I’m interested in an integrated and imaginative approach to treating psychosis and schizophrenia which is free from a world view wedded to one type of approach. I do agree that it includes a mismatch between interior bottom up signals and top down ones. These signals contain hundreds of neurotransmitters and millions of neurons and their glial cells, so it has a chemical level component. It can also be described at the functional or cognitive level (the word cognitive I find very steampunk to be honest with you). Many neuroleptic drugs have helped people live normal lives when they stay on them. These people in recovery then move away from the healthcare system and are no longer labelled as schizophrenic. This creates a smaller and smaller group that can create the illusion that the condition cannot be cured. Most relapses occur when people come off their meds because they feel well and don’t think they need them anymore. Many people who are labelled “treatment resistant” simply aren’t taking the medication and not telling anyone that they aren’t. They behave like this because they don’t trust authority figures. Kindness and compassion are surprisingly effective in these situations. Psychotics respond really well to compassion, even if they are unable to articulate this at the time. Your research sounds really interesting and I urge you to do a study if you are able. For the human and psychological component of recovery, do have a look at some of our blogs on here for a window into our lives as people, not patients ❤️
Hi Sarah - thanks so much for reading and your thoughtful comment. I took a look through your blog and enjoyed it very much, thank you for sharing!
I agree that there are many patients who are labeled as "treatment resistant" who are simply not taking their medications, and that there are many who are able to live normal lives on antipsychotics. That said, I think the scientific literature shows pretty clearly that there is a large proportion of individuals with schizophrenia who will only get a modest benefit from antipsychotics, and a good chunk who really are treatment resistant.
More importantly, though, I agree even more with your point that kindness, compassion, and empathy are important tools in these situations. Learning how to empathize and form an alliance with a psychotic individual, so that we can work together to make their lives more worth living, is one of the hardest but most rewarding parts of my job.
Thank you for replying. I’m trying to make some constructive connections between psychiatrists and people in recovery outside the confines of the doctor and patient relationship. I like Substack and feel it’s a good platform to achieve this while avoiding stressful chat room rants back and forth. I can also recommend Mat’s Blog on Schizoaffective Disorder and The Manic Messiah. Nathan Harmon has just started his. I’m building working support between us for peer support but also to access good advice and best practice. I’m aware that you are probably too busy to be across what we are all writing about, but I just want you to know about us in case we can help people with their recovery after they are discharged
Great find in that Xie article. I've been using tirzepatide extensively in pts on clozapine and anecdotally I see improvement in schizophrenia symptoms as well. Of course to justify to insurance they have to have diabetes, BMI, etc etc, but who on long term clozapine maintenance doesn't have severe metabolic syndrome really. I really want to run a small trial just adding it as an adjunct.
One of the saddest things about prescribing antipsychotics to me is that Im basically taking 15 years off a persons life because of metabolic syndrome. Really excited about the potential of these agents to mitigate that.
Note that if you do start to use the GLP1s the data we do have show tirzepatide is the far more effective agent than semaglutide and with a more favorable side effect profile. I don't use semaglutide anymore and its not because im an eli lilly shill, its because the data are pretty conclusive. Unfortunately I still see a lot of physicians just instinctively going for the semaglutide because it came out first I think.
In general I think your theory is cool but idgaf. Give me a randomized trial, it works or it doesn't regardless of mechanism.
Also, I don't think you should be so gloomy about prescribing antipsychotics, at least for schizophrenics. There are many studies that show that all-cause mortality is improved for patients with schizophrenia on antipsychotics vs. not, which wouldn't be the case if metabolic syndromes were actually knocking off 15 years.
Though maybe your point is that in the absence of metabolic effects those patients would also get those 15 years back? I don't think the evidence actually supports this. If you look at Figure 4 of this paper (https://pmc.ncbi.nlm.nih.gov/articles/PMC6953552/), there's no evidence that the more metabolically active antipsychotics reduce all-cause mortality less relative to less metabolically active ones. In fact, clozapine and olanzapine are two of the best at reducing all-cause mortality, outperforming risperidone.
