Assessing the predictability of self-harm in a high-risk adult prisoner population: a prospective cohort study

Background: Prisoners are at increased risk of self-harm and when either intent is expressed, or an act of self-harm carried out, prisoners in the UK are subject to self-harm/suicide monitoring (referred to as “open ACCT” monitoring). However, there is a paucity of validated instruments to identify risk of self-harm in prisoner populations. In response to the need to support prison staff to determine who is at increased risk of self-harm or repeat self-harm, the aim of this study was to determine whether any pre-existing, standardised instruments could usefully identify future self-harm events in prisoners undergoing ACCT monitoring.

Methods: A multi-stage prospective cohort study was conducted, where the Prison Screening Questionnaire (PriSnQuest), a modified Borderline Symptom List-23 (BSL-23), Self-Harm Inventory (SHI), Patient Health Questionnaire-9 (PHQ-9) and Clinical Outcomes in Routine Evaluation – Outcome Measure (CORE-OM) instruments were administered to prisoners aged 18 and above, who were judged to be at an increased risk of self-harm (on open ACCT monitoring) during the recruitment phase. A 6-month follow-up determined self-harm occurrence since baseline, and Area-Under-the-Curve (AUC) analysis examined the ability of the instruments to predict future self-harm.

Results: Prison records established that 29.1% self-harmed during the follow up period, involving a total of 423 self-harm events reported from 126 individuals, followed up for 66,789 prisoner days (median 167 days; IQR 71–207.5 days). This translated to an ‘event incidence’ of 6.33 per 1000 prisoner days of those who had been placed upon an ACCT, or ‘prisoner incidence’ of 1.89 per 1000 days, with considerable variation for both gender and participating prisons. None of the summary scores derived from the selected instruments showed a meaningful ability to predict self-harm, however, exploratory logistic regression analysis of individual background and instrument items revealed gender-specific item sets which were statistically significant in predicting future self-harm.

Conclusions: Prospective self-harm was not predicted by any of the pre-existing instruments that were under consideration. Exploratory logistic regression analysis did reveal gender-specific item sets, producing predictive algorithms which were statistically significant in predicting future self-harm; however, the operational functionality of these item sets may be limited.