Risk formulation after NG225
NICE says not to stratify risk into low, medium and high, and not to use scales to predict suicide. Here is the reasoning, and what a formulation looks like in practice.
If you trained more than a few years ago, you were probably taught to end a risk assessment with a rating. Low, medium or high. Perhaps a number from a scale. It felt rigorous, it fitted neatly in a box on the form, and it gave the next clinician something definite to act on.
Current UK guidance says not to do it. NICE guideline NG225 is explicit, and the reasoning is better than most clinicians realise.
What NG225 actually says
NICE guideline NG225, Self-harm: assessment, management and preventing recurrence, published September 2022, contains three recommendations that between them rule out the familiar practice.
- Do not use risk assessment tools and scales to predict future suicide or repetition of self-harm.
- Do not use risk assessment tools and scales to determine who should and should not be offered treatment, or who should be discharged.
- Do not use global risk stratification into low, medium or high risk to predict future suicide or repetition of self-harm, or to determine who should be offered treatment or discharged.
In place of them, NICE asks clinicians to focus the assessment on the person's needs and how to support their immediate and longer-term psychological and physical safety, and states that mental health professionals should undertake a risk formulation as part of every psychosocial assessment.
NHS England's guidance on staying safe from suicide takes the same line, describing static stratification as unacceptable practice.
The evidence behind it
The reasoning rests on work from the National Confidential Inquiry into Suicide and Safety in Mental Health, and it is uncomfortable reading.
NCISH examined the clinician's judgement of risk at the last contact before death for patients who went on to die by suicide. The large majority had been rated as low risk or no immediate risk at that final encounter — a finding usually summarised as around eight in ten.
This is sometimes called the low-risk paradox, and it is not a story about poor clinicians. It is a story about base rates. Suicide is, statistically, a rare event. Any instrument attempting to predict a rare event produces enormous numbers of false positives, and still misses most of the true cases. The people who die are drawn overwhelmingly from the large "low risk" group simply because that group is so much larger.
A "low risk" rating carries almost no predictive information, but it carries a great deal of reassuring information. It changes how the next clinician reads the note, how urgently a review is arranged, and how a family's concerns are weighed. That is the harm: not the number itself, but what everyone downstream does with it.
What to do instead of a score
The alternative is not vagueness. It is a narrative formulation that does four things a score cannot.
- Describes what was found, in enough detail that another clinician could form their own judgement rather than inheriting yours.
- Identifies what is modifiable — the things that could change this week, as opposed to historical facts that cannot.
- Records what was agreed with the person, in their words where possible.
- States what you did, and when it will be reviewed.
The distinction between static and dynamic factors is where formulation earns its place. Previous attempts, a family history, chronic illness — all relevant, none changeable. Access to means, current alcohol use, an eviction next week, whether anyone is at home tonight — these are the ones an intervention can touch, and they are what the formulation should be organised around.
Writing a risk formulation
A workable structure, which fits comfortably into a paragraph or two.
| Element | What goes in it |
|---|---|
| Current presentation | What was disclosed today, in the person's own words where possible. Include denials, since a blank is read as never asked. |
| What is driving it | The current, changeable circumstances — loss, housing, substances, sleep, a discontinued medication. |
| Background factors | History that raises long-term concern, kept brief and clearly marked as historical. |
| Protective factors | Specific, not generic. "His sister will stay with him" is useful; "has family support" is not. |
| What was agreed | The safety plan, co-produced. Who holds medication, who they call, where they will be tonight. |
| What you did | Escalation, referrals, observations, and the review interval. |
Asking about suicide
Asking does not plant the idea. That concern has been examined repeatedly and the evidence does not support it. Not asking, on the other hand, reliably produces a record in which the question was never put.
Move from general to specific, and stop to respond to what you hear rather than completing a sequence.
"Have things ever got so difficult that life did not feel worth living?"
"Have you had thoughts of ending your life?"
"Have you thought about how you might do it?"
"Have you taken any steps towards it, or made any preparations?"
"How close have you come?"
"What has kept you going so far?"
"Is there anyone you would tell if things got worse?"
