Hackett Award: How accurate do suicide risk prediction algorithms need to be?

Suicide is the 10th most common cause of death in the United States; despite substantial efforts to prevent suicide, suicide rates have not fallen over the past 15 years. One challenge is accurately identifying the patients who are at highest risk of suicide: fewer than half of people who die by suicide have a diagnosed mental health condition, and only one-third are connected with specialized mental health care. To address this challenge, increasingly sophisticated algorithms are being developed to predict which patients are at high risk of suicide, but it's not known how accurate these models need to be before they're worth using in clinical practice.

This talk will describe a recently published cost-effectiveness analysis which answers this question. We will start by reviewing the theory and principles of health care cost-effectiveness analysis, and then review the published literature on the health and economic consequences of suicide attempts in the United States. Finally, we will discuss the implications of our findings for implementation of suicide risk prediction algorithms in real-world clinical practice.