Discussion

Our findings have implications for the use of the HCR-20 and VRAG in predicting aggression occurring within psychiatric hospitals. In keeping with research in community follow-up, ^{Reference Coid, Ullrich and Kallis5} it would appear that these instruments are not performing markedly well across aggression types with those with a personality disorder detained in conditions of secure psychiatric care. Indeed, PLRs generally ranged from minimal to small, with the only exception a moderate value in relation to the VRAG and its prediction of physical aggression against staff at 4 years, with verbal aggression against patients at 4 years closely following. The findings are broadly consistent with other research examining aggression occurring within hospitals, although previous research has used considerably shorter follow-up periods. ^{Reference O'Shea, Picchioni, Mason, Sugarman and Dickens7,Reference Arbach-Lucioni, Andrés-Pueyo, Pomarol-Clotet and Gomar-Soñes9} Nevertheless, there is evidence for lower predictive accuracy with participants with a personality disorder, particularly in relation to physical violence, ^{Reference O'Shea, Picchioni, Mason, Sugarman and Dickens7} although the VRAG does seem to have some utility, particularly in relation to physical aggression against staff and with predicting patient self-harm.

Overall, it was the historical component of the HCR-20 that was presenting with some potential to accurately discriminate, even though this was limited to the 4-year follow-up and not producing PLRs beyond small values. The marginal results for the historical components as a whole were perhaps supportive of the more favourable results indicated by the actuarial risk instrument, the VRAG. The VRAG does not include dynamic risk factors but a wider range of historical factors. It suggests that the more dynamic elements of risk assessment (HCR-20) are not contributing to aggression outcomes with this very specific population, and certainly not when longer time frames are being applied. Our results indicated that the more historical and static VRAG was a better predictor than the HCR-20. Interestingly, the VRAG also demonstrated moderate (closer to good) discriminatory potential in identifying patients likely to self-harm across both time points (12 months and 4 years); it also performed moderately to good in terms of predicting verbal aggression towards patients and staff at 4 years, and total aggression at 4 years.

Our results do perhaps suggest some degree of caution in the application of risk assessment instruments to patient groups characterised by enduring challenges in personality functioning. The difficulty for clinicians is that such difficulties are associated with an increased risk for aggression and yet the risk instruments more commonly applied do not seem to be discriminating beyond at least moderate with this patient group across the longer term. Our findings also suggest that actuarial assessments cannot be considered completely without merit ^{Reference Hart, Michie and Cooke3} when focus is on hospital-based aggression.

The environment is also an important consideration. Although placement in a secure psychiatric setting is arguably a protective factor against the expression of overt aggression, ^{Reference Chu, Daffern and Ogloff6} nothing is reliably known about how these risk instruments are converted, if at all, into clinical practice in the longer term to manage potential aggression risk. ^{Reference Vojt, Thomson and Marshall10} The current study did not, for example, identify any means of reliably assessing the content or quality of risk management plans put in place following these assessments; such plans are varied and can be expected to change over time. This is an obvious limitation for a long-term study. Examining this in future research, however, would be valuable and perhaps future revisions of risk assessment tools could consider a rating centred on the quality of risk plans put in place following such assessments, to what extent they were implemented and how such strategies could be evaluated effectively over time. Research has not comprehensively addressed these issues; the limited research to date has focused on very brief time periods (e.g. 24 h follow-up ^{Reference Ogloff and Daffern13} ) and not used all components of structured risk assessments (e.g. only considering the clinical component of the HCR-20). ^{Reference Daffern, Howells, Hamilton, Mannion, Howard and Lilly14} Thus it would be a valuable direction for future research to pursue in more detail over longer time periods.

Psychiatrists who are required to complete violence risk assessments need, however, to be mindful of the debates and associated potential limitations in using risk instruments with patients with a primary diagnosis of personality disorder in high secure psychiatric populations. Further research needs to expand on these issues by going beyond what the current study was able to provide and examining the impact of medication, incorporating neuroimaging variables and comorbid conditions of potential interest such as epilepsy. Indeed, we recognise that the current sample is a highly specialised one, namely a high secure sample with long-standing issues relating to aggression; this does make it particularly challenging for a generally validated risk assessment tool to predict aggression owing to the specifics of this population. The lack of generalisability outside of high secure settings is certainly acknowledged. There is also a need to expand the current research by considering the role of personality clusters in determining aggression risk. Although controlling for those individuals presenting with more than one cluster may be challenging among a sample where more than one personality disorder is common, it would still remain a valuable avenue to explore. Future research could examine the specific nature of individual personality disorder traits and how they associate with aggression in more detail. It could also extend to considering the role of personality functioning, as promoted by DSM-5, ¹⁵ and where functioning challenges may impact on aggression risk. This would represent a novel and valuable area of study to consider.