Hacker News Reader

The article mentions sensitivity, but I’m also curious about this method’s specificity too…

Yes it sounds like people came in at high risk of suicide and then they gave them an app that surveyed them and would predict they were going to commit suicide. Given they were at high risk for suicide attempts, how does this compare to just predicting the people would always try to commit suicide in the following week?

Suicide attempt PPV = .16. This means, and correct me if I'm wrong, 84% of positive results would be wrong if the test was applied to the general population (population of psychiatric patients?).

> The most accurate prediction was achieved using bidirectional long short-term memory and simple lasso-penalized logistic regression models, with the best performing model using bidirectional long short-term memory to predict SRE, which with specificity at .90, had area under the curve = .94, sensitivity = .87, and positive predictive value = .30, and SAs with area under the curve = .90, sensitivity = .74, positive predictive value = .16.

Where SA is "Suicide Attempt" and SRE is "suicide-related event".

https://psycnet.apa.org/doiLanding?doi=10.1037%2Fabn0001117

I also wonder if having to answer this daily annoying survey a actually make you more wanting to kill yourself...

Reply on news.ycombinator.com