Reducing restrictive practice via the reduction of seclusion has become one of the top priorities of many health authorities (1–3). Although rates of seclusion have fallen worldwide (4, 5), rates of violence on psychiatric wards have not substantially changed with evidence denoting an upward trend (6, 7), arguably indicating that seclusion is a therapeutic option increasingly reserved to contain more extreme cases of violence. The use of seclusion is controversial (8, 9). Generally implemented as a last resort treatment (10), at times alone, more often in combination with other, mainly pharmacological approaches (11, 12), seclusion use is criticized as an unjustifiable and potentially traumatic practice for patients (13–17). These claims, however, remain hotly debated (18). Seclusion, in fact, can be an effective and safe therapeutic tool in selected cases (19, 20) with limited evidence of harmful effects to patients (21–23). A lack of appropriate psychometric tools that standardize an objective seclusion approach to violent patients based on degree of risks is identified as one of the major flaws by its detractors.

In spite of the fact that assessment of risk of violence on psychiatric wards has become much easier and more reliable thanks to the implementation of validated instruments like the Broset Violence Checklist (BVC) (24), or the Staff Observation Aggression Scale Revised (25), the decision to seclude a patient remains mostly an ad hoc process. This is not surprising considering the dynamic nature of a decision-making process (26) that reflects a complex interplay of multiple patient-related (5, 27–29), environmental-related (9, 30–32), and team-related factors (33–35).

Indeed, assessment tools that may predict seclusion as part of a therapeutic care plan have become available (36, 37), but they are research-oriented, and far too impractical to be used in every day clinical practice, as they depend on prior collection of detailed risk information regarding demographic, historical, and behavioral factors. The complexity and length of these tools means they cannot meet the requirements of a quick decision-making process by the frontline staff on the ward, particularly needed when dealing with rapidly unfolding situations (38).

Based on our clinical experience of Psychiatric Intensive Care Units (PICUs), and supported by evidence from the literature on seclusion, we extended on our previous exploratory work (39) on the development of a modified version of the BVC, the East London Modified-Broset (ELMB), taking into consideration two factors affecting staff’s seclusion decision-making when containing high acuity behaviors, that is, patient’s degree of compliance with PRN (pro re nata) medications, per os (P.O.) or intramuscular (I.M.), and response to de-escalation (verbal and/or physical) or lack thereof.

In this case-control study, we retrospectively employed ELMB in a sample of patients consecutively admitted to two PICUs over a 3-year period. The aims of this study were to analyze the predictive characteristics of ELMB with respect to secluding adult psychiatric inpatients at high risk of violence; to describe ELMB’s accuracy levels; its ability to discriminate patients’ degrees of risk; to delineate ELMB as a seclusion decision-making tool. In the absence of a gold-standard seclusion tool, the BVC was used as a benchmark instrument to compare the discriminatory characteristics of ELMB.

The results of this study support ELMB as a promising seclusion decision-making and prediction tool for hic et nunc use on acute psychiatric wards. Three factors significantly predicted the occurrence of seclusion in a sample of acutely disturbed patients, namely attacking objects, physical threats, and PRN compliance. The effect size of these findings was moderate for the first factor, with a large effect for the second and the third factors. In addition, results were consistent across the male and female samples. This can be partly explained on one hand, by the involvement in emergency situations of composite teams with members responding from both PICUs (as well as from other acute wards within the hospital), who, therefore, share common approaches to violent behavior, on the other, by the characteristics common to patients generally admitted to these highly specialized wards. These results are in line with findings from previous studies (5, 27, 28, 55).

