4 Classification and assessment in psychiatry4.3 Clinical assessment
4.3.7 Risk assessment and safety planning
3 min read2 figures
Written by Resident Doctor Committee · Published
Principles of risk assessment
Defining the outcome
Risk assessment covers harm to the person and to others, including suicide, self-harm, violence, self-neglect, exploitation and harm from restrictive care itself.
Specify the outcome, who is affected, the severity, the timeframe and the setting.
Deal first with immediate safety: injury, overdose, current violence, imminent suicidal intent and access to lethal means.
Learning from past behaviour
Previous sexual offending is the strongest predictor of future sexual offending.
It outweighs psychiatric, substance and developmental factors, which are general violence risk factors or have little specific predictive value.
Past behaviour informs the assessment of similar outcomes but does not settle an individual's future.
For each previous incident, establish recency, frequency, severity, intent, triggers and similarity to the current situation.
Types of risk factor
Type | Timescale | Example |
|---|---|---|
Static | Historical or largely fixed | Previous serious harm |
Stable dynamic | Weeks to months | Substance dependence |
Acute dynamic | Hours to days | Intoxication, weapon access |
Management focuses on the factors that can change.
Approaches to judgement
Approach | Features |
|---|---|
Unstructured clinical judgement | Flexible, but prone to omission and inconsistency |
Actuarial | Fixed statistical rules whose estimates apply to groups |
Structured professional judgement | An evidence-informed framework combined with formulation and management planning, as in the HCR-20 |
NICE advises against using scales or low, medium and high categories to predict suicide or repeated self-harm, decide treatment or justify discharge.
Rare outcomes give poor positive predictive value even when test accuracy looks reasonable.
Formulation and safety planning
Protective factors do not simply cancel risk factors; each should interrupt a specific route to harm.
A risk formulation explains how vulnerabilities, triggers and protective factors interact, names warning signs and uncertainty, and is revised as circumstances change.
A safety plan lists warning signs, coping strategies, distractions, supportive contacts, professional and crisis contacts, and steps to make the environment safer.
Recognised risk assessment tools
What the HCR-20 assesses
The HCR-20 Version 3 is a structured professional judgement instrument for the risk of future violence to other people in adults.
Its scope stops at interpersonal violence: it does not assess suicide, self-harm or relapse, and it is not an actuarial probability calculator.
Relapse, suicide risk, stalking and sexual violence or recidivism each have their own approach or tool, and separate guides cover intimate partner violence and adolescents.
The twenty items
Domain | Items | Timeframe | Examples |
|---|---|---|---|
Historical (H1 to H10) | 10 | Past | Violence, substance use, personality disorder |
Clinical (C1 to C5) | 5 | Present | Insight, violent ideation or intent, instability |
Risk management (R1 to R5) | 5 | Future | Living situation, personal support, stress or coping |
Rating and judgement
Each item is rated for presence (absent, possibly or partly present, present) and separately for relevance to future violence.
When information is insufficient, the rater records that, rather than treating the item as absent.
Clinicians do not reach conclusions by adding item scores or applying a cut-off, and one highly relevant factor can outweigh several well-managed problems.
Scenarios describe how violence might occur: likely harm, victims, setting, timeframe, triggers, warning signs and protective conditions.
Summary judgements address case prioritisation, serious physical harm and imminence, and they are contextual, not permanent labels or individual probabilities.
Limitations
No tool guarantees prediction: false positives and negatives depend on the tool's performance, the threshold and how common the outcome is.
Validation in one population does not transfer automatically to another.