Preventing and De-escalating Aggressive Behavior Among Adult Psychiatric Patients: A Systematic Review of the Evidence

Searches of all sources identified a total of 1,983 potentially relevant citations [see online supplement]. Twenty-two studies that otherwise met selection criteria were deemed ineligible because of a pre-post design. We identified 17 eligible controlled studies (described in 22 articles) that provided data for this review; the studies included more than 3,628 participants, and the samples ranged in size from 20 to 973 participants (not all studies reported sample size) (26–47). Thirteen studies were randomized trials (eight RCTs and five CRTs), two were NRCTs, and two were retrospective cohort studies (Table 1). Nearly half took place in the United States; most interventions took place in public psychiatric hospitals or inpatient psychiatric treatment units or facilities. For studies reporting demographic characteristics for their patient populations, the mean age ranged primarily between 38 and 40 years, the distribution of men and women varied widely across studies, and race or ethnicity was sparsely reported.

There are no agreed-upon categories for stratifying violence prevention and de-escalation interventions, so we sorted the interventions described in the eligible studies into the following five broad categories on the basis of the intervention’s main focus: staff training, risk assessment, multimodal programs, environmental or group psychotherapeutic interventions, and medication protocols. Table 2 describes the design of the eligible studies in each of the intervention categories.

Staff training interventions for clinical staff who provide acute care to patients with psychiatric symptomatology aim to equip staff with new skills or to promote staff attitudes that can help prevent or de-escalate aggression. Risk assessment interventions involve clinical staff’s use of structured assessment of individual patients’ risk of becoming actively aggressive. Multimodal programs involve a combination of various intervention types, such as enhanced administrative review of patients with high restraint use and staff training in strategies to better manage patients’ difficult behavior; the goal of the programs is to decrease the occurrence of active aggression or use of seclusion or restraint for managing active aggression.

Environmental or group psychotherapeutic interventions involve changes to the physical environment of the acute care setting or the introduction of group psychotherapeutic interventions meant to diminish precursors of active aggression. Finally, medication protocols encompass any medication-focused intervention to de-escalate active aggression, ranging from hospital- or unitwide policies specifically affecting how or which medications can be used to manage active aggression to the use of one or more emergency medications. Key characteristics of the eligible studies are listed in Table 3 [see the online supplement for more details about the studies and their risk of bias assessments].

The highest SOE grade for any outcome was low. The outcomes with low SOE involved two types of preventive interventions; two studies compared the benefits of risk assessment and one study compared the benefits of a multimodal intervention with usual care (Table 4). The SOE was insufficient for all other preventive interventions, for de-escalating interventions, and for modifying characteristics. For all ratings, the supporting evidence had medium risk of bias, had unknown consistency (because each finding was supported by a single study), was direct, and had precision. Of note, the CRTs in this review did not control for clustering in their statistical analyses, which weakened the SOE grade for those interventions.

Information about variables that might modify the effectiveness of interventions was limited. One RCT found significantly greater increases in weight, triglycerides, and cholesterol levels among black patients treated with clozapine compared with white or Hispanic patients treated with haloperidol, but the sample size did not reach the minimum OIS (insufficient SOE) (41).

Our review aimed to fill gaps in available literature about the comparative effectiveness of various strategies to prevent aggressive behavior, de-escalate aggressive behaviors, or decrease reliance on seclusion or restraint in acute care settings. An overarching objective of these strategies, of course, is to improve health outcomes for patients who are actively aggressive or at risk of acute aggressive behavior.

Overall, the evidence base was extremely limited. We identified 17 studies (mainly RCTs and CRTs) for which we could grade the SOE of one or more outcomes. Most evidence addressed preventive, unitwide programs rather than interventions specifically targeting actively aggressive patients; this focus represented the core difference between the CRTs (which randomize groups) and the RCTs (which randomize individuals). Moreover, some of these analyses included patients who were not actively aggressive These factors prevented us from attributing reduction of aggressive behavior among actively aggressive patients to any particular intervention. Furthermore, inexact descriptions of many interventions made it difficult to attribute a change to particular components. For example, the multimodal intervention had components of risk assessment and staff training, and distinguishing between their components was challenging.

None of the comparative data from the studies supported an SOE grade of higher than low, and all findings with low SOE were from studies of preventive interventions. The two studies of risk assessment protocols identified fewer aggressive incidents (34) and lower rates of severe aggressive incidents and physical attacks (27) compared with the usual care conditions. The protocols overlapped somewhat but differed in important ways. Both trials used the Brøset Violence Checklist as part of the protocol, but the trial from the Netherlands used a more comprehensive protocol that included completing a crisis monitor form and the Kennedy Axis V (short version) on a daily basis and the full version of the Kennedy Axis V, the Brief Psychiatric Rating Scale, the Dangerousness Scale, and the Social Dysfunction and Aggression Scale on a weekly basis (34). The trials also differed in the length of time during which they evaluated their risk assessment protocols. For example, one study (27) implemented the risk assessment protocol for the first three days of the patient’s hospital stay, whereas the other study (34) used the risk assessment protocol throughout each patient’s hospital stay.

