Risk models could help correctional health teams make more consistent decisions about suicide and self-harm prevention, but only when connected to treatment, safety planning and professional judgment.
Seena Fazel, professor of forensic psychiatry and director of the Centre for Suicide Research at the University of Oxford, presented the Aug. 6 seminar “Risk assessment and prediction modelling for self-harm and suicide risk in correctional populations” at UTMB’s Health Education Center.
Moving beyond checklists
The seminar addressed a difficult point in correctional care, determining what support a person may need after an immediate risk-management plan ends.
Screening instruments developed for other clinical settings have generally performed poorly when used to predict repeat self-harm in prisons. In England, the National Institute for Health and Care Excellence now advises against using risk tools to predict suicide or repeated self-harm. Fazel considers that an overcorrection, noting the evidence behind it came largely from emergency-department studies of older instruments that sorted people into categories rather than estimating probability.
A stronger approach, Fazel argued, begins with large datasets from the population in which a model will be used. Predictors and outcomes should be defined in advance, the model should be tested in a separate population, and the required information should be routinely available.
Probability estimates may also be more useful than fixed “high-risk” and “low-risk” categories. They acknowledge uncertainty while giving professionals a consistent reference point.
Assessing the risk of repetition
The presentation highlighted RAPSS, a structured risk-assessment model for people in prison. The model estimates the probability that a new risk-management plan will be opened within 90 days after a previous plan closes.
The development and validation study used 754 people across seven English prisons and tested the model prospectively among 390 people across 13 prisons. About 22% experienced the outcome within three months.
The final model used nine routinely collected factors. Its ability to distinguish between people who did and did not experience the outcome was moderate, while its predicted probabilities aligned reasonably well with observed outcomes. The researchers called for further validation in other settings.
The findings also suggested that factors associated with repeated risk-management episodes differ from those associated with an initial episode. No single cutoff is appropriate for every setting, and the choice would depend on available resources and the support provided after a score is calculated.
Connecting assessment to care
Assessment alone does not improve outcomes, Fazel said. A model should support professional decisions, not replace them.
RAPSS could prompt renewed safety planning or help identify people who may benefit from additional follow-up, multidisciplinary review or psychological services. A lower score should not be used to deny care.
The broader point was that prediction should not become the model’s sole purpose. Its value may also lie in making decisions more transparent and connecting assessment with therapeutic support.
The seminar preceded UTMB’s Inaugural National Symposium on Correctional Healthcare on Aug. 7, where Fazel will deliver the closing keynote. View the 2026 Correctional Healthcare Symposium agenda
Seena Fazel is Professor of Forensic Psychiatry and Director of the Centre for Suicide Research at the University of Oxford, where his research examines suicide, self-harm, violence and crime among people with mental illness. He holds an honorary appointment with Oxford Health NHS Foundation Trust, worked for more than 13 years as a visiting psychiatrist in a local prison, and serves as an expert member of the UK’s Independent Advisory Panel on Deaths in Custody. The RAPSS study was funded by the Health Technology Assessment programme of the National Institute for Health and Care Research.