The assessment of suicide risk attracts a fair amount of controversy and polarises opinion.
There are some who see it as futile and a waste of valuable clinician time. Others view it as one of the most important, complex and difficult tasks in mental health. Machines can undertake routine tasks quickly and can make light work of complexity. Is machine learning the future of risk assessment?
In this talk I will begin by trying to understand what we mean by suicide risk assessment. I will discuss some of the challenges of traditional approaches to risk assessment and whether machine learning can overcome them. I will take a closer look at recent literature and consider how we might best use the vast computational powers at our disposal in the assessment and management of people who present with suicidal thoughts and ideas.
In the end suicide risk assessment means different things to different people. Everything from a thorough clinical assessment of a person who presents with suicidal thoughts or behaviours to an actuarial consideration of risk factors and their influence on outcome. Problems related to poor positive predictive values, the low risk paradox, and the exclusion of people from services are unlikely to be overcome with machine learning. Suicide risk prediction will remain a fallacy. But perhaps the machines might help us by allowing clinicians to do the simple things better – the accessibility and continuity of health information and knowledge support systems are two examples.