Episode 4 - Can we predict suicide?

So the topic for today is can we predict suicide? And this is something that I get asked to do almost daily as a clinician.

I'm told someone is potentially suicidal and, uh, called down to the unit to see, um, whether or not they will or will not and what we can do about it.

And prediction is one of those things that I think people expect out of suicide risk assessments and psychiatrists doing them.

But, um, prediction actually goes back for quite a while as having a, a poor track record.

And, um, my question for today is really trying to frame it on, has any of the recent advances in technology improved the detection algorithms to the degree that we can rely on them for prediction? It's always important to remember that prediction drives real decisions.

In 2026, in most environments, uh, a high rating of suicide leads to increased rates of admission, um, involuntary holds, medications being given with or without consent, um, all sorts of administrative decisions.

Uh, in the United States, it's required to do a suicide risk assessment on people for psychiatric admissions, um, which, uh, I'll comment on later.

Um, but um, this is why it's always so pertinent.

This is not just some academic exercise, who can we predict or not? Real world implications occur from prediction, and unfortunately, so much is, is put on a prediction.

And if the prediction actually doesn't have that much value, it makes you wonder like, why do administrators require suicide risk assessments if they're not that predictive? If prediction is of high risk or low risk is of no value, then, then why do we have a, a unit for high risk suicides? And why do we treat patients differently when they're deemed to be a high risk for suicide? So going back to the, the hook at the beginning, um, Alex Pokorny's study in the VA, um, followed these 4,800 patients for a number of years, and 17% of them were rated as high risk.

96.3% of them did not die by suicide over the multiple years that he followed.

And 37 of the 67 suicides came from the group that he had categorized using all of the studies, all of the tools, all of the risk prediction rubrics, the Beck's suicidality scale and hopelessness scale and all these things.

And his own conclusion was that identifying the individual is not feasible and the tools are just too limited given the low base rate of suicide.

Um- He basically said he could live with missing half the suicides if he could find the other half without all the false alarms.

But he couldn't find all the other half without all the false alarms.

He basically had 803 alarms and only 4% of that 803 ended up dying by suicide.

So needless to say, the false alarms were just reigning supreme.