Seclusion and Psychiatric Intensive Care Evaluation Study (SPICES): combined qualitative and quantitative approaches to the uses and outcomes of coercive practices in mental health services

The Biomedical Research Centre Clinical Records Interactive Search database

Psychiatric intensive care unit cohort

Identification of psychiatric intensive care unit cohort cases

Psychiatric intensive care unit cohort services in SLaM are provided by five wards: four in general adult services (three male only and one female only) and one within the forensic inpatient service. PICU cases (n = 986) were defined as all transfers of patients from general adult acute wards to a (non-forensic) PICU ward occurring during the study period of April 2008 to April 2013. Figure 1 summarises how these PICU cases relate to all admissions during the same period. Direct admissions to a PICU from the community were excluded primarily because our method depended on being able to code behaviours in the 3 days prior to entry into the PICU and the available notes for patients just admitted to hospital were considered unlikely to suffice. In addition, as we hypothesised that factors predicting PICU transfer would differ substantially among general adult and forensic wards (and our primary interest was in the former) we also excluded transfers from forensic wards to any PICU and all transfers into the forensic PICU.

FIGURE 1.

Identification of PICU cases (treated) and controls (untreated).

Seclusion cohort

Seclusion cases

The use of seclusion is recorded in free-text progress notes within the electronic patient record rather than in any structured field. Relevant free-text data were therefore extracted and used to identify seclusion spells. Using SQL, we initially extracted progress notes containing the words ‘seclusion’, ‘supervised confinement’ or ‘solitary confinement’. We manually cleaned these to create a database comprising details of all seclusion spells occurring during the study period (n = 1478). Based on the assumption that seclusion practices across general and forensic wards might also differ substantially, seclusions occurring on forensic wards (n = 240) were subsequently excluded from the data set. We also excluded seclusions on other non-PICU wards, which formed only a small proportion of the non-forensic seclusions (n = 248), in order to reduce heterogeneity. Seclusion episodes examined in the current study were therefore those that occurred on the four non-forensic PICU wards (n = 990). Figure 2 summarises the procedure used to construct the seclusion databases.

FIGURE 2.

Identification of seclusion cases (treated) and controls (untreated).

Consequences of how cases and controls were defined

It is important to note that at the analysis level, ‘cases’ refer to transfers into PICU treatment and into seclusion, and not to unique patients. Similarly, as set out above, controls are non-transfers of patients into a PICU or seclusion, rather than patients who are not transferred to a PICU or into seclusion. The same patient could, in principle, feature within a data set multiple times as a case (each time they are secluded or sent to a PICU) and also as a control (on any day during which they are not transferred). This respects the fundamental principle of case–control design, which is that any patient who is at risk of the outcome of interest should be eligible for selection as a control. The fact that an individual person may contribute more than one observation to the data set may create clustering within the data, a feature that must be taken account of by the selected method of analysis.

Extraction of predictor variables

Statistical analyses

All analyses were conducted using Stata. The same procedure was used for the analysis of PICU and seclusion use; in all analyses, the BRC ID was included as a random effect in order to account for clustering at the patient level (i.e. within each data set a single patient could represent multiple cases or controls, by including the BRC ID as a random effect, correlations between data observations obtained by the same patient were accounted for). Univariable logistic regression analyses were first performed to examine associations between all predictor variables (demographic/clinical factors and behavioural precursors) and PICU/seclusion status. We then performed a multivariable analyses for each outcome (PICU and seclusion) that included all predictor variables, irrespective of whether or not they were significantly associated with the outcome in univariable analyses. We subsequently conducted exploratory multivariable analyses that included interaction effects between sex and all other predictor variables; that is, all predictor variables and all interaction effects (sex with all other predictors) were entered simultaneously, with subsequent removal of all effects with p ≥ 0.05. In the analyses presented in this chapter, we did not apply probability weight – as is typical in case–control studies – and therefore the intercept values are uninterpretable (weights were used in the derivation of propensity scores in Chapter 3).

Predictors of psychiatric intensive care unit transfer

Predictors of seclusion use

Strengths and limitations

Our analyses examining the predictors of transfer to a PICU and seclusion have a number of strengths. First, the sample sizes that we achieved were distinctly in excess of those in other previous studies, increasing the precision of the estimates that we obtained. Second, we combined novel and robust measurement and sampling techniques so that we could estimate the effect of the time-varying behavioural factors that we assumed to be critically involved in selection for treatment, as well as the usually available demographic and basic clinical factors. There were, however, a number of notable limitations.

The current study was conducted within a single NHS trust, which potentially limits the extent to which the current findings can be generalised to other psychiatric hospitals (particularly those outside the UK). However, within this single NHS trust, we were able to examine practices across four PICU wards, including a female-only ward; thus, our findings may have greater generalisability than those of previous studies.

Although we examined a wide range of behavioural precursors that were identified directly from clinical events preceding treatment, we may have failed to account for some low-frequency behaviours that were not present in the initial screening subset but that may, nonetheless, be important precursors of treatment. Of note, we did not examine suicide or self-harm behaviours, both of which have been reported as antecedents of transfer to a PICU and of seclusion in previous studies. A further limitation relates to the fact that we focused only on patient characteristics. It is likely that a range of environmental factors influence the decision to initiate PICU transfer and seclusion (e.g. number of staff, staff sex, bed numbers); however, it was beyond the scope of the current study to examine these variables. Thus, we were unable to determine the impact of patient factors after accounting for external factors beyond the patient’s control. Environmental factors are particularly important as they are often dynamic (i.e. amenable to change) and, therefore, offer the opportunity to identify ways by which PICU and seclusion practices might be modified. A more difficult issue may be the potential for treatment selection to determine the recording of apparently relevant behaviours; these records may be as much accounts of a decision already taken as disinterested reports of behaviour. Naturally, there are normal concerns about the accuracy of data taken from electronic patient records, although in most cases these would be expected not to result in bias.

Finally, based on previous studies, we explored whether or not there might be important interactions between patient sex and other predictors of PICU/seclusion status. However, we did not examine other interaction effects identified in the extant literature (e.g. interactions with ethnicity and diagnosis).

Demographic and clinical predictors of treatment

Behaviours associated with use of a psychiatric intensive care unit and seclusion

Implications

Previous studies have indicated that specific patient subgroups are at increased risk of PICU transfer and seclusion. However, