Suicide risk: what clinical assessment captures

A 34-year-old woman arrives at the emergency department after taking more pills than prescribed. She is clinically stable. The on-call physician applies a risk scale, adds up the points and reaches "low risk". In the conversation, however, she says she lost her job, that she lives alone and that this was the second time in a year.

Assessing suicide risk is one of the hardest tasks in clinical practice. The physician has to decide, in a short time, who needs more care. The temptation is to turn that decision into a number, with a scale or a high and low risk classification.

This text discusses what clinical assessment and scales actually capture, why individual prediction fails and what the evidence suggests putting in their place: an assessment centered on the person's needs. It is a text about clinical reasoning and does not replace local protocols or specialized assessment.

Key points

• The high and low risk classification separates groups, but it predicts poorly what happens to each person.
• Very sensitive scales err on the side of excess, and scales with low sensitivity let most of those who return slip through.
• Self-harm is one of the strongest risk factors, and the first year after the episode is the most critical period.
• Risk factors describe populations, they do not decide a patient's fate.
• The assessment should serve to understand needs and plan care, not only to classify.

Why risk is so hard to predict

Suicide is a serious outcome and, at the same time, relatively rare in each group of patients. This combination creates a known statistical problem: even a strong risk factor has low predictive value when the base rate of the outcome is low.

The size of the risk after self-harm

A systematic review gathered 177 articles on what happens after an episode of self-harm (Carroll et al., 2014). Suicide occurred in 1.6 per 100 people at 12 months and in 3.9 per 100 at 5 years. Non-fatal repetition reached 16.3 per 100 in the first year.

The numbers show two things at once. The risk is much higher than in the general population, and that is why self-harm deserves attention. But the vast majority of people who go through an episode do not die by suicide in the following years (Carroll et al., 2014), which makes individual prediction very difficult.

Returning to the service is common

Suicidal thinking, even without self-harm, also carries risk. A registry from Northern Ireland followed 14,695 hospital visits for suicidal ideation (Griffin et al., 2020). Over 5 years, the cumulative incidence of a new visit was 40.5 per 100, with a risk of later self-harm of 18.3 per 100. Each new visit is a chance to review care.

Rare in the individual, frequent in the service

There is a paradox that confuses many students. For each patient, suicide is an unlikely event in the short term. For the service that sees many people in crisis, however, it is not rare: over time, some of these patients will die, and the team will remember each one of them.

This contrast explains why the pressure to predict is so great and why prediction fails. The question "is this patient going to kill themselves?" will almost always have the statistical answer "probably not", and even so care must be offered. The more useful question becomes another one: what does this person need now, and what can be done to reduce their risk, whatever it is?

What the high and low risk classification delivers

The central question is simple: does separating patients into high and low risk help to decide? A meta-analysis tried to answer with data from 53 samples in 37 studies (Large et al., 2016).

Different groups, weak prediction

The group classified as high risk had a chance of suicide about five times greater, with an odds ratio of 4.84 (Large et al., 2016). During follow-up, 5.5 per 100 high-risk patients died by suicide, against 0.9 per 100 of the low-risk ones.

The difference between groups is real, but the prediction for each patient is weak. Combined sensitivity was 56 per 100 and specificity was 79 per 100 (Large et al., 2016). A large share of the deaths occurred among people classified as low risk, and most of those classified as high risk did not die by suicide. The authors concluded that classification based on several risk factors is not much stronger than the association of each factor alone with suicide.

Clinical intuition also has limits

An Australian study compared clinicians' assessment, made with an 18-item checklist, with a model built from medical record data in 7399 patients (Tran et al., 2014). The area under the curve for the clinical assessment was 0.58, against 0.79 for the model using medical record data.

The most striking finding is that 66.8 per 100 patients had been seen in the emergency department in the previous year (Tran et al., 2014). The history of previous contacts with the service, often available in the medical record itself, carries information that the interview alone does not capture.

What the scales show in practice

Risk scales were created to standardize assessment. The studies that tested these scales in emergency services show a consistent pattern: none of them predicts well what will happen.

A prospective study with several scales

An English study applied several scales to 483 episodes of self-harm and followed repetition over 6 months (Quinlivan et al., 2017). Repetition occurred in 30 per 100 episodes. Sensitivity ranged from 1 per 100, with the SAD PERSONS scale, to 97 per 100, with the Manchester scale.

The positive predictive value ranged from 13 to 47 per 100. The best area under the curve, 0.74, came from the clinician's own global assessment (Quinlivan et al., 2017). The authors concluded that the scales should not be used to decide treatment.

The same pattern in 4000 episodes

A multicenter study assessed 4000 episodes of self-harm in English emergency services (Steeg et al., 2018). Repetition within 6 months occurred in 28 per 100. The Manchester and ReACT scales had high sensitivity, 98 and 94 per 100, but specificity of only 15 and 23 per 100.

The SAD PERSONS scale, in turn, identified only 24 to 29 per 100 of the people who repeated self-harm (Steeg et al., 2018). A very sensitive scale classifies almost everyone as high risk and does not help to prioritize. A scale with low sensitivity misses most of those who return.

Study	Outcome	What the scales showed
Quinlivan et al., 2017	Repetition within 6 months, 483 episodes	Sensitivity from 1 to 97 per 100; positive predictive value from 13 to 47 per 100
Steeg et al., 2018	Repetition within 6 months, 4000 episodes	Manchester and ReACT very sensitive, not very specific; SAD PERSONS not very sensitive
Quinlivan et al., 2016	Review of 8 articles	Sensitivity from 6 to 97 per 100; positive predictive value from 5 to 84 per 100
Harris et al., 2019	Adolescents, 11 studies	Sensitivity from 27 to 95.8 per 100

Table source: Quinlivan et al., 2017; Steeg et al., 2018; Quinlivan et al., 2016; Harris et al., 2019.

The reviews reach the same conclusion

A systematic review of scales for assessing risk after self-harm included 8 articles and found sensitivity between 6 and 97 per 100 and positive predictive value between 5 and 84 per 100 (Quinlivan et al., 2016). The authors concluded that it is hard to be sure which scale, if any, is the most useful, and that they should not be used alone to decide management.

In adolescents, a review of 11 studies with 10 instruments found sensitivity between 27 and 95.8 per 100 (Harris et al., 2019). The variation is too large for a single score to support a decision, a typical problem of diagnostic uncertainty.

Why high sensitivity does not solve the problem

A scale can be adjusted to catch almost every patient who will repeat self-harm. The price is classifying most of those who will not repeat as high risk too. In the emergency department, this means the "high risk" category stops separating patients and starts describing almost everyone.

The opposite move also has a cost. If the scale is adjusted to be more specific, it starts leaving out many of those who return, as happened with SAD PERSONS (Steeg et al., 2018). There is no cutoff point that escapes this dilemma when the outcome is rare and the risk factors are common. That is why reviews recommend that no scale alone decide discharge, admission or referral (Quinlivan et al., 2016).

Risk factors: what they describe and what they do not decide

Risk factors remain important. The mistake is using them as if, added together, they produced an individual prediction.

The strongest factors

A meta-analysis of psychological autopsy studies gathered 37 studies from 23 countries, with 5633 suicide deaths and 7101 controls (Favril et al., 2022). The factors with the strongest effect were the presence of a mental disorder, with an odds ratio of 13.1, and a history of self-harm, with 10.1. Family, life-event and social factors had more modest effects.

In specific groups, the risk is also high. A review on schizophrenia estimated the lifetime risk of suicide at about 5 in 100 (Hor and Taylor, 2010). For repetition of self-harm, a review of 129 studies with 329,001 participants found a consistent association with previous self-harm, personality disorder, hopelessness, previous psychiatric treatment, schizophrenia, alcohol and drug abuse and living alone (Larkin et al., 2014). Composite scales tended to have high sensitivity and low specificity.

Why adding up factors is not enough

The presence of a risk factor raises the probability in groups, but each patient has a unique combination of factors, context and resources. The reasoning of pretest probability helps to understand the problem: a factor present in many people who will never attempt suicide has little power to separate those who will from those who will not.

Risk factors are more useful for guiding the conversation and the care plan than for generating a score. They point to what needs to be asked and what can be changed.

Fixed factors and modifiable factors

A practical way to organize risk factors is to separate what does not change from what can change. A history of self-harm, a diagnosis of schizophrenia or a personality disorder are facts in the patient's history. Hopelessness, alcohol and drug use, lack of treatment and social isolation can be addressed in the care plan (Larkin et al., 2014).

This separation shifts the focus of the assessment. Instead of adding up points to reach a category, the clinician identifies what is within reach of the team, the person and the support network. A patient who lives alone and has lost a job, for example, may benefit from close follow-up, contact with a family member or referral to social support.

