Predicting suicide attempts and suicide

Predicting suicide attempts and Suicide is very difficult since they are relatively rare events. The majority of individuals with suicidal ideation will not make an attempt and the majority of individuals who make attempts will not die by suicide.

This presents challenges when predicting these outcomes. Scales and algorithms will often have low positive predictive value; most individuals identified as high risk will not go on to make a suicide attempt or die by suicide. A choice needs to be made between sensitivity and specificity.

Approaches §

Retrospective analysis §

Commonly used instruments in suicide risk do not have adequate sensitivity and specificity But the PHQ-9 is still decent: The PHQ-9 item on suicidality is predictive of suicide attempt or death

A controlled examination of acute warning signs for suicide attempts among hospitalized patients

Risk models §

Risk of death by suicide following self-harm presentations to healthcare

Machine learning §

• Machine learning-based prediction for self-harm and suicide attempts in adolescents
  • A random forest model achieved an AUC of 0.73 when predicting adolescent self-harm two years later
  • Random forest prediction of suicide attempts in adolescents reached an AUC of 0.72