The Truth About Predicting Suicide
What the most recent research tells us about suicide risk.
I was completing my Ph.D. and training at an inpatient psychiatric hospital in 2003. After completing an assessment on a young man with depression, it came time to make a decision — release him from hospital or deny his request and keep him for more observation and testing. My supervisor and I were in agreement — he was probably mentally well enough to return home. Within 48 hours of release he stepped in front of a train.
Nearly 14 years later and I can’t help but wonder whether the current version of me is capable of better predicting a suicide attempt.
The latest research makes me doubt that I am any more capable. My field of clinical psychology has a number of imperfections, but few are more glaring and frustrating than the inability to reliably predict suicide attempts and deaths.
Approximately 25 million people* annually try to end their lives prematurely — no statistic better reveals the truth that life is suffering. Of this massive group, an estimated one million will “succeed.” More than HIV/AIDS. More than war. More than car accidents. More than homicides.
Having thoughts about suicide (called ideation) is relatively common, as an estimated 140 million people contemplate it every year. In clinical practice, suicide ideation is very common and most of it is passive in nature. Passive suicide ideation refers to an individual thinking about death and suicide, but not seriously planning to do anything (ex: “death would be a relief”). Active suicide ideation refers to a person planning the means by which to end their life. The former is relatively common in therapy and fairly low risk, while the latter makes one pay more attention as the risk of an attempt is higher.
Or is it?
Suicide ideation is just one of dozens of suicide risk factors that come into play when deciding whether a person is at risk of making an attempt. There are a range of guidelines for health care workers to use for clinical decision-making, based on reviews and the opinion of expert panels. However, a comprehensive meta-analysis was just published in the top psychology journal — Psychological Bulletin — and surely it would provide a more quantitative guide to practitioners and concerned loved ones on what to look for and consider when someone is presenting as suicidal. Right?
Wrong. Unfortunately, the meta-analysis by Franklin and associates showed what seasoned practitioners have known for some time — any attempt to predict who will attempt and complete suicide falls far below the standard that should exist after over 50 years of suicide research.
Here’s what we know (and mostly don’t).
What is a Risk Factor?
The authors of the aforementioned analysis do an excellent job of setting the scope of what it is actually meant as a risk factor for suicidal thoughts and behaviour (STB).
To start, there is a difference between a correlate and a risk factor. A correlate is something that correlates with the occurrence of an STB. For example, having depression may be associated with the occurrence of an STB, but this doesn’t mean that depression and STBs are related in any meaningful way.
A risk factor is a correlate that provides more useful information. A risk factor is different from a correlate in that a risk factor is known to precede an STB. So, if a research study followed people over a period of time and found that depression tended to occur before an STB, this would make depression a risk factor because it is possible that there is now a causal relationship. It also allows us to start to categorize people as being low and high in risk for a particular issue, in this case STBs.
The only way to know whether a risk factor is causal (ie, depression causes suicide to occur sometime in the future), we would have to use an experimental design that manipulated the risk factor. For example, if we could cure depression in one group and let another group go untreated, we could then observe whether this causally influences STBs at a later time.
Unfortunately, there is not much known about causal risk factors, and so we are left speculating on which identified risk factors directly influence suicide risk.
In this study, the authors completed a comprehensive search for research articles on risk factors for STBs, and included only those using a longitudinal design (ie, one that follows people over time to see what happens to them). This allowed the researchers to separate correlates from risk factors. They settled on 365 papers that met certain quality criteria, drawing from research going back to 1965.
There were simply too many individual variables to assess and report on, so the researchers created 16 broad categories of risk, which were composed of various subcategories. For example, a ‘cognitive problems’ category was one of the 16 risk domains, and it was composed of subcategories like intelligence, problem-solving ability, school performance, etc.
The 16 categories of risk (with examples of subcategories) were:
1. Biology (ex: genes; hormones)
2. Screeners (specific suicide screening questionnaires)
3. Cognitive problems (ex: problem-solving ability)
4. Demographics (ex: age; gender)
5. Externalizing psychopathology (ex: aggressive behaviours; impulsivity)
6. Family History (ex: maternal depression; paternal alcoholism)
7. General Psychopathology (ex: presence of psychiatric symptoms or diagnoses)
8. Implicit/explicit processes (ex: implicit attention bias)
9. Internalizing psychopathology (ex: anxiety and mood disorders)
10. Personality traits (ex: Type A; extraversion)
11. Physical illness (ex: heart disease; weight)
12. Psychosis (ex: Schizophrenia)
13. Prior STBs (ex: prior self-harm; suicide attempt)
14. Social factors (ex: abuse history; isolation)
15. Exposure to STBS in others (ex: suicide attempt by friend)
16. Treatment history (ex: prior admission to psychiatric hospital)
The Results
Before discussing the results of arguably one of the most important research papers on suicide to be published in some time, we must first address the issue of “clinical significance.” Finding a statistically significant association between two variables, like depression and suicide, is not necessarily useful. It is the size or strength of the relationship that is important, especially in risk research. It is also very important to consider something called the base rate.
It is common for the media to report on associations among a whole host of variables, and most of it is garbage reporting because rarely do the media use the base rate to explain the meaning of the results. For example, it is common for people to hear that variable X (ex: eggs) increases one’s risk of disease Y (ex: cancer) by a certain amount (ex: 50%). This might sound alarming, but usually it is not because these results are relative risk statistics. Using our example, this would mean that your risk of getting cancer by eating eggs is 50% higher than someone who doesn’t eat eggs. This number (50%) is only useful if we know what the rate of cancer is among those not eating eggs (called the base rate).