What the conversation captures that the scale does not

If individual prediction fails, what is left for the clinician? The answer from the reviews is an assessment focused on the person's needs, not on classification.

Assessment as an encounter

An English qualitative study gathered 28 family physicians in focus groups on suicide risk assessment in young people (Michail et al., 2016). The authors found wide variation in how physicians understood and operationalized risk, and argued for a broad assessment that considers individual, social and contextual influences.

The review on scales points in the same direction: it cites the recommendation to carry out full clinical assessments of each patient's situation and needs, instead of classifying into high and low risk, and notes that a focus on classification can harm the therapeutic relationship (Quinlivan et al., 2016).

The conversation reveals elements that no scale lists: what has changed in the person's life, what they think about the future, who is nearby. Empathy in clinical reasoning and narrative medicine are not decorations, they are tools for gathering this information.

Asking directly and respectfully

Many students avoid asking about suicidal thoughts for fear of causing offense or of not knowing what to do with the answer. A direct question, asked calmly and without judgment, is usually received as a sign of interest. Beating around the bush and euphemisms, by contrast, can give the impression that the subject cannot be discussed.

A useful sequence starts with the general and moves to the specific. First, how the person is feeling and how they see the future. Then, whether they have already thought that life was not worth living, whether they have thought about hurting themselves and whether they have made any plan. It is also worth asking what made them give up at other times, because the reasons to keep living are as much part of the plan as the risk factors.

The answer should not be recorded as an isolated "yes" or "no". What matters is the context: what has changed, what the person expects from help and who can be with them in the coming days.

Watching out for one's own judgment

With a patient who comes back repeatedly, the clinician may be swayed by fatigue or irritation. Affective bias and the labels discussed in stigma and language can lead to the repetition being read as "attention seeking", when the data show that repetition is precisely a risk marker (Carroll et al., 2014).

From classification to the care plan

The meta-analysis concluded that there is still no statistically strong and reliable method to distinguish high-risk patients (Large et al., 2016). The English prospective study reached the same conclusion for the scales (Quinlivan et al., 2017). What the assessment can offer is an understanding of the person's needs, which guides the care offered to everyone, not only to those classified as high risk.

Building the reasoning step by step

Suicide risk assessment can be organized in a sequence that does not depend on a score.

1. Treat the clinical emergency first and ensure immediate safety.
2. Listen to the story of the episode: what happened, what the person expected and how they feel now.
3. Ask about previous episodes and prior contacts with the service.
4. Explore mental disorders, substance use, recent losses and support network.
5. Identify what can be changed and what the person wants or accepts as help.
6. Use scales, if used, as support for the conversation, never as an isolated decision.
7. Record the assessment and the plan, with agreed follow-up, since the first year concentrates the risk.

What to record in the medical chart

Recording the risk assessment is part of care, not just a protection for the team. A good record describes the story of the episode, the factors identified, what the person said about the future and the agreed plan. A line with the risk category, without context, says little to the next professional who sees the same person.

Because many patients return to the service (Griffin et al., 2020), the medical chart is also the team's memory. Whoever handles the next episode needs to know what was tried, what worked and what the person accepted last time. This history is exactly the kind of information that, in the Australian study, carried more predictive power than the clinician's isolated assessment (Tran et al., 2014).

Returning to the patient in the case

The 34-year-old patient was classified as "low risk" by the scale. But she has a repeated episode within a year, a recent loss and social isolation. Repetition in the first year is common after an episode (Carroll et al., 2014), living alone and prior self-harm appear among the factors linked to repetition (Larkin et al., 2014), and according to the review of psychological autopsies self-harm is one of the strongest factors (Favril et al., 2022).

The low score should not end the assessment. The conversation about what has changed, what she expects and who can help her is worth more than the sum of points. The care plan should start from her needs, not from the risk category (Quinlivan et al., 2016). This kind of case, reviewed by the team, is also valuable material for morbidity and mortality sessions.

How NEXO helps train this reasoning

In NEXO, the case of the day is fictional, with a real clinical presentation and up to six progressive clues, and you propose the diagnosis by searching for the ICD-10 code.

The limits of using the game are described on the NEXO Economy page. If you research the teaching of clinical reasoning, the page for Researchers explains how to get in touch. Professionals who write or review cases find information on Curator and Creators, and institutions can learn about the proposal on the partnerships page. Training clinical decision-making and searching for the right ICD-10 code complete this practice.