Hmmm very interesting doc, very interesting, I will look into this more although I dont buy the its because they have better adherence to healthy lifestyle and medical care explanation, I think that data is confounded im just not sure how
I think the issue might be the follow up period still isnt long enough at 20 years, like is whats killing schizophrenics really being captured? gotta crack the hood on this one
I mean, average age at entry was ~45 and average follow up was 14 years. If >50% of your study population is being followed into their late 50s, that's pretty much hitting the average life expectancy for individuals with schizophrenia, so I don't think you can say that the issue is that follow up wasn't long enough.
I've had this conversation with a few people and they make the point that this can't possibly be true and that the data must be confounded, but also nobody has any good explanations for why the populations on risperidone and olanzapine should be all that different.
That's all to say that I'm confused by it too! Let me know if you find anything that makes sense.
"I think your theory is cool but idgaf" is generally the same sentiment I operate under! Very cool to hear that there might be something there in actually psychotic patients, but like you said, we need some trials. Maybe in a decade or so
I was sort of trying to parse this and started by thinking I'd maybe found a flaw when I think I probably just found another way to frame it that ends up making it plainly consistent. Took a quick look at Reconsidering Dopamine to see if you agreed, but I've seen Scott Alexander theorize elsewhere also that dopamine acts as an element of predictive confidence, so its involvement in schizophrenia may be related to overconfidence in low-quality predictions that overcook-the-books on the sense data to serve up conscious hallucinations.
If we start by suggesting that GLP1RAs might modulate surprisal by nudging prediction vs sense data into baseline alignment, for example by working for food/crack appetite by decreasing predictive confidence in "we should really be hungry/jonesing now" versus the sensory-received "nah we're actually good here", we could see a possible, straightforward similar relationship between decreasing schizophrenic-style broad severe errors of predictive confidence versus the normal generic sensory data.
Reconsidering Dopamine was all about a paper that showed exactly that function of dopamine, so yeah I definitely agree. In fact, I'm pretty into to the idea that pretty much every neurotransmitter system is either (1) passing on error signal or (2) modulating gain on those signals.
Andy Clark makes me think that the nudge here might not be in the way that we expect. He points to data from the force-feedback stuff mentioned above, plus smooth eye pursuit studies (Levy et al. 2010 for a review), and suggests that perhaps the fundamental problem with precision weighting in schizophrenics is that there is a "weakening of the influence of prior expectations relative to the current sensory evidence." The idea being that "normal" priors end up being cast aside for "delusional" priors, which are intransigent because they offer explanations for almost every conceivable experience. (If it turns out that the GLP1RAs don't do anything for people who are already psychotic, maybe this is why. Maybe this is why early intervention doesn't work?!)
This makes me wonder if the nudge that the GLP1RAs are giving to the system is something unintuitive like making top-down models *more* sticky rather than less. I think this makes sense when you realize that a prior like "I'm hungry" doesn't specify *how* hungry, and it would make sense that there is a huge range of "I'm hungry" priors that differ in the intensity of the top-down prediction. Maybe GLP1RAs make it harder to switch from a low-intensity "I'm hungry" prior, to a higher intensity one?
Yeah I suppose I'd be open to the idea that the specific relationship is unintuitive like that given how whacky bits of predictive processing are, like the stuff on movement makes sense once you ponder it for a bit but it just conceptually sounds so weird and backwards. If it's something like your top-down model "forgetting" which is the actually sensible "I'm hungry" prior in the space of hungry priors that sorta fits too, though if you squint that still kind of looks like it's restating "error of prediction confidence", hmm.
Oh it's prediction error all the way down! In this instance, it's just a more specific comment on what side of the top-down/bottom-up the adjustment is working on.
This is a fascinating article. Predictive processing is new to me--I just learned about it here--I really like the concept, I want to learn more, but I'm still trying to wrap my head around it. I have two thoughts for now:
Would the predictive processing framework predict the feelings of nausea that people on GLP-1 agonists report? My guess is no? So, then GLP-1 would be doing something extra in addition to possibly changing the error signals as you suspect.
