the first study to be on the smaller side, and that's reflected in the confidence intervals.
Personally, I don't trust control variables and lean heavily towards randomization instead. :) (https://dynomight.net/control/) My guess is that they just didn't predict that so many people would decline to participate in the last experiment. It's very hard to anticipate these kinds of things!
agree on controls and it's sort of part of the primary issue i guess i have with studies like these.
an observed or unobserved factor can wildly swing a small sample, non-randomized study in many directions and magnitudes. you demonstrated what it would look like to control for different BMI segments and that swung the finding one way - some other factor/control observed or unobserved could have swung it another.
Couldn't they find any people who had more severe drinking problems? If somoeone is consuming close to the threshhold then they're holding things together pretty well, if not totally abstaining.
Looks like they excluded most people who had severe problems:
> Key exclusion criteria included currently seeking treatment for alcohol problems or actively attempting to reduce consumption; past use of GLP-1 receptor agonists; weight loss medications; body mass index (BMI, calculated as weight in kilograms divided by height in meters squared) less than 23; past-year substance use disorder other than AUD, tobacco use disorder, or mild cannabis use disorder; recent (30-day) use of illicit drugs except cannabis; history of diabetes, and current medical or neurological illness precluding participation based on physician judgement
Since they were actively giving the people alcohol, I imagine it would have been quite hard to get approval if they included people with severe drinking problems.
Thanks for this. I'm guilty of sharing the graphs from this study online *without* reading the appendix!!
I think the primary outcome is poorly chosen for voluntary participation. What sort of person wants to drink booze by themself in a hospital in (I assume) daylight hours? The answer is mostly the kind of person who they already excluded: hardcore alcoholics.
I would love to see participation in the voluntary drinking session by time of day. If bet if people's drinking session was meant to begin at 11am they'd be more likely to just go home than if it began at, say, 330pm.
Now that you mention it, you could actually treat the decision to agree to do this experiment itself as an outcome!
If I'm reading the paper right:
- There were 24 people in each group at the start of the trial.
- At the end of the trial, 22 people from the semaglutide group showed up for the final visit and 19 from the control group.
- Of those, 13 people in the semaglutide group and 12 in the placebo group agreed to the experiment.
That would make it 13/22=59% for the semaglutide group and 12/19=63% for the control group. So... uh.... I guess that doesn't tell us very much after all.
Is it possible that we have a causal effect like (GLP-1) -> (weight loss) -> (less drinking), where the second arrow just follows from getting drunk faster (maybe in addition to a direct arrow)?
Interesting! I feel like I slightly disagree with the general conclusion that this study provides evidence against semaglutides being good (or miraculous) for alcohol use disorder for a couple reasons:
-The placebo was SO good (reducing alcohol use by >1/2 over a few weeks) that it's hard to imagine that even if GLP-1 was working optimally it could perform any better! I think the takeaway here is less that we can learn anything about semaglutide's utility and more that bringing alcoholics into a lab and injecting them once a week is a miraculously good intervention against alcohol use disorder. But it's possible that this could fail to generalize over long time horizons, and obviously seems impractical, whereas semaglutides could continue to work as well as the placebo is working in this study (we have no evidence either way to this).
-I basically think that outcomes 2 and 3 from the study are epistemically near-useless in determining whether semaglutides are effective. Like, the delayed drinking experiment is obviously so silly and I don't think would provide any evidence in any direction regardless of the outcome. Like, if you told me that every person in the control group drank the alcohol immediately and every person in the experimental group waited the entire 50 minutes, I still probably wouldn't index on it at all, just because of how meaningless it is as a proxy for alcoholism, and I would assume the drugs or the lab setting were just having some second-order effect on people's patience.
I agree that the strong reductions in the placebo group reduce the evidence that this provides against semaglutide working. In a way it's more than just placebo—in all studies it seems like any kind of regular contact with doctors helps. Or it could be that just the act of *recording* how much you drink causes reductions.
That said, I'd still have hoped that semaglutide would have provided some benefit on top of that placebo (or whatever) effect. It would be a somewhat strange mechanism for semaglutide to be helpful by itself, but have those benefits totally masked by placebo etc.