Frequently asked questions

Do suicide risk scales work?

To predict what will happen to each person, they do not work well. Studies show scales that are very sensitive and not very specific, or the opposite (Steeg et al., 2018), and a review concluded that they should not be used alone to decide on management (Quinlivan et al., 2016).

Is it safe to classify the patient as low risk?

It should not be used as a guarantee. In the meta-analysis, the sensitivity of the high-risk classification was 56 out of 100, meaning a large share of deaths occurred in people classified as low risk (Large et al., 2016).

Does prior self-harm matter that much?

Yes. It is among the factors with the strongest effect in psychological autopsies (Favril et al., 2022), and the risk of suicide after an episode is much higher than in the general population (Carroll et al., 2014).

Does the clinician assess better than the scale?

Both have limits. The clinician's assessment with a checklist performed modestly (Tran et al., 2014), but the clinician's global scale had the best area under the curve among the instruments tested in the English prospective study (Quinlivan et al., 2017).

So what is the assessment for?

To understand the person's needs and plan care. The review on scales cites the recommendation to assess each patient's situation and needs instead of classifying them (Quinlivan et al., 2016).

Take-home messages

• Separating patients into high and low risk distinguishes groups, but it predicts each person's outcome poorly.
• No suicide risk scale performs well enough to decide treatment on its own.
• Prior self-harm and mental disorder are among the strongest factors.
• The first year after an episode of self-harm concentrates the risk.
• The conversation captures context, losses and support network that no score lists.
• A care plan should exist for everyone, not only for those classified as high risk.

If you or someone close to you is going through a difficult time, do not wait for the crisis to pass on its own: seek the help you need from health and emergency services in your area. Talking to someone you trust and seeking care is a step of care, not of weakness.

References

1. Carroll R, Metcalfe C, Gunnell D. Hospital presenting self-harm and risk of fatal and non-fatal repetition: systematic review and meta-analysis. PLoS One. 2014. PMID 24587141.
2. Favril L, Yu R, Uyar A, et al. Risk factors for suicide in adults: systematic review and meta-analysis of psychological autopsy studies. Evid Based Ment Health. 2022. PMID 36162975.
3. Griffin E, Kavalidou K, Bonner B, et al. Risk of repetition and subsequent self-harm following presentation to hospital with suicidal ideation: A longitudinal registry study. EClinicalMedicine. 2020. PMID 32529177.
4. Harris IM, Beese S, Moore D. Predicting future self-harm or suicide in adolescents: a systematic review of risk assessment scales/tools. BMJ Open. 2019. PMID 31494608.
5. Hor K, Taylor M. Suicide and schizophrenia: a systematic review of rates and risk factors. J Psychopharmacol. 2010. PMID 20923923.
6. Large M, Kaneson M, Myles N, et al. Meta-Analysis of Longitudinal Cohort Studies of Suicide Risk Assessment among Psychiatric Patients: Heterogeneity in Results and Lack of Improvement over Time. PLoS One. 2016. PMID 27285387.
7. Larkin C, Di Blasi Z, Arensman E. Risk factors for repetition of self-harm: a systematic review of prospective hospital-based studies. PLoS One. 2014. PMID 24465400.
8. Michail M, Tait L. Exploring general practitioners' views and experiences on suicide risk assessment and management of young people in primary care: a qualitative study in the UK. BMJ Open. 2016. PMID 26758263.
9. Quinlivan L, Cooper J, Davies L, et al. Which are the most useful scales for predicting repeat self-harm? A systematic review evaluating risk scales using measures of diagnostic accuracy. BMJ Open. 2016. PMID 26873046.
10. Quinlivan L, Cooper J, Meehan D, et al. Predictive accuracy of risk scales following self-harm: multicentre, prospective cohort study. Br J Psychiatry. 2017. PMID 28302702.
11. Steeg S, Quinlivan L, Nowland R, et al. Accuracy of risk scales for predicting repeat self-harm and suicide: a multicentre, population-level cohort study using routine clinical data. BMC Psychiatry. 2018. PMID 29699523.
12. Tran T, Luo W, Phung D, et al. Risk stratification using data from electronic medical records better predicts suicide risks than clinician assessments. BMC Psychiatry. 2014. PMID 24628849.