Can predictive tools tell us who will benefit from antidepressants?

I'm just a layperson with strong, strong thoughts about this. So... take it with a grain of salt.

  a U.S. push for “deprescribing”

Always fun to remember that the person in charge of the federal health system isn't a fan of drugs based on vibez. Still waiting to be but in a "camp" to get off my prescribed adderal, as he casually mused in 2025...

  Doctors and researchers know antidepressants don’t work for everyone, and some even agree the drugs are overprescribed.

Kinda absurd framing, even if it's followed by a semi-related "But...". Some scientists doctors and researchers agree that climate change isn't real, that ADHD doesn't exist in adults, and that v-cc---s are a scam cooked up by aliens; you wouldn't so casually make those seem like reasonable positions!

  finding the right treatment is often a frustrating matter of trial and error

That is the very intentional shape of modern psychiatric care, and will be forever until we can gradually decompose each psychiatric disorder into one or more neurological diseases with thoroughly-specified physical characteristics. In other words, it will and should be like this until we have a dramatic scientific revolution in neuroscience.

  One method for doing so, based on brain imaging, is already moving toward approval from the U.S. Food and Drug Administration (FDA) as a tool to guide prescribing.

Presumably they mean EEG, here...? We already have tons of neurocognitive tests that "guide prescribing", but adding a new tech couldn't hurt, especially as EEG & fMRI explode in utility thanks to DL. Still, frustrating that this is framed as an alternative or fundamentally new development, rather than just one more (very expensive) tool in a psychiatrist's repetoire.

   a statistical and AI tool called PETRUSHKA, which they trained on the sociodemographic details and health histories

Hoooooly shit. "Sociodemographic details"?! Did they miss every single controversy around ML in the 2010s, and every public health textbook since then? We've been down this road a dozen times in a dozen contexts; absurd that professionals are trying something so resoundingly harmful.

By the way, the study (linked in article) is shocking. The fact that at least two PhDs signed off on a study with such an obvious glaring issue ("lack of a double-blind design and the large amount of missing data limit the validity of these results") to get a paper out and promote their vibe-coded webapp is a massive indictment of our academic system. Or at least Britain's, I guess. In case it's not obvious: of course people who are told their antidepressants were asigned by a super intelligent machine based on a detailed quiz will be more patient with them than people who get the "usual standard of care", which, again, is very intentionally and openly "guess-and-check" and often opaque to the patient.

  He and his team have trained a machine learning model on patient records from a clinical trial and used it to predict who would do better on CBT or one of two antidepressants.

The next study is also frustrating :(. To cut it very very short: A) CBT vs. medication is not a coherent dichotomy, B) it's repeating very basic mistakes around unintentional discrimination as discussed above, C) the "machine learning model" is just linear regression (!!!), and D) their attempt to cross-validate has serious, serious issues.

  Other researchers are looking for clues to treatment response in the brain activity thought to underlie depression. “That’s really what we’re treating,” says Leanne Williams, a psychiatrist at Stanford University. Her research has identified six distinct “biotypes” of depression based on functional magnetic resonance imaging (fMRI).

Oh jesus christ. They don't teach philosophy of psychiatry at Stanford?? Does she know that she's departing from the sanctioned, licensed framework for all psychiatry? And why is a psychiatrist doing this instead of a neuroscientist, or ideally a specialist in neuropsych?

  The tool might also identify people who will likely never respond to either psychotherapy or first-line antidepressants, Williams says—potentially allowing them to skip straight to second-line therapies such as transcranial brain stimulation, electroconvulsive therapy, or ketamine.

Well that's fucking terrifying. "Whelp your history and race puts you under the curve in our linear regression model, so let's just skip the boring steps in the middle!" - a statement uttered by the deranged.

Sorry for the spam. Just scary and frustrating. I'm at least the author here is trying to frame this in a way that might redirect some of the unscientific, hateful, ignorant urges of the Health Czar.