Tuesday, 10 May 2016
Working out who’s gonna die. Or why suicide risk assessment is a waste of time.
This post is by Chris Ryan (pictured above), a psychiatrist and Clinical Associate Professor with the University of Sydney and its Centre for Values Ethics and the Law in Medicine. Though primarily a clinician he maintains an active research agenda focusing on issues at the interface of ethics, law, and psychiatry. In this post he writes about his recent work on risk assessment for suicide.
Imagine you are me – a psychiatrist working in a hospital with a large emergency department. This morning, like most mornings, you arrive at work to find that ten people have been seen overnight after presenting in some sort of psychological crisis. Many have attempted suicide. How do you work out who should be admitted to hospital and who should go home? On what basis should you make that call?
Here is one tempting answer: admit the people that are at the highest risk of actually killing themselves in the future. If this strikes you as a sensible approach, you’re in good company – indeed many psychiatrists are inclined to that opinion. In fact ‘suicide risk assessment’, often aided with a series of checkbox tools, is often seen as a mandatory component of proper clinical care.
If, on the other hand, you think that trying to categorise patients into those at relatively high and low risk of future suicide is a fool’s errand, you are in even better company – I couldn’t agree more.
Since many people have difficulty seeing why this is so, let me spell out the reasoning.
It’s all about the math. Risk is probability multiplied by loss. If we are talking about suicide risk assessment then the magnitude of loss is held more or less constant. Though there will be exceptions to this, most people will agree that most suicides will be a very significant loss, so significant they can be treated as more or less of the same magnitude of badness. If we are prepared to except that the loss is a constant, then all we are doing when we do a suicide risk assessment is trying to gauge the probability of this person suiciding with a certain period T.
If you want to put your emergency department patients’ probability of suicide into context, there’s good news and bad news. The bad news is that, as a class, your ten patients are at a greatly increased probability of suicide at some point in the future compared to people that have never presented to a emergency department in a psychiatric crisis. Studies vary, but it is reasonable to regard the population of such patients at about 30 times the population-likelihood of eventual suicide. The good news is that each individual patient is at quite a low absolute probability of suicide. Though the probability will vary as T varies, and again empirical studies vary, its probably reasonable to say that, as a class, such patients have about a 1% likelihood of dying by suicide within twelve months and about a 1 in 20,000 probability of dying by suicide tomorrow.
If you were a fan of the risk assessment approach to this dilemma, you probably opted for it because you thought that there might be a useful way of determining which of the ten patients (who as a group are all at very high relative risk and very low absolute risk) were more ‘at risk’ than others. You probably reasoned ‘there must be a factor (or combination of factors) that allow one to usefully categorise these patients by their likelihood’. If you did think that, it was a reasonable first pass assumption, but it was wrong.
There are now heaps of studies demonstrating that this is wrong and my group has done many of them, but, the truth is, we knew that it would be wrong before we even started. Here’s why.
Consider suicidal ideation, might not that be a great candidate as a differentiating factor. Psychiatrists often mistakenly assume that if a patient in this situation is, or was recently, suicidal this state would increase their likelihood of future suicide. In fact though suicidal ideation is no help at all.
It won’t surprise you to learn that almost of all of your ten patients in psychiatric crisis will be, or will recently have been, suicidal, but, as we know, hardly any of them will go onto kill themselves. It was never going to be likely that a risk factor present in almost everyone in a group, was going to be useful in predicting something that almost no one in the group would do. You might be tempted to say, ‘well perhaps not suicidal ideation alone. What if it were combined with depression, despair, hopelessness etc etc?’ Again this is common response, but a moment’s reflection will having you guessing what studies show to be true – very few people who are suicidal are not depressed, despairing, or hopeless. All these factors co-vary enormously. While it is probably possible to use a range of factors to categorise patients presenting in acute psychiatric crisis into those who are statistically significantly more likely to kill themselves in the future, given that all such patients are at greatly elevated relative risk of suicide, the meagre additional risk posed to the so-called “high-risk” group is too slight to allow any useful management decisions to hinge on the categorisation.
Suicide risk assessment in this patient group is a waste of time.
‘Fair enough’, I hear you say, ‘How do I decide who is coming into hospital and who is going home?’
To that I’d say two things. First, you are asking the wrong question, and, second, the answer to the right question will require a whole ‘nother blog.
Large, M. and Ryan, C. 2014: 'Suicide risk assessment: myth and reality'. International Journal of Clinical Practice. Vol. 68, no. 6, pp. 679–81.
Large, M. and Ryan, C. 2014: 'Suicide risk categorisation of psychiatric inpatients: what it might mean and why it is of no use'. Australasian Psychiatry. Vol. 22, no. 4, pp. 390–2.
Ryan, C. 2015: 'Suicide explained!' Australian & New Zealand Journal of Psychiatry. Vol. 49, no. 1, pp. 83–4.
Ryan, C., Large, M., Gribble, R. et al. 2015: 'Assessing and managing suicidal patients in the emergency department'. Australasian Psychiatry. Vol. 23, no. 5, pp. 513–15.