Predicting self-harm in prisoners: Risk factors and a prognostic model in a cohort of 542 prison entrants

Background. Self-harm is common in prisoners. There is an association between self-harm in prisoners and subsequent suicide, both within prison and on release. The aim of this study is to develop and evaluate a prediction model to identify male prisoners at high risk of self-harm. Methods. We developed an 11-item screening model, based on risk factors identified from the literature. This screen was administered to 542 prisoners within 7 days of arrival in two male prisons in England. Participants were followed up for 6 months to identify those who subsequently self-harmed in prison. Analysis was conducted using Cox proportional hazard regression. Discrimination and calibration were determined for the model. The model was subsequently optimized using multivariable analysis, weighting variables, and dropping poorly performing items. Results. Seventeen (3.1%) of the participants self-harmed during follow up (median 53 days). The strongest risk factors were previous self-harm in prison (adjusted hazard ratio [aHR] = 9.3 [95% CI: 3.3-16.6]) and current suicidal ideation (aHR = 7.6 [2.1-27.4]). As a continuous score, a one-point increase in the suicide screen was significantly associated with self-harm (HR = 1.4, 1.1-1.7). At the prespecified cut off score of 5, the screening model was associated with an area under the curve (AUC) of 0.66 (0.53-0.79), with poor calibration. The optimized model saw two items dropped from the original screening tool, weighting of risk factors based on a multivariable model, and an AUC of 0.84 (0.76-0.92). Conclusions. Further work is necessary to clarify the association between risk factors and self-harm in prison. Despite good face validity, current screening tools for self-harm need validation in new prison samples.