My second thought is that there seems to be a pretty straightforward way to test your hypothesis. Just do the external force finger pressure test on a case-control of people on GLP-1. I doubt anyone would do this experiment, but it's pretty cool that it's possible to test.
The predictive processing framework is basically a proposal for how the entire nervous system works: that there are top-down models that exist at virtually every level of processing that are trying to accurately predict the incoming bottom-up signals, and the goal of the system overall is to minimize the error signals.
I don't see why the experience of nausea with the GLP-1s couldn't be incorporated into that framework. It seems like they are able to adjust the sensation of fullness, which in PP terms could be thought of as the result of a top-down model that is integrating a bunch of bottom up signals about how much food is in the stomach (e.g. levels of leptin, ghrelin, NPY, etc.). In this instance, you might conceptualize nausea as a sign of error mismatch similar to hunger, where there is a strong preference in the system to resolve the mismatch by changing the bottom-up signal instead of the top-down model.
To bring the GLP-1s into this conjecture, one way that they might act in this system is to increase the strength of one (or many) of those bottom-up signals. Again, there are many ways by which it might do this. Maybe it directly reduces the release of ghrelin, but it could also reduce the strength of the signal produced by the neurons that sense ghrelin.
If you find this interesting I really can't recommend Surfing Uncertainty enough! I am probably not giving a totally accurate account of all of these ideas.
Agree that the finger pressure test would be very interesting to do, seems like it would be pretty low cost, too.
To clarify my point about nausea, I think that nausea not only suppresses the desire to eat, but also increases the desire to take a different action: to vomit. And vomiting also happens more frequently in people who use GLP-1s.
GLP-1 isn’t turning down the gain on that error signal! It’s turning it up!
I see this as a direct counterexample to the hypothesis in your article: "GLP-1RAs help with turning down the gain on these interoceptive error signals".
(sorry, I should've been more clear in my first comment. When I asked "would the predictive processing framework predict..." I should have asked "would that hypothesis specifically predict...")
Again, I still very much like your article, and I'm looking forward to reading that book! Thank you for introducing me to the concept!
Ah, ok, yeah I missed the point that you were making with the nausea then. It's a good counterexample; I definitely oversimplified for the sake of speculation here.
This is a fascinating theory, thank you. You conclude that the effect may be driven by reduced creation of misaligned models. Would you then expect the effects to continue post cessation of GLP-1? Or are you imagining that these models are re-created so frequently that this wouldn't persist? I suppose if the positive effects appear quickly then we would assume the models are updated frequently or that there are modulation levers being tweaked.
Most people who discontinue GLP1s have weight regain and some describe feeling more hungry than they ever have before (presumably because their actual weight is so far from an internal set point), *however* Ive seen data showing that people who begin at the highest BMIs are the most likely to continue losing weight post discontinuation-- this might align with your theory of model creation as a driver of disregulation and then re-regulation.
I also wanted to mention that in addiction, GLP1s have shown clear reductions in consumption and harms among people who already have SUDs-- the effect is not just preventative. (GLP1s for addiction is the core focus of my substack and organization, btw.)
My explanation hand-waves a lot of complicated theory away so it's hard to provide a better explanation without giving a deeper explanation of Predictive Processing. Reducing creation of misaligned models doesn't really give us any clues into how sticky any improvements from the GLP1s would be since there are multiple avenues by which you could achieve improved alignment.
I've elided a lot of the complicated theory here, but it might be helpful to know that top-down models exist at every level of the processing hierarchy (we see evidence of top-down models just barely upstream from the basic sensory cells themselves!). It's not just one big model sitting at the top of the processing hierarchy.
I would guess that there is a lot of heterogeneity in terms of where the "problem" is in the processing hierarchy, and that itself would have significant influence on whether or not the activity of the GLP1s allows for "retraining" of an individual's network.
I'm glad to hear confirmation that the GLP1s are effective in active SUDs. Thanks for turning me on to your blog!