I mean, reducing alcohol consumption to about 1 drink per day across two dozen participants might just be a practical limit to the efficacy of basically any intervention (with this experimental setup and recruitment, at least). And to be fair, it's not the experimental group's fault that they started with a lower baseline; they did end up at a lower final value in the end, and perhaps % reduction is a less relevant figure of merit than final consumption, since you might get diminishing returns below a certain threshold, due to some fraction of the sample size just not showing any benefits, and another (perhaps very significant, due to the DSM classification) fraction not aiming to cut their alcohol consumption below, say, 1 drink a day.
Also, you could imagine a world, though it's slightly far-fetched, where:
>Semaglutide is very effective at reducing "cravings" or something, and thus is effective at treating alcohol use disorder.
>The placebo (contact w doctor, study participants trying to "get serious" about dealing with alcohol use, etc) is very effective.
>Semaglutide also reduces the effectiveness of these other placebo considerations for some yet-to-be-understood reason (makes patients' "cravings" to succeed at the initial goal of reducing alcohol use go down, while still diminishing their "cravings" for alcohol, idk lol)
>It appears that semaglutide is no better than or even WORSE than the placebo at treating alcohol use disorder, despite it actually being efficacious.
Yup, that's totally possible! It's a judgement call, but I personally would rate that plausible. To put it another way, if there was no decline in either group, then the results of this paper would be *much* worse for the prospects of semaglutide being helpful for alcoholism. Future studies should try to avoid this somehow.
That being said, I guess I feel like there are a lot of moving parts in this story and I wouldn't have predicted it in advance. It's something that I rate as plausible mostly because it's the easiest way to reconcile the experiment with the (pretty strong) anecdotal evidence. If I start from a baseline of already having that anecdotal evidence, I still tend to see this as bad news.
Isn't this just regression to the mean? Presumably people with alcohol use disorder seek help at their lowest point.
Maybe, although I'd think most people would join AA or something? Also, the average person was only averaging around 3 drinks/day at the start. That's probably not super healthy, but I also know a *lot* of people who do that for years and still live productive lives. (I'd also have thought there would be a significant delay between recruitment and the study actually starting, although from reading the paper I can't actually tell if that's true.) My guess is that the decline in the placebo group is more due to the placebo effect or contact with doctors, etc.
Thank you for this! It's very thought-provoking for me. I have been following the research on this as closely as I can. I am on tirzepatide, not semiglutide. I have been obese for most of my life and definitely fall into the "alcohol use disorder" level of consumption. My experience has been fairly dramatic weight loss, and that alcohol is less appealing. My alcohol consumption has dropped some, but not as much as I hoped it effortlessly would. Basically, fried food and sugar do nothing for me now, and I get full fast when I eat, but I still like my drinks, and they still go down easy.
Glad to hear it is at somewhat helpful for you! The possible mechanisms here are quite mysterious. For example, I've noticed that when I'm jetlagged, I find sugar (and everything else) much harder to resist. The brain is a strange thing.
Does citing a DSM-V diagnostic help you pass IRB? Is there any other reason to open the DSM these days? It feels like citing a dictionary definition at the start of a term paper.
Anyway I've been on this family of drugs for a couple years now and lost 20% of my starting body mass, which is REALLY nice, but that is not what excites me most about GLP1 family drugs. The biggest difference to me is learning about my body without the shame of "failing" a "diet". Incidentally I DETEST the term 'losing weight' - what you should mean is 'improve body composition'. I can lose weight by chopping off a leg after all. Or put in more realistic terms, everyone's weight can fluctuate by 2-5% weekly just by chance. If you are thinking in terms of losing weight, the scale will only confuse you.
What makes me really excited is the idea that researchers and clinicians might start taking human perceptions and sensations seriously.
If I won the lottery I would fund studies into perceptions associated with hunger and satiation. It sounds very simple - hungry is hungry, right? But there are a bundle of hormones and neurotransmitters mediating the feeding response. I'm most curious as to whether there are sensations that can be differentiated to the different hormone states.
See 'Hungry Brain' by Guyenet
That is, can people learn to feel or notice:
* High and low blood sugar (somewhat yes, though the feeling of hypoglycemia is more associated with blood sugar decreasing rapidly)
* increased hunger over different time periods after consuming alcohol
* fat metabolism mode (higher CO2 in blood?)
* ghrelin or leptin more directly
* the increase in hunger while eating a meal and being shown a new food item like desert
* others that are not coming to mind right now
I think using the DSM criteria is most helpful for not giving reviewers any excuse to complain about how you picked the patient population! I agree with you that hunger is very complicated. Most of all, I learned this from trying the potato diet: https://dynomight.net/potato-diet/ (I also recommend "The Hungry Brain")
"(People think the problem with science is that it’s too woke. While I don’t really disagree, I still think the bigger problem is screwed up incentives that force everyone oversell everything, because that’s what you have to do to survive. But that’s a story for another time.)"
