Changes in Dynamic Risk Factors Predict Short-Term Institutional Violence

There is general consensus that dynamic factors ought to be considered in the assessment of violence risk, but little direct evidence exists to demonstrate that within-individual fluctuations in putative dynamic factors are associated with changes in risk. We examined these issues in a sample of 30 male forensic psychiatric inpatients using a pseudoprospective design. Static and dynamic factors were coded on the basis of chart review using 2 structured measures of violence risk: Version 2 of the Historical–Clinical–Risk Manage- ment—20 (HCR–20; C. D. Webster, K. S. Douglas, D. Eaves, & S. D. Hart, 1997, HCR–20: Assessing risk for violence, Version 2, Vancouver, BC, Canada: Mental Health, Law, and Policy Institute, Simon Fraser University) and the Short-Term Assessment of Risk and Treatability (START; C. D. Webster, M. L. Martin, J. Brink, T. L. Nicholls, & S. L. Desmarais, 2009, Short-Term Assessment of Risk and Treatability [START], Version 1.1, Coquitlam, BC, Canada: British Columbia Mental Health and Addiction Services). HCR–20 and START assessments were repeated every 3 months for a period of 1 year. Institutional violence in the 3 months following each assessment was coded using a modified version of the Overt Aggression Scale (S. C. Yudofsky, J. M. Silver, W. Jackson, J. Endicott, & D. W. Williams, 1986, The Overt Aggression Scale for the objective rating of verbal and physical aggression, The American Journal of Psychiatry, Vol. 143, pp. 35–39). Dynamic risk and strength factors showed predictive validity for institutional aggression. Results of event history analyses demonstrated that changes in dynamic risk factors significantly predicted institutional violence, even after controlling for static risk factors. This is one of the first studies to provide clear and direct support for the utility of dynamic factors in the assessment of violence risk.

violence risk assessment, dynamic factors, inpatient aggression, HCR–20, START

Archival file data for 30 male inpatients who had been hospitalized after being deemed unfit/incompetent or insane/not criminally responsible and who had participated in a previous study of institutional violence at the hospital were randomly selected for inclusion in this study: 15 from the pool of potential participants who had engaged in at least one instance of institution violence between January 1 and December 31, 2004 (cases) and 15 who did not engage in any violence during that same period (controls).

The study used a design (pseudoprospective, case—control design) wherein the HCR-20 and the START (structured professional judgment risk assessment instruments) were coded on the basis of archival file data for the three months immediately preceding and these ratings were then used to predict institutional violence over the next three months. This was done for a total of 4 consecutive assessment and 4 follow-up periods at regular 3-month intervals between October 1, 2003 and December 31, 2004.

“Dynamic factors were assessed every 3 months over a 12-month period, affording the opportunity to examine whether these factors truly are dynamic in nature and whether change in dynamic variables is associated with institutional violence risk” (p. 385).

Results indicate that the dynamic items of the HCR-20 (C and R items) and the START (Strength and Vulnerability ratings) predicted future institutional aggression. In addition, the dynamic risk factors showed incremental predictive validity over the historical risk factors.

“The findings of this study provide strong evidence for the predictive validity of dynamic risk factors for institutional violence. The dynamic items of both the HCR-20 and the START showed that predictive validity and changes in dynamic risk factors were reliably associated with institutional violence risk even after controlling for observed and unobserved static factors…These current findings provide clear and direct support for the inclusion of dynamic factors in the process of assessing and managing institutional violence risk” (p. 384).

The results of this study “demonstrated that dynamic factors fluctuated over time and that these changes are associated with future aggression beyond what can be predicted with historical variables alone. Specifically, when the dynamic risk factors were included in the prediction model, the historical factors were no longer significant predictors of short-term inpatient aggression” (p. 385).

This study has two important implications for violence risk assessment.  “First, violence risk assessments should include explicit evaluation of dynamic risk factors. [These] findings add to a growing body of research demonstrating the unique and added value of considering dynamic factors in addition to static factors in the prediction of violence risk” (p. 385).

Second, “violence risk assessments should be time limited. [Because] dynamic factors change over time, they should be reassessed regularly” (p. 385).

Taken together, these two implications are elaborated by the study authors for professionals involved in making treatment and management decisions:

“In contrast to the violence-prediction approach, in which assessors make a one-time prediction of risk for violence, [these] findings support use of a violence-prevention model. That is, instead of a one-time assessment of risk for violence, efforts should be made to reevaluate and manage an individual’s level of risk to reduce the likelihood of violence over time. Repeated assessment of risk and (re-)examination of changes in risk level would afford the opportunity to (a) monitor a patient’s risk and treatment needs and (b) make more informed decisions regarding when treatment is needed, what type of treatment to implement, whether treatment is working, and what modifications to the treatment plan may be required (as risk increases or decreases). Evaluators should explicitly communicate that risk may change over time, identify dynamic risk factors that may be important to monitor and reassess, and specify target dates for reassessment” (p. 385).

This article provides an excellent model of a research design for conducting high-quality, clinical research with finite resources. The researchers considered numerous factors and built control for many of these into their research design. Examples include using multiple, independent raters who were blind to outcome; using a case—control design to maximize statistical power for a given sample size; using multiple risk measures; making multiple measurements (i.e., repeated assessments at regular 3-month intervals); and counter-balancing within and across patients.

The analytic strategies undertaken in this research are also worth noting. The researchers used multiple analytic strategies, including event history analysis wherein “information about how variables change over time was incorporated into the analysis, and the ratings of dynamic risk factors at each assessment time were used to predict violence during the subsequent follow-up period.” A frailty term was included in the model to account for unobserved static or stable risk factors across follow-ups.

As always, please join the discussion below if you have thoughts or comments to add!