Lack of response to de-escalation was statistically significant and has been associated in previous studies with violent behavior leading to seclusion (54, 55). Mann-Poll et al. (57) found that patient’s uncooperativeness played a paramount role in the decision-making process over seclusion. In our study, however, the true magnitude of the effect of response to de-escalation on the final odds of being secluded could not be fully estimated due to issues of separation. Problems of separation in logistic regression are common in situations in which one or more factors in the model perfectly predict the outcome (58). We believe that perfect prediction was at play in this case, as the preliminary assumptions of the BRM were fully met. In addition, results were consistent across the male and female samples. This indicates that response to de-escalation had a disproportionate weight on the decision to seclude a patient and is largely independent of the combined effects of the other behavioral predictors (attacking objects, physical threats, and PRN compliance), because the inclusion in a non-constrained BRM of response to de-escalation alone did not significantly change the final odd ratios. This primacy of response to de-escalation over other factors in driving the decision-making process is a likely reflection of the strict compliance with the Mental Health Act Code of Practice in England (59) and the view that seclusion is a last-resort approach to violent behavior. This is also supported by the large discrepancy between ELMB’s identified optimal threshold to seclusion (Youden Index associated criterion > 4) and staff’s perception of patients’ risk levels appropriate to seclusion (ELMB>6). The items of confused, irritable, boisterous, and verbal threats did not predict the occurrence of seclusion, in line with our clinical experience that patients exhibiting these behaviors do not generally meet the threshold to being secluded.

Patients with history of seclusion were twice as likely to be secluded during subsequent admissions compared to never secluded patients, although the small to moderate effect size of the odds, indicated that a positive history of seclusion had a negligible influence on the decision to seclude an acutely unwell patient. Although this association has not so far been fully investigated, these results are broadly in line with those reported by other investigators. For instance, while Swett (60) described a non-significant trend, Thompson (61) indicated that patients with history of seclusion were more likely to be secluded during future re-admissions, between 43% and 81% of the cases. Georgieva et al. (55) also found that patients subjected to measures of coercion (seclusion and/or restraint) were 2.4 times more likely to have had a history of previous exposure to coercive measures. In the latter study, coerced patients were more frequently male, younger, more severely unwell, and uncooperative. It could be that, as proposed by Thompson (61), known violent patients might make staff less tolerant to aggressive behaviors, which may thus lead more easily to seclusion. However, Thompson did not employ measures of current violent behavior, which makes it difficult to draw a full comparison with the attitude of staff in the present study. Our findings showed that patients who were secluded scored significantly higher on ELMB (>6) than the best estimated threshold to seclusion (Youden Index associated criterion >4), indicating that staff tolerated very high levels of disturbed behavior before using seclusion to manage aggression. Unlike Thompson’s study, our sample included only the first admission to PICU of patients with multiple admissions, which minimized the chances that staff had secluded the same patient during a previous admission and thus reducing the influence of this confounder on the decision-making process. In addition, the effect of this confounder was minimized by the collective nature of the decision-making process that took place within a composite team of staff, several of whom generally did not have any prior knowledge of the patients’ circumstances at the time of the incident.

Substance abuse, common among psychiatric patients (62), and an important contributor to aggression (63–65), has been linked to violent behavior leading to seclusion (27, 66). In our study, no significant difference was found among secluded and non-secluded patients in substance abuse consumption. There was no other difference in demographic and other clinical characteristics among secluded and non-secluded individuals. These results are not surprising and echo the weak and variable associations between violence, use of seclusion, and demographic and clinical characteristics described in other studies (67, 68). However, these results could also be explained by the homogenous characteristics of patients admitted to PICUs. This arguably indicates that the decision to seclude a patient was mainly motivated by the degree of aggressive behavior regardless of other patients’ clinical or demographic background.

East London Modified-Broset and BVC scores were significantly higher among patients who were secluded as compared to non-secluded patients with large ES indicating that both tools were very good at discriminating degrees of aggressive behavior, with similar results in the male and female groups. Despite these similarities, we believe that ELMB has an important additional advantage over the BVC in that it allows a dynamic assessment of the patient’s response to staff’s interventions (last resort component), and allows staff to review behaviors in order to decide on initiating seclusion. Interestingly, these elements are also rehearsed in some of the major international psychiatric guidelines on the use of seclusion (10, 69, 70).