I AM SO READY.
Seems like a bigger problem might be that a 300 page application is needed to do research that is extremely important and we want ASAP!
This seems like the kind of research where have some medium/high amount of regulation is more reasonable. ("Let's give synthetic hormones and alcohol to people with alcohol use disorders!") But still, it does seem like the regulatory burden is quite enormous! I'd love to hear from people doing this kind of research if you couldn't have a 50 page application that still provided the same level of protection.
Also, it seems like that application preregisters a lot of the statistical analysis. I think preregistering statistical analysis is GREAT, but I'm not sure why that needs to be part of the same process...
There's a lot of research like this - a lot - and so much of it gets cited as peer-reviewed expert opinions. The underlying problem with this paper in particular is how small the study sample is.
Even if it were randomized, you still wouldn't be statistically balanced anyway and there could be some unobservable characteristic that could still mess with the results. In this case we were lucky to have an observable characteristic to invalidate the finding. At minimum you would have thought they would control for these covariates when presenting the findings.
I wish at the top of every research paper they would say:
randomized/not-randomized
sufficiently powered/not sufficiently powered
how people were recruited into the study
how were outcomes collected
i think some sort of clear ratings system based on the above would help average joe navigate good empirical results from maybe good empirical results.
Honestly, despite my quibbles about the *writing*, I feel the study itself was quite well done. They did about as good a job as you could ask of balancing the two groups. They give stats on sex/age/race/alcohol/cigarettes/BMI/blood pressure and it all looks pretty good. It's a small-ish sample, sure, but it makes sense for the first study to be on the smaller side, and that's reflected in the confidence intervals.
Personally, I don't trust control variables and lean heavily towards randomization instead. :) (https://dynomight.net/control/) My guess is that they just didn't predict that so many people would decline to participate in the last experiment. It's very hard to anticipate these kinds of things!
agree on controls and it's sort of part of the primary issue i guess i have with studies like these.
an observed or unobserved factor can wildly swing a small sample, non-randomized study in many directions and magnitudes. you demonstrated what it would look like to control for different BMI segments and that swung the finding one way - some other factor/control observed or unobserved could have swung it another.
then what? ¯\_(ツ)_/¯
Couldn't they find any people who had more severe drinking problems? If somoeone is consuming close to the threshhold then they're holding things together pretty well, if not totally abstaining.
Looks like they excluded most people who had severe problems:
> Key exclusion criteria included currently seeking treatment for alcohol problems or actively attempting to reduce consumption; past use of GLP-1 receptor agonists; weight loss medications; body mass index (BMI, calculated as weight in kilograms divided by height in meters squared) less than 23; past-year substance use disorder other than AUD, tobacco use disorder, or mild cannabis use disorder; recent (30-day) use of illicit drugs except cannabis; history of diabetes, and current medical or neurological illness precluding participation based on physician judgement
Since they were actively giving the people alcohol, I imagine it would have been quite hard to get approval if they included people with severe drinking problems.
good point - but sadly renders the test close to useless
Thanks for this. I'm guilty of sharing the graphs from this study online *without* reading the appendix!!
I think the primary outcome is poorly chosen for voluntary participation. What sort of person wants to drink booze by themself in a hospital in (I assume) daylight hours? The answer is mostly the kind of person who they already excluded: hardcore alcoholics.
I would love to see participation in the voluntary drinking session by time of day. If bet if people's drinking session was meant to begin at 11am they'd be more likely to just go home than if it began at, say, 330pm.
Now that you mention it, you could actually treat the decision to agree to do this experiment itself as an outcome!
If I'm reading the paper right:
- There were 24 people in each group at the start of the trial.
- At the end of the trial, 22 people from the semaglutide group showed up for the final visit and 19 from the control group.
- Of those, 13 people in the semaglutide group and 12 in the placebo group agreed to the experiment.
That would make it 13/22=59% for the semaglutide group and 12/19=63% for the control group. So... uh.... I guess that doesn't tell us very much after all.
Is it possible that we have a causal effect like (GLP-1) -> (weight loss) -> (less drinking), where the second arrow just follows from getting drunk faster (maybe in addition to a direct arrow)?