> In a population of 10,000 individuals this model would catch only 44 true positives, but 365 false positives. That’s ~8 false positives for every true positive.
Thanks for bringing this part up. As someone who doesn't get into statistical weeds often enough, I forget about these things. I do wonder what the equivalent screening specificity for EASA programs is. I didn't read the study, but maybe the ML program is trying to identify clients before any symptoms and EASA screenings are higher specificity due to waiting for actual symptoms of psychosis. IDK.
I don't have access to the AN capacity assessment study, but that paper, and your comment on the complexities of capacity assessments, is intriguing to me. If you are ever feeling the itch to write a digestible synopsis on those complexities, know you will have at least one reader!
> My first instinct is to say that there is some hidden variable that correlates strongly with likelihood of being on an SNRI, but I can’t think of anything all that obvious
Could it just be that clinicians see SNRIs as second line therapies, so patients on them generally have more severe / treatment resistant depression, and ketamine has larger effect sizes in more severe depression / TRD?
I'm surprised the ERB allowed a trial to give depression patients a inflammatory drug which makes their depression worse! I would have assumed stuff like that would only be for animal models.
Thanks for this. Perfect journal club material while I’m away from the office.
LSD for ADHD? You’d really have to misunderstand both conditions to think that could work. What is interesting, though, is whether microdosing might have an effect on dementia. Could all that supposed “neurogenesis” actually matter? And does it even happen with microdosing?
The ketamine + SNRI finding is genuinely surprising. That said, I’ve grown increasingly skeptical of ketamine for depression in general. Something feels off, and I admit that may be my bias as an addiction psychiatrist. I see too many cases where long-term ketamine use makes psychiatric symptoms much worse, in ways I’ve never seen with SSRI/SNRI or ECT. My hunch is that the biggest mistake was stripping ketamine from the “psychedelic-assisted psychotherapy” framework and turning it into something clinics could churn out more easily.
I might've thought "Predicting Diagnostic Progression to Schizophrenia or Bipolar Disorder via Machine Learning" was about LLMs causing schizophrenia or bipolar, which would be fairly relevant now but I guess is a new enough phenomenon that it's not especially likely a paper for it would exist yet.
There's a few ways the telehealth thing could be relevant, eg if we thought it might be responsible for a COVID-era uptick in drug abuse. "It might be easier for people to get drugs if they're facilitated by apps" is a quaint idea made politically expedient if all this can be blamed on the tech sector, but in my experience the people who abuse drugs have no trouble finding apathetic doctors in person so there's not really a result that would surprise me here.
The clozapine REMS change was an insane personal blessing because my current PIC is a bit of a luddite who had been dragging his feet on getting me access at our hospital for over half a year AND for whatever reason was requiring us to manually track everything by hand in a thick binder despite us already having EMR software and a clinicals tracking website on top of the damn REMS website grumble grumble. Pharmacy handled that at the state prison I was contracted with and I'm sure they're happy to have less work for free too.
I don't think it's a good sign that the biggest study in the MDD group showed an improvement with 30 units, but not 50 units relative to placebo. That makes no sense whatsoever. Maybe there's something here, but I think this evidence is pretty weak.
> In a population of 10,000 individuals this model would catch only 44 true positives, but 365 false positives. That’s ~8 false positives for every true positive.
Thanks for bringing this part up. As someone who doesn't get into statistical weeds often enough, I forget about these things. I do wonder what the equivalent screening specificity for EASA programs is. I didn't read the study, but maybe the ML program is trying to identify clients before any symptoms and EASA screenings are higher specificity due to waiting for actual symptoms of psychosis. IDK.
I don't have access to the AN capacity assessment study, but that paper, and your comment on the complexities of capacity assessments, is intriguing to me. If you are ever feeling the itch to write a digestible synopsis on those complexities, know you will have at least one reader!
> My first instinct is to say that there is some hidden variable that correlates strongly with likelihood of being on an SNRI, but I can’t think of anything all that obvious
Could it just be that clinicians see SNRIs as second line therapies, so patients on them generally have more severe / treatment resistant depression, and ketamine has larger effect sizes in more severe depression / TRD?
I'm surprised the ERB allowed a trial to give depression patients a inflammatory drug which makes their depression worse! I would have assumed stuff like that would only be for animal models.
Thanks for this. Perfect journal club material while I’m away from the office.
LSD for ADHD? You’d really have to misunderstand both conditions to think that could work. What is interesting, though, is whether microdosing might have an effect on dementia. Could all that supposed “neurogenesis” actually matter? And does it even happen with microdosing?
The ketamine + SNRI finding is genuinely surprising. That said, I’ve grown increasingly skeptical of ketamine for depression in general. Something feels off, and I admit that may be my bias as an addiction psychiatrist. I see too many cases where long-term ketamine use makes psychiatric symptoms much worse, in ways I’ve never seen with SSRI/SNRI or ECT. My hunch is that the biggest mistake was stripping ketamine from the “psychedelic-assisted psychotherapy” framework and turning it into something clinics could churn out more easily.
I might've thought "Predicting Diagnostic Progression to Schizophrenia or Bipolar Disorder via Machine Learning" was about LLMs causing schizophrenia or bipolar, which would be fairly relevant now but I guess is a new enough phenomenon that it's not especially likely a paper for it would exist yet.
There's a few ways the telehealth thing could be relevant, eg if we thought it might be responsible for a COVID-era uptick in drug abuse. "It might be easier for people to get drugs if they're facilitated by apps" is a quaint idea made politically expedient if all this can be blamed on the tech sector, but in my experience the people who abuse drugs have no trouble finding apathetic doctors in person so there's not really a result that would surprise me here.
The clozapine REMS change was an insane personal blessing because my current PIC is a bit of a luddite who had been dragging his feet on getting me access at our hospital for over half a year AND for whatever reason was requiring us to manually track everything by hand in a thick binder despite us already having EMR software and a clinicals tracking website on top of the damn REMS website grumble grumble. Pharmacy handled that at the state prison I was contracted with and I'm sure they're happy to have less work for free too.
What’s happening with Botox for depression? I feel like it’s been in the small trial stage for 15 years or more now and yet it keeps showing promise.
The whole "botox improves depression by limiting frowning" just has never passed the smell test for me. I haven't looked into it closely.
Indications of efficacy + minimal side effects (especially compared with existing medications) sound like a good combination to me
https://www.mdpi.com/2072-6651/16/4/191
This still leaves me very unimpressed.
I don't think it's a good sign that the biggest study in the MDD group showed an improvement with 30 units, but not 50 units relative to placebo. That makes no sense whatsoever. Maybe there's something here, but I think this evidence is pretty weak.