Aaron (Tim) Beck, Harvey Resnik, and Dan Lettieri are the editors of The Prediction of Suicide.  The assembled work brings together the best minds in the prediction of suicide in 1974.  The arguments made then are like the arguments that could be made today.  In the preface, they state, “Despite the voluminous research reports, there is a very flimsy basis of knowledge that can contribute in a scientific sense to the problems of the worker in this field.”  It’s a challenge that hasn’t changed substantially in the fifty years since this publication – but hopefully it’s one that will change soon.

The Process

The point is made that, “Suicide is the end result of a process, not the process itself.”  This belies the problem of prediction and identification.  We speak of the outcome, but even today, we struggle to articulate the pathways that lead to this outcome.  It’s understanding these pathways that provides hope for our ability to do some level of prediction of suicide.

The one differentiation that can be made about the process – even in 1974 – is related to the outcomes.  “But the unsuccessful suicides are no doubt quite different from the successful, and the former cannot be regarded as representative of the latter.”  The categorization is that attempts must be categorically different than deaths, because the outcomes are different.

I think this hides the reality of the randomness to the process.  Silvia Plath arguably wanted to be found and her attempt to be aborted.  (See The Savage God and Suicide and Its Aftermath.)  Even though she eventually died, her process may have been closer to that of an attempter who didn’t die.  In short, while we can presume that there’s a difference between attempts and those who die, we can’t really know.

Zeigarnik

Blume Zeigarnik was a student and colleague of Kurt Lewin.  She noticed an odd thing about the memory of servers.  They could remember orders without writing them down – until they relayed the order to the kitchen.  After that point, they promptly forgot the order.  This led to the discovery of what we call the “Zeigarnik effect,” where uncompleted tasks are held more prominently in memory.

Joseph Subin, in the first chapter, hypothesizes that the Zeigarnik effect may have an influence on attempters, providing some subtle draw towards “completion.”

Call Centers

Suicide call centers are an important part of the overall system of care to try to prevent suicide, but the book notes that “only 4 percent of suicide attempts and even a smaller percent of the eventually successful suicides called suicide prevention centers.”  So, they’re an important part of the overall strategy – even if the overall match to those who make attempts is low.  We see this same sort of calling pattern in 988 today.

The Perception of Control

We often underestimate our need for the feeling of control.  The belief that someone has control and the presence of options has consistently demonstrated a positive effect on mood for people.  We see this in places where there are suicide options for those with terminal illnesses.  The number of people who use the suicide option after having been approved is very low.  (See November of the Soul.)  They’ll go through great lengths to acquire the ability to die by suicide – and simultaneously decide not to use the option.

At its heart is our perceptions of control.  When we feel we have control, we have a greater capacity for self-soothing and down-regulation of fears.

Mental Health Disorders

Mental health disorders are, for the most part, time-limited with or without therapy.  That’s striking, but not totally unexpected, news.  For most of human evolution, mental health disorders have occurred before the introduction of psychotherapies and the like.  This is not to say that mental health assistance is a bad thing – far from it.  Antibiotics, in most cases, merely decrease the time it takes to heal, but we still use them anyway.  Similarly, mental health supports are good things.  But understanding that mental health disorders typically self-resolve can help us to understand how suicidal crisis can also self-resolve.

To be clear, this is not to say that all mental health disorders will self-resolve – they won’t.  However, the argument made by Zubin is that they largely self-resolve.

Actuarial Versus Clinical

One of the big challenge in the prediction of suicide is the difference between aggregating various risk factors to develop a risk score for an individual and the need to sit next to someone and make a decision about whether they are a risk to themselves.  (Ideally, sit next to them rather than across from them, as still often happens – see Motivational Interviewing and Managing Suicidal Risk.)

In the development of actuarial risk, demographics and history are combined into a single score based on previous research and factors that can be identified to raise or lower the risk.  Being an “old white guy” raises one’s risk – my risk.  Other factors are loaded into the assessment to create a score.  However, this score has nothing to do with me personally and everything to do with the statistical abstractions made for groups of people.

Time and time again, we’ve demonstrated that such actuarial risk summarizations have almost no utility in the assessment of individuals.  Compiling the most comprehensive profile still doesn’t yield the ability to predict which individuals are at risk.  The statistical (actuarial) process simply has eliminated all of the distinctiveness in the data and with it the ability to see the risk of individuals.

Later in the volume, Beck states it clearly: “The belief that suicidal behaviors are predictable can be valid only as a belief in principle, not in fact.”