If people who don’t eat eggs have a cancer rate of 1% per year, then eating eggs means your rate of getting cancer would be 1.5% per year (50% more than 1%). Hardly significant. The point here is that when it comes to understanding risk, we need to consider the base rate — which is the rate at which the disease/disorder/injury occurs for some comparison group.
Let’s use suicide to explain this further. As mentioned earlier, suicide affects millions of people every year. However, the actual rate of suicide is relatively low, as only 0.013 out of every 100 people die by suicide each year. This is the base rate for suicide in the general population. If suicide rates jumped by 200% (or 2xs the base rate) this would sound scary, but it would only mean that 0.026 people out of every 100 now die by suicide each year. This is important to know because it affects how we think about suicide risk factors.
If a research study found that depression increased risk of suicide by 50%, it sounds significant, but it is not actually that serious because death by suicide is such a relatively unlikely event to occur.
This issue had to be considered carefully by the researchers when interpreting their findings because there were many significant associations found between the potential risk factors and STBs, but this doesn’t translate into useful clinical information unless the size of the association is quite large.
Unfortunately, large associations were not found between any of the 16 domains of risk and STBs, nor between any of the subcategories and STBs.
Having a prior experience with suicidal thoughts or behaviours was the biggest predictor of suicide ideation and attempts in the future. In fact, the odds of engaging in suicide ideation or making a suicide attempt were both slightly less than 2.5 (ie, nearly 2.5 times more likely) if the person had a history of STBs. This sounds like a very significant finding, but as mentioned above it doesn’t significantly impact absolute risk because the base rates for suicide ideation (2/100 people) and attempts (0.33/100 people) are relatively low.
In terms of subcategories of risk, having a history of non-suicidal self-injury (ex: cutting) was the biggest predictor of making a suicide attempt, as it increased the risk by 4.15. The top predictor of suicidal death was having a prior admission to a psychiatric hospital, which increased the odds of death by 3.57 (ie, 3.57 times more likely to die by suicide compared to someone who never had a previous psychiatric hospital admission).
What Does This All Mean (and not mean)?
There are a few key points to take from this research on suicide risk. One of the main implications is that suicide research needs to change course. The researchers found an odd and discouraging trend in suicide research. Although research has increased over the decades, which is encouraging, researchers have tended to narrow their focus to a smaller group of risk factors, and this smaller group had some of the smallest effect sizes. Bluntly put, suicide researchers have developed the habit of studying the worst predictors of suicide more frequently, to the exclusion of other possible factors. This is embarrassing and more innovative thinking is needed.
A second key point to keep in mind, and one made repeatedly by the research team, is that these results are limited by the research designs used in the study of suicide. The typical suicide study looks like this — researchers measure a particular factor at a single point in time and then wait for years to see if the person engages in STBs. This may (likely?) reflect a poor model of suicide risk. For example, it is probably more likely that multiple factors coming together in a short period of time (days, hours or even minutes) best explain suicidal behaviour, and therefore would help us better understand risk. Unfortunately, this is not how research on suicide seems to be done.
This is important to keep in mind because the failure to find big risk factors in this research study does not mean that these risk factors do not predict suicide. They may very well (and likely do) predict suicide, but they may do so only in combination with other variables, and over short periods of time.
Indeed, the final point to be made is that the current research findings do not mean that it is impossible or worthless to try and predict suicide. Just because research designs fail to study it the right way, and therefore fail to find significant results, does not mean we just ignore all signs of suicide. For example, I suspect that I have helped to prevent a few suicides over the years, and some patients have later told me that they were alive because of how their risk was handled. I would venture to guess that my colleagues have had similar experiences.
As such, I have provided the following things to look for based on existing guidelines for suicide risk and prevention. Suicide remains a very difficult issue to predict and understand, but that does not mean it cannot be prevented or understood at all. If you are concerned about a patient or loved one with the following indicators, take steps to talk to them about it and consider (a) taking them to the nearest emergency room, or enlisting the help of others (possibly police, if necessary) to take the person to the ER, (b) calling the mobile crisis unit in your city to report your concerns (if your city or town has one), and/or (c) encouraging the person to seek immediate consultation with a mental health worker, and/or phoning a suicide hotline.
Major Risks/Warning Signs of Suicide
1. Previous history of suicide attempts
2. Currently engages in active suicidal ideation (ex: making a plan and getting access to the means to do it).
3. There has recently been a significant loss or life stress (ex: break-up; legal problems; failure experience)
4. Increased substance abuse
5. Person reports no reason for living
6. Hopelessness
7. Person engages in risky/reckless behaviours
8. Dramatic change in mood (ex: they become suddenly very sad)
9. Signs of leaving — giving away prized possessions, saying goodbye, writing a note(s)
10. The presence of a serious mental illness
11. Appearing restless/feeling trapped/angry/impulsive
*The majority of facts and information reported in this article, including all the statistics come from the article reviewed:
- Franklin et al. (2016). Risk Factors for Suicidal Thoughts and Behaviours: A Meta-Analysis of 50 Years of Research. Psychological Bulletin.