I’m interested in an integrated and imaginative approach to treating psychosis and schizophrenia which is free from a world view wedded to one type of approach. I do agree that it includes a mismatch between interior bottom up signals and top down ones. These signals contain hundreds of neurotransmitters and millions of neurons and their glial cells, so it has a chemical level component. It can also be described at the functional or cognitive level (the word cognitive I find very steampunk to be honest with you). Many neuroleptic drugs have helped people live normal lives when they stay on them. These people in recovery then move away from the healthcare system and are no longer labelled as schizophrenic. This creates a smaller and smaller group that can create the illusion that the condition cannot be cured. Most relapses occur when people come off their meds because they feel well and don’t think they need them anymore. Many people who are labelled “treatment resistant” simply aren’t taking the medication and not telling anyone that they aren’t. They behave like this because they don’t trust authority figures. Kindness and compassion are surprisingly effective in these situations. Psychotics respond really well to compassion, even if they are unable to articulate this at the time. Your research sounds really interesting and I urge you to do a study if you are able. For the human and psychological component of recovery, do have a look at some of our blogs on here for a window into our lives as people, not patients ❤️
Hi Sarah - thanks so much for reading and your thoughtful comment. I took a look through your blog and enjoyed it very much, thank you for sharing!
I agree that there are many patients who are labeled as "treatment resistant" who are simply not taking their medications, and that there are many who are able to live normal lives on antipsychotics. That said, I think the scientific literature shows pretty clearly that there is a large proportion of individuals with schizophrenia who will only get a modest benefit from antipsychotics, and a good chunk who really are treatment resistant.
More importantly, though, I agree even more with your point that kindness, compassion, and empathy are important tools in these situations. Learning how to empathize and form an alliance with a psychotic individual, so that we can work together to make their lives more worth living, is one of the hardest but most rewarding parts of my job.
Thank you for replying. I’m trying to make some constructive connections between psychiatrists and people in recovery outside the confines of the doctor and patient relationship. I like Substack and feel it’s a good platform to achieve this while avoiding stressful chat room rants back and forth. I can also recommend Mat’s Blog on Schizoaffective Disorder and The Manic Messiah. Nathan Harmon has just started his. I’m building working support between us for peer support but also to access good advice and best practice. I’m aware that you are probably too busy to be across what we are all writing about, but I just want you to know about us in case we can help people with their recovery after they are discharged
Great find in that Xie article. I've been using tirzepatide extensively in pts on clozapine and anecdotally I see improvement in schizophrenia symptoms as well. Of course to justify to insurance they have to have diabetes, BMI, etc etc, but who on long term clozapine maintenance doesn't have severe metabolic syndrome really. I really want to run a small trial just adding it as an adjunct.
One of the saddest things about prescribing antipsychotics to me is that Im basically taking 15 years off a persons life because of metabolic syndrome. Really excited about the potential of these agents to mitigate that.
Note that if you do start to use the GLP1s the data we do have show tirzepatide is the far more effective agent than semaglutide and with a more favorable side effect profile. I don't use semaglutide anymore and its not because im an eli lilly shill, its because the data are pretty conclusive. Unfortunately I still see a lot of physicians just instinctively going for the semaglutide because it came out first I think.
In general I think your theory is cool but idgaf. Give me a randomized trial, it works or it doesn't regardless of mechanism.
Also, I don't think you should be so gloomy about prescribing antipsychotics, at least for schizophrenics. There are many studies that show that all-cause mortality is improved for patients with schizophrenia on antipsychotics vs. not, which wouldn't be the case if metabolic syndromes were actually knocking off 15 years.
Though maybe your point is that in the absence of metabolic effects those patients would also get those 15 years back? I don't think the evidence actually supports this. If you look at Figure 4 of this paper (https://pmc.ncbi.nlm.nih.gov/articles/PMC6953552/), there's no evidence that the more metabolically active antipsychotics reduce all-cause mortality less relative to less metabolically active ones. In fact, clozapine and olanzapine are two of the best at reducing all-cause mortality, outperforming risperidone.