The CRT of a multimodal intervention examined the safety and effectiveness of seclusion or restraint reduction strategies (modeled after the Six Core Strategies [11]) in the setting of high-security psychiatric units of a Finnish state hospital (32). Unlike prior examination of the Six Core Strategies in the United States (48), this CRT included data from a control group; thus, it provided the first outcome data eligible for an SOE assessment. Specifically, it reported decreases in the proportion of patient-days in seclusion, restraint, or room observation and in the duration of seclusion or restraint use. Importantly, both reductions were achieved without a concomitant increase in violent incidents, and these results were demonstrated in a patient population inherently at high risk of aggression (males with schizophrenia and history of violent behavior).

Our work is consistent with prior findings. Earlier reviews emphasized the lack of high-quality intervention studies of strategies to prevent the development of aggressive behavior in acute care settings (49–52). An absence of relevant literature on interventions for actively aggressive behavior has been similarly reported, regardless of whether alternative strategies to seclusion and restraint were being compared with each other (49–51,53) or with seclusion and restraint (49,51). A lack of literature about comparative harms of these interventions has also been identified (54). Our review updates and confirms these findings. Of particular relevance for clinical and administrative audiences, the review expands considerations of potentially relevant interventions to include staff training, environmental or group psychotherapeutic, multimodal, and pharmacologic interventions not previously reported.

Our review adds to existing reviews by including findings highlighting the potential benefits of two preventive interventions. First, a general application of a strategy that involves a risk assessment component for all individuals on inpatient psychiatric units—not just actively aggressive patients—may produce less aggressive behavior and less use of seclusion and restraint compared with usual care. The former finding extends what is known about the relationship between risk assessment and subsequent behavior. Earlier reviews of risk assessment found that using such a service was associated with decreased agitation (12), often considered an intermediate precursor of more dangerous aggressive behavior. Lowering agitation may or may not lead to decreased aggression, but our review found that risk assessment may lead to reduced subsequent aggression (as indicated by fewer aggressive incidents).

Second, both risk assessment and multimodal interventions may lower use of seclusion and restraint (as indicated by duration of seclusion or restraint and by use of forced treatment, including seclusion and restraint). These findings highlight a key potential benefit relevant to practice guidelines and quality-of-care measures advocating decreasing use of seclusion and restraint. Nevertheless, the SOE for these findings (all from CRTs) was limited by data analyses that did not account appropriately for the clustering of these data; this drawback likely affected each trial’s results, for example, by increasing the risk of a type I error. Further, the three studies forming the basis for the low SOE findings (risk assessment [27,34] and multimodal interventions [32]) were conducted outside the United States. How substantially clinical practice outside the United States differs from current U.S. practice is unclear, which may bring into question the applicability of findings from non-U.S. studies.

The handful of findings that we graded as low SOE may provide some clinical or policy implications. In particular, a limited number of risk assessment interventions subsequently led to less aggressive behavior and reduced the use of seclusion and restraint. These findings suggest that clinicians must consider carefully the role of these strategies on psychiatric inpatient units. Specifically, acute care practitioners and administrative staff must balance the low SOE with the reality that violence is a pressing—indeed, growing—concern and poses significant disruptions to quality of care in such settings. Broad implementation of a well-validated, structured risk assessment instrument in acute care settings illustrates a recommendation that could facilitate the prevention of aggression. Similarly, some evidence supports using multimodal interventions consistent with the Six Core Strategies to reduce seclusion and restraint, even in populations with psychiatric diagnoses or symptomatology that may be difficult to treat.

Several questions may arise, however. Is currently available limited evidence sufficient for evaluating effectiveness? Should implementation decisions be delayed until more evidence becomes available? How should the substantial barriers to conducting RCTs in these populations, involving both the challenges of obtaining informed consent as well as limited funds supporting such research, be considered in weighing and acting on the available evidence? What is the role of quality measures, designed to create incentives for improving quality of care, if the evidence base for those measures is unclear?

Regarding the last question, we are unaware of any ongoing trials that will add to the current sparse body of evidence addressing the benefits of risk assessment protocols and multimodal interventions. Furthermore, we cannot comment on potential harms or costs associated with implementing risk assessment protocols. Therefore, determining how to apply interventions from other countries to settings in the United States—and determining the modifications that might be necessary to do so—are key next steps, given the absence of SOE findings from inpatient psychiatric settings in the United States.

