Another lesson on basic (suicide) statistics.

Self harm is a risk factor for suicide. A recent paper using Medicaid records notes that such people are at an almost forty fold risk of suicide, and this is worse for adults. The more idiotic journalists will talk about a 3700% increase. to a risk of 0.4 percent.

During the first year following the index self-harm event, the suicide rate for the follow-up cohort was 439.1 per 100,000 person-years. The suicide rate of the age-, sex-, and race/ethnicity-matched U.S. general population was 11.8 per 100,000 person-years. The 1-year suicide rate for the follow-up cohort was therefore 37 times greater than the suicide rate in the general population.

A more interesting idea is to consider the number needed to detain. THis is the inverse of the base rate, which is 1/0.0004391. or 227. You would have to intervene or detain over two hundred people to save one life. About twice as bad as the clinical lore I was taught: that one in a hundred who self harm will suicide.

Suicide is rare, and that rarity confounds most secondary intervention.

3 thoughts on “Another lesson on basic (suicide) statistics.

  1. Yes. The efforts in the military to “prevent” suicide drive me crazy. From an epistemological standpoint, the only way you can know if you “prevented” a suicide is if you had a time machine. You go back in time and withold the intervention and see if they die.

    Self-harm maps pretty nicely onto Joiners “acquired ability to commit suicide” factor which is a powerful predictor. But his model requires the other 2 factors (thwarted belongingness and perceived burdensomness) to also be pegged in order for the ingredients of a suicide to be present.

    In the end though the data says this: the probability that someone who is imminently about to commit suicide is going to walk into your office and tell you about (or you will detect it) approaches zero in most settings.

  2. The VA/DOD clinical practice guidelines in conjunction with the self-directed violence classification system

    https://www.mirecc.va.gov/visn19/docs/Clinical_tool.pdf

    Is at least a step in the right direction. But more than standardizing and operationalizing these terms, what needs to happen next is weighted values for these SDV events needs to be established.

    What good is it for me to know to use the phrase “suicidal ideation with undetermined suicidal intent” versus “non-suicidal SDV without injury interrupted by self” if I have no idea which one is a more dangerous or more powerful predictor?

Comments are closed.