I was being a bit facetious with the 15 years
Hmmm very interesting doc, very interesting, I will look into this more although I dont buy the its because they have better adherence to healthy lifestyle and medical care explanation, I think that data is confounded im just not sure how
I think the issue might be the follow up period still isnt long enough at 20 years, like is whats killing schizophrenics really being captured? gotta crack the hood on this one
I mean, average age at entry was ~45 and average follow up was 14 years. If >50% of your study population is being followed into their late 50s, that's pretty much hitting the average life expectancy for individuals with schizophrenia, so I don't think you can say that the issue is that follow up wasn't long enough.
I've had this conversation with a few people and they make the point that this can't possibly be true and that the data must be confounded, but also nobody has any good explanations for why the populations on risperidone and olanzapine should be all that different.
That's all to say that I'm confused by it too! Let me know if you find anything that makes sense.
"I think your theory is cool but idgaf" is generally the same sentiment I operate under! Very cool to hear that there might be something there in actually psychotic patients, but like you said, we need some trials. Maybe in a decade or so
I was sort of trying to parse this and started by thinking I'd maybe found a flaw when I think I probably just found another way to frame it that ends up making it plainly consistent. Took a quick look at Reconsidering Dopamine to see if you agreed, but I've seen Scott Alexander theorize elsewhere also that dopamine acts as an element of predictive confidence, so its involvement in schizophrenia may be related to overconfidence in low-quality predictions that overcook-the-books on the sense data to serve up conscious hallucinations.
If we start by suggesting that GLP1RAs might modulate surprisal by nudging prediction vs sense data into baseline alignment, for example by working for food/crack appetite by decreasing predictive confidence in "we should really be hungry/jonesing now" versus the sensory-received "nah we're actually good here", we could see a possible, straightforward similar relationship between decreasing schizophrenic-style broad severe errors of predictive confidence versus the normal generic sensory data.
Reconsidering Dopamine was all about a paper that showed exactly that function of dopamine, so yeah I definitely agree. In fact, I'm pretty into to the idea that pretty much every neurotransmitter system is either (1) passing on error signal or (2) modulating gain on those signals.
Andy Clark makes me think that the nudge here might not be in the way that we expect. He points to data from the force-feedback stuff mentioned above, plus smooth eye pursuit studies (Levy et al. 2010 for a review), and suggests that perhaps the fundamental problem with precision weighting in schizophrenics is that there is a "weakening of the influence of prior expectations relative to the current sensory evidence." The idea being that "normal" priors end up being cast aside for "delusional" priors, which are intransigent because they offer explanations for almost every conceivable experience. (If it turns out that the GLP1RAs don't do anything for people who are already psychotic, maybe this is why. Maybe this is why early intervention doesn't work?!)
This makes me wonder if the nudge that the GLP1RAs are giving to the system is something unintuitive like making top-down models *more* sticky rather than less. I think this makes sense when you realize that a prior like "I'm hungry" doesn't specify *how* hungry, and it would make sense that there is a huge range of "I'm hungry" priors that differ in the intensity of the top-down prediction. Maybe GLP1RAs make it harder to switch from a low-intensity "I'm hungry" prior, to a higher intensity one?
Yeah I suppose I'd be open to the idea that the specific relationship is unintuitive like that given how whacky bits of predictive processing are, like the stuff on movement makes sense once you ponder it for a bit but it just conceptually sounds so weird and backwards. If it's something like your top-down model "forgetting" which is the actually sensible "I'm hungry" prior in the space of hungry priors that sorta fits too, though if you squint that still kind of looks like it's restating "error of prediction confidence", hmm.
Oh it's prediction error all the way down! In this instance, it's just a more specific comment on what side of the top-down/bottom-up the adjustment is working on.
This is a fascinating article. Predictive processing is new to me--I just learned about it here--I really like the concept, I want to learn more, but I'm still trying to wrap my head around it. I have two thoughts for now:
Would the predictive processing framework predict the feelings of nausea that people on GLP-1 agonists report? My guess is no? So, then GLP-1 would be doing something extra in addition to possibly changing the error signals as you suspect.