Receiver operating characteristic analyses indicated that the discriminatory power of ELMB was over 12 SDs higher than the BVC’s. The inclusion of patient’s response to staff’s pharmacological and behavioral interventions are the two key-factors setting it apart from the benchmarked BVC for the purpose of seclusion assessment. Regression models indicated that higher scores were more predictive of seclusion outcome, with almost a sixfold increase in the probability of being secluded with ELMB score > 6. The choice to consistently seclude a patient at a higher (>6) than the best identified threshold (Youden Index associated criterion > 4) speculatively delineates a minimum expected standard that is likely to have acted upon by staff to use seclusion as the very last resort, potentially at the cost of increased risks toward staff. Because of the important implications that the adoption of this course of action can have in terms of ward safety, it is not immediately apparent the rationale behind this decision. Beside the strict compliance with the Mental Health Act Code of Practice England (59), an additional reason might be found in the higher therapeutic cost (FPR = 16%) at the lower threshold > 4, which would have meant that up to 16 in 100 incidents could have been potentially misled seclusion events. Because of the inverse relationship between sensitivity and specificity, the choice of a lower threshold would have driven up sensitivity and FPR to the detriment of specificity (FPR = 1—specificity). The FPR is much higher at ELMB > 4 than that at a cut-off > 6 (FPR = 0.7%), where less than 1 incident in 100 was dealt with seclusion. This might have represented a far better cost-benefit option for the staff decision-making. It is the view of these authors, however, that an increase in the FPR would not automatically reflect an inappropriate increase of seclusion events. This is indicated by the ROC model, which identifies as a “true positive” any seclusion incident that has achieved a high score on the ELMB, compared to non-seclusion incidents that were scored low on the ELMB (i.e., “true negatives”). The seclusion of a patient with a low ELMB score (for instance, a patient secluded following an actual physical assault might score as low as 2 or 3) would be identified by the classification model as a false positive when it would indeed be a “true positive” (i.e., the patient needed to be secluded as immediate precaution for staff and patient safety).

An additional limitation lays in the retrospective nature of the study, which limits the generalizability of the findings. However, the sample size was generous and the results consistent across the male and female samples. Employing the BVC as standard of reference against the ELMB might be regarded as controversial given that the BVC was not designed as seclusion prediction tool. This limitation notwithstanding, the BVC displayed reasonable discriminatory properties non-dissimilar from that of other specific, albeit lengthy, seclusion instruments, like the Resident Assessment Instrument-Mental Health (71). Despite these limitations, our study presents several strengths. ELMB isolates well recognized behavioral components of violence and, unlike many other seclusion assessment tools, is short and focuses on how patients respond to staff’s interventions regardless of other demographic and clinical characteristics. As such, our tool is novel and has the potential to fill an important gap in seclusion clinical practice where it can standardize a consistent approach to the implementation of seclusion on the ward.

In the future, a prospective case-control study looking into the validation of ELMB as seclusion prediction and decision-making tool will be required in order to examine its discriminatory properties in real-time analysis. Finally, in a clinical panorama characterized by the absence of any seclusion discontinuation tools, it would be equally important to test ELMB in this aspect of the clinical assessment in light of the important repercussions that such an assessment can have on the safety of the ward and the patient’s care plan.

This study was carried out in accordance with the recommendations of the Joint Research Management Office of Queen Mary Innovation Centre and East London NHS Foundation Trust who gave Ethical and R&D approval. Since the present research was limited to secondary use of information previously collected in the course of normal care (without an intention to use it for research at the time of collection), it was excluded from Research Ethics Committee (REC) review and approval. As such, the present research required only R&D approval by East London NHS Foundation Trust (ELFT). The study was sponsored by East London NHS Foundation Trust: protocol reference KO1308/1.

FL and KM conceptualized the work, interpreted the data, and wrote the manuscript. FL designed the research and acquired and analyzed the data. All authors approved the final version of the present manuscript to be published.

The present research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.