The paucity of evidence means that most implications of our review pertain to future research rather than to clinical or policy judgments. Major evidence gaps exist in this increasingly worrisome clinical arena; they point to important next steps for research in preventing and de-escalating aggressive behavior in acute care settings. The SOE grades informing decision making in this area were minimal. A major void is the lack of well-designed, adequately powered, properly analyzed comparative trials that address questions of preventing and de-escalating aggressive behavior. The validity of findings from the three reasonably well-designed CRTs was constrained by analyses that did not properly control for the clustered nature of the data. We applaud the efforts to conduct comparative trials, but this evidence base does not convincingly show the efficacy of most of these strategies; that fact complicates the design of strong comparative studies and reflects a gap in efficacy data that may need to be addressed first.

Nonetheless, head-to-head trials that compare various interventions with each other rather than with usual care are needed to guide decision making. Most critical is identifying the “right” interventions to compare, which would allow the most efficient use of research time and funding for this topic. More evidence about the differential effectiveness of interventions would allow clinicians and administrators to balance evidence of effectiveness with implementation and resource costs.

Investigators who lead future trials must clearly describe their interventions. Only in this way can other research teams sensibly try to reproduce or replicate such studies and help confirm which components of the interventions are the most (or least) effective. Risk assessment strategies and multimodal interventions, which have some evidence for preventing aggressive behavior, must be described in more detail to allow comparisons with each other and to allow variations within these approaches to be compared as well.

Currently, clinicians and investigators do not know the accuracy of risk assessment tools. Because these tools are necessary to identify patients at high risk of aggressive behavior and, hence, to develop an effective plan to manage aggressive behavior, more work on documenting the measurement capabilities of these tools is needed.

In the future, all trials must report on consistently defined and clinically meaningful outcomes, both short term and long term. Crucial short-term outcomes include reliable and valid measures of aggressive behavior and of seclusion and restraint actions. Using well-established, reliable, and valid assessments of aggression that can be harmonized across studies (and ideally countries) is crucial, as well, for future systematic reviews on these topics. In addition, research teams should increase adherence to the Consolidated Standards of Reporting Trials statement regarding the reporting of clinical trials, including CRTs (55).

Key long-term outcomes must involve more patient-centered outcomes, such as health-related quality of life. Patient perspectives of harms, including treatment preferences, in acute care settings are largely missing from the literature; this gap should be remedied. Measures of the use of health services are important, as are cost implications. Investigators should incorporate factors involving implementation of interventions, such as acceptability, feasibility, and sustainability, into their designs for intervention research in acute care settings.

Available acute care data are almost entirely from inpatient psychiatric settings and settings outside the United States. In the latter case, standard practices, patient populations, insurance coverage, costs, and various other variables may differ in the United States, perhaps considerably. Future well-designed studies of inpatient psychiatric settings must be conducted in U.S. settings. In addition, informative data must be collected from acute care medical and surgical units and from emergency department settings.

Finally, we had no informative data on modifiers of treatment effectiveness. Future studies, including comparative trials, must assess how variables may modify or mediate the effects of the interventions studied. These variables could include age and other sociodemographic or economic factors, specific primary diagnoses (and perhaps coexisting conditions), and explicit treatment components.

Our review had limitations. First, we bounded its scope to focus on data relevant to adults in acute care settings. This emphasis left out consideration of data from chronic care and psychiatric residential settings; it also omitted treatment of children and adolescents. In both these clinical areas, however, use of seclusion and restraint is common and potentially concerning (56). Second, to allow a meaningful synthesis of outcomes, we required that studies report at least one of our main outcomes—change in aggressive behavior or in seclusion and restraint use. This restriction may have reduced the number of eligible studies and the number of patient-centered outcomes we could examine, for example, by omitting outcomes such as improved quality of life and improved therapeutic relationship. However, synthesizing such data with the other collected outcomes would have been difficult and would likely have not affected our SOE findings.

Finally, we excluded reviews and primary studies that examined agitation as the primary outcome when evaluating the effectiveness of inpatient or acute care risk assessment protocols (12). This exclusion limited consideration of interventions to reduce agitation, which may also lead to decreased aggression. An evidence base for reducing agitation exists (39,57), and it may inform aggression management. The decision to focus on aggression and not agitation, although narrowing the scope of our review, also reduced the heterogeneity of the outcomes under examination. Each exclusion decision was made with the intention of focusing the review and controlling for important sources of heterogeneity. By better delineating situations in which agitation leads to aggression, future research can better guide selection of a specific intervention, for example, a psychotherapeutic approach versus medication.