My second thought is that there seems to be a pretty straightforward way to test your hypothesis. Just do the external force finger pressure test on a case-control of people on GLP-1. I doubt anyone would do this experiment, but it's pretty cool that it's possible to test.
The predictive processing framework is basically a proposal for how the entire nervous system works: that there are top-down models that exist at virtually every level of processing that are trying to accurately predict the incoming bottom-up signals, and the goal of the system overall is to minimize the error signals.
I don't see why the experience of nausea with the GLP-1s couldn't be incorporated into that framework. It seems like they are able to adjust the sensation of fullness, which in PP terms could be thought of as the result of a top-down model that is integrating a bunch of bottom up signals about how much food is in the stomach (e.g. levels of leptin, ghrelin, NPY, etc.). In this instance, you might conceptualize nausea as a sign of error mismatch similar to hunger, where there is a strong preference in the system to resolve the mismatch by changing the bottom-up signal instead of the top-down model.
To bring the GLP-1s into this conjecture, one way that they might act in this system is to increase the strength of one (or many) of those bottom-up signals. Again, there are many ways by which it might do this. Maybe it directly reduces the release of ghrelin, but it could also reduce the strength of the signal produced by the neurons that sense ghrelin.
If you find this interesting I really can't recommend Surfing Uncertainty enough! I am probably not giving a totally accurate account of all of these ideas.
Agree that the finger pressure test would be very interesting to do, seems like it would be pretty low cost, too.
Thanks for reading!
To clarify my point about nausea, I think that nausea not only suppresses the desire to eat, but also increases the desire to take a different action: to vomit. And vomiting also happens more frequently in people who use GLP-1s.
GLP-1 isn’t turning down the gain on that error signal! It’s turning it up!
I see this as a direct counterexample to the hypothesis in your article: "GLP-1RAs help with turning down the gain on these interoceptive error signals".
(sorry, I should've been more clear in my first comment. When I asked "would the predictive processing framework predict..." I should have asked "would that hypothesis specifically predict...")
Again, I still very much like your article, and I'm looking forward to reading that book! Thank you for introducing me to the concept!
Ah, ok, yeah I missed the point that you were making with the nausea then. It's a good counterexample; I definitely oversimplified for the sake of speculation here.
This is a fascinating theory, thank you. You conclude that the effect may be driven by reduced creation of misaligned models. Would you then expect the effects to continue post cessation of GLP-1? Or are you imagining that these models are re-created so frequently that this wouldn't persist? I suppose if the positive effects appear quickly then we would assume the models are updated frequently or that there are modulation levers being tweaked.
Most people who discontinue GLP1s have weight regain and some describe feeling more hungry than they ever have before (presumably because their actual weight is so far from an internal set point), *however* Ive seen data showing that people who begin at the highest BMIs are the most likely to continue losing weight post discontinuation-- this might align with your theory of model creation as a driver of disregulation and then re-regulation.
I also wanted to mention that in addiction, GLP1s have shown clear reductions in consumption and harms among people who already have SUDs-- the effect is not just preventative. (GLP1s for addiction is the core focus of my substack and organization, btw.)
My explanation hand-waves a lot of complicated theory away so it's hard to provide a better explanation without giving a deeper explanation of Predictive Processing. Reducing creation of misaligned models doesn't really give us any clues into how sticky any improvements from the GLP1s would be since there are multiple avenues by which you could achieve improved alignment.
I've elided a lot of the complicated theory here, but it might be helpful to know that top-down models exist at every level of the processing hierarchy (we see evidence of top-down models just barely upstream from the basic sensory cells themselves!). It's not just one big model sitting at the top of the processing hierarchy.
I would guess that there is a lot of heterogeneity in terms of where the "problem" is in the processing hierarchy, and that itself would have significant influence on whether or not the activity of the GLP1s allows for "retraining" of an individual's network.
I'm glad to hear confirmation that the GLP1s are effective in active SUDs. Thanks for turning me on to your blog!
Great points, and thank you! A bunch of big news will be arriving this year on GLP-1s and addiction. Hopefully some of it coming from us (CASPR.org)