Prediction of institutional aggression among personality disordered forensic patients using actuarial and structured clinical risk assessment tools: prospective evaluation of the HCR-20, VRS, Static-99, and Risk Matrix 2000

Abstract

Entry to the Dangerous and Severe Personality Disorder (DSPD) service in England and Wales is heavily determined by risk status, and therefore requires valid procedures for monitoring changes in risk over time in order to make risk management decisions and determine patients’ suitability for transfer to lower security settings. Yet little is known about the validity of current risk assessment tools with the new DSPD population. This study reports a prospective evaluation of the predictive accuracy of the HCR-20, VRS, Static-99, and Risk Matrix 2000 with 44 consecutive admissions to the DSPD unit at a high secure forensic psychiatric hospital. Thirty eight per cent of the sample exhibited interpersonal physical aggression (IPA) on one or more instances over an average 1.5 year period following admission, and a similar percentage caused damage to property (DTP) on one or more occasions over the same period. All tools predicted DTP. HCR-20 Total and scale scores predicted IPA with structured final risk judgements also predicting repetitive (2 + incidents of) IPA. HCR-20 Risk Management scores were significantly associated with imminence of IPA. Results were discussed in terms of the practical utility of these tools with high risk forensic psychiatric inpatients.

Keywords:

• personality disorder
• psychopathy
• DSPD
• risk assessment
• inpatient aggression

Acknowledgements

The authors would like to acknowledge the work of the clinical staff at the Peaks Unit, Rampton Hospital, and the involvement of the patients. The authors would like to thank Sylvia Langton for assistance preparing the manuscript. Opinions or points of view expressed are those of the authors and do not necessarily reflect the official position or policies of Nottinghamshire Healthcare Trust or any other agency.

Notes

1. Although a number of indices of predictive validity are reported in the literature, the area under the Receiver Operating Characteristic (ROC) Curve (AUC) is the accepted measure of predictive and diagnostic accuracy (Swets et al., 2000), representing an index of the trade-off between sensitivity and specificity as a function of test score (Harris, Rice, & Quinsey, 2007). There is not, however, agreement regarding an accepted standard for interpreting AUC values. Rice and Harris (2005) provide a table (pp. 616–617) for comparing the most common current measures of effect sizes. They report that an AUC value of 0.556 equates to a Cohen's d of 0.200, considered a small effectr size, an AUC value of 0.639 equates to a Cohen's d of 0.500, considered a medium effect, and an AUC value of 0.714 equates to a Cohen's d of 0.800, considered a large effect size.

2. Items in individual cases were omitted when insufficient information to score the item was available or conflicting information could not be reconciled. No adjustment was made to scores on the Static-99 or RM 2000 scales when items were omitted. PCL-R Total, Factor, and Facet scores were prorated when items were omitted according the manual (Hare, 2003) as were the VRS Total, Static, and Dynamic Factor scale scores (Wong & Gordon, 2000). HCR-20 Total, Historical, Clinical, and Risk Management scale scores were prorated when items were omitted as per the general recommendation in the manual (Webster et al., 1997). In the absence of specific instructions, the procedure used was the same as that used for the VRS: where x is the total or scale score, y is the number of items used to calculate the total or scale score, and z is the number of omitted items.

3. It was decided to analyse the data using both the full follow-up period and an initial 12-month fixed follow-up period in order to ensure that potentially important findings were not missed. Some studies simply report an average follow-up time and do not attend to the possibility that follow-up times between patients in the sample may vary considerably. Where all patients have been followed-up for a reasonable minimum period this may be an adequate approach. But, if present, differences in follow-up time could confound associations between predictors and outcomes if patients with lower scores on the measures are followed for longer periods than patients with higher scores; incidents of aggression might be missed for higher scoring patients but detected for those with lower scores. A full follow-up period, sometimes referred to in other studies as a ‘variable’ follow-up period, was used in the present study in order to utilize all available data and facilitate comparisons with other studies. An initial 12-month fixed follow-up was also used because it controls for unequal follow-up times between individuals. Harris and Rice (2003) have shown that fixed and equal follow-up is one methodological feature that maximizes predictive accuracy. But there is a trade-off if the total sample is small: when the fixed follow-up is short (e.g. a month) most of the sample can be retained but fewer incidents of interest occur. If the fixed follow-up period is longer (e.g. 12 months) more of the focal incidents occur but the sample size can drop (because those followed-up for less than 12 months who are incident-free are removed from the analyses). Associated with this reduced sample size is the loss of statistical power and less confidence in the generalizability of the findings. These concerns likely go some way to explaining differences in findings for individual tools between the two follow-up periods in the present study. Given the need for empirical evaluations of these tools with psychiatric patients in high secure settings such as those of the DSPD programme, the use of both follow-ups was considered informative as a first step in this line of research.