Psychological Autopsy

Even in 1974, the limitations of psychological autopsies was well known.  Alex Pokorny explains the difficulty of discovering intent: “It also appears to require a ‘psychological autopsy,’ which is not practical for general use and which also introduced the possibility of circular reasoning.”  He first identifies the effort and therefore cost of doing psychological autopsies.  They’re time consuming.  They require willing participants of the survivors, which isn’t always the case.  That makes them somewhat impractical for broad use.

The more challenging aspects of psychological autopsies are the problems of retrospective reasoning.  After a determination of suicide is reached (preliminarily), the scales tilt towards that, and there is some bias towards confirmatory evidence.  This is held back by the stigma and extra pain associated with suicide death, but the degree to which one of these forces is more powerful than another is both situational and effectively immeasurable.

We’re left with serious doubts about whether psychological autopsies create a real picture of the person’s mind or whether they create a fiction that roughly fits the facts.  This fiction may help us feel better about understanding – but it does not necessarily create actual understanding.

The Categories

One of the challenges of creating good research on suicide is the need for clear and consistent categories.  The categories proposed are completed suicide (CS), suicide attempt (SA), and suicide ideas (SI).  These are good, broad categories, but they miss some of the nuances and challenging situations.

In particular, non-suicidal self-injury (NSSI) is problematic in this framework.  There is a relationship between NSSI – particularly cutting – and later suicidal behavior, but the narrow and coarse framework proposed here doesn’t connect NSSI to suicide.

Screening

Aaron (Tim) Beck was one of the earliest proponents of finding scales to measure risk.  He was developing what became the first risk screening tools – some of which are still used today because of their efficacy.  However, he states, “Nevertheless, even the best of these produces a very high proportion of false positive errors, that is, cases that are unjustifiably labeled as high suicide risks.”  Later, he continues by saying, “For there is currently no detection scheme that can be set to identify half of the available genuine suicide risks without erroneously identifying along with them a lot of people who are not suicide risks at all.”  He acknowledges that because suicide is a statistically rare (and tragically too common) event, it’s hard to develop tools to identify it.

He argues that, in order to get sufficient sensitivity to detect people who may have suicide in their immediate future, many must be identified and later assessed out of the system.  My “back of the napkin” calculations put the false positive rate at about 300-600 times the number of actual positives based on current tooling.  Despite the insistence on the use of these tools, the behavioral health system can’t cope with the false positives that must be screened out.  Even if these clinical assessments were 100% accurate, the sheer volume of work puts a strain on an already burdened system.  The tragedy is that even clinician assessment is a poor predictor of outcomes, as is explained in The Practical Art of Suicide Assessment.

Predictors and Postdictors

Hindsight is 20-20.  It’s a common cliché that pushes us towards an understanding that we can see things in the past that might have never been identifiable before the event.  We can understand the factors and methods that lead to outcomes only after the kind of careful study and clarity that comes after the event.  One of my great frustrations is with lists of suicidal risks, because they include things that frequently occur, including in a proportion of those who attempt suicide.

Things like a change in mood or behavior is often listed.  The problem is that, when applied to teenagers, this is almost universal – with or without suicide risk.  Also listed are statistics like 95% of people with suicide have a mental illness.  That’s misleading, because a very small percentage of those with mental illness will die by suicide (<5%).

David Lester makes the point that what we call “predictors” are all assessed after-the-fact and therefore should more accurately be called “postdictors.”  They have little predictive value.  They do, however, encourage a great deal of guilt and shame on the part of loved ones who feel that they missed signs that they should have seen.

Infrequency

Chapter after chapter in the book has authors saying that suicide is a statistically rare event and is therefore nearly impossible to predict at an individual level.  George Murphy explains how a statistically good screener would be unacceptable clinically owing to the intersection of statistics and outcomes: “From the numerical standpoint, a prediction of ‘no suicide’ in every case would be highly accurate (1,336/1,350 x 100 = 98.96%). It would also be entirely unacceptable clinically.”

Extending out some basic math approaches, he concludes, “More to the point, the predictive accuracy assumed (80 percent) is far beyond our present capabilities. The population chosen for the example (suicide attempters) is one of relatively high risk, and yet prediction of the infrequent event, suicide, is poor. It would be very much poorer in a population unselected for risk.”  The threshold he used of 80% accuracy exceeds the capacity of our tooling even today, 50 years later.  Screening is still required by accrediting bodies in high – and not so high – risk situations despite our awareness that they simply aren’t effective.

The funniest thing is that the more we pay attention to the details, the more we can recognize that it’s a fool’s errand to believe in The Prediction of Suicide.