"What would help you feel safer tonight?"
"What has helped when you have felt like this before?"
It surfaces protective factors in the person's own words rather than yours, and it moves the conversation from enumerating danger to building a plan. The answer is frequently the most quotable line in the whole assessment.
The safety plan
A safety plan is co-produced, specific, and written in language the person would recognise as theirs. Generic plans are a documentation exercise; specific ones are an intervention.
Cover: what the early warning signs are for this person; what they can do themselves; who they will contact and how; what makes the environment safer tonight; and what professional contact is arranged and when.
Restricting access to means is one of the better-evidenced suicide prevention measures, and it belongs in the plan. Record it in terms of the action agreed — who is holding what, and what has been removed — rather than cataloguing methods in the notes.
A worked example
Synthetic. Note that it contains no rating and no score, and that a reader could still act on it.
She described passive thoughts that life is not worth living, without plan or stated intent, present most days for around three weeks. She said: "I would not do anything, I just want it to stop." No preparatory acts were described and no previous attempts were disclosed.
These thoughts are closely tied to circumstances that have changed recently: she lost her tenancy six weeks ago after a bereavement and is staying temporarily with a cousin. She describes disrupted sleep and has not been taking her prescribed medication for the past fortnight.
Protective factors she identified were her daughter, who lives locally, and a wish to return to work. She was clear that she would tell her daughter if things worsened.
Agreed together: her daughter will call daily and has been given the crisis line number, which has also been saved to her phone. Her cousin will hold the medication and dispense it weekly. She has agreed to restart the prescription tomorrow. Discussed with the duty consultant; housing referral submitted today; review arranged in five days, sooner if her daughter is concerned.
Common objections
"My trust's form has a low/medium/high box."
Many still do; the guidance changed faster than the paperwork. Complete what your employer requires, and make sure the narrative formulation is present and substantial, because that is what supports the decision if it is ever examined. Raising the discrepancy with your clinical governance lead is reasonable and increasingly common.
"Without a rating, how does the next clinician know how worried to be?"
By reading what you found. A formulation stating that a person disclosed active thoughts, has stopped medication, is drinking heavily and is alone tonight conveys concern far more precisely than "high risk" — and unlike the label, it says what to do about it.
"Isn't this just defensive practice in reverse?"
The opposite. A rating is the defensive option: brief, definite, and difficult to argue with afterwards. A formulation commits you to reasoning that can be examined. That is more exposing and more useful, and it is what current guidance asks for.
"What about structured tools like the Columbia scale?"
NG225 is directed at using tools and scales to predict and to determine treatment or discharge. Structured questioning that helps you enquire thoroughly is a different use from a score that decides who gets seen. If your service uses an instrument, be clear about which of those two things it is doing.
Common questions
Does NICE NG225 ban risk assessment?
No. It rules out using tools and scales to predict suicide or repetition of self-harm, and to decide who is offered treatment or discharged, including global stratification into low, medium or high risk. It asks instead for a risk formulation as part of every psychosocial assessment. Assessment continues; the numerical output does not.
Why is 'low risk' considered unsafe wording?
Because it is both uninformative and reassuring. Work from the National Confidential Inquiry found that the large majority of patients who died by suicide had been judged low or no risk at their final service contact. The rating carries little predictive value but strongly shapes how urgently everyone downstream responds.
What should replace low, medium and high risk in my notes?
A narrative formulation: what was disclosed today including denials, what current and changeable circumstances are driving it, brief background factors, specific protective factors, the co-produced safety plan, and what you did with a review interval. It is longer than a rating and considerably more actionable.
Does asking about suicide increase risk?
The evidence does not support that concern, and it has been examined repeatedly. Asking directly and calmly is standard practice. What does cause difficulty is not asking, which produces a record in which the question was never put and leaves the person without an opening to answer honestly.
Our trust form still requires a risk rating. What should I do?
Complete what your employer requires and make sure the narrative formulation is present and detailed, since that is what supports your decision if it is later examined. Many services are revising documentation to match NG225, and raising the discrepancy through clinical governance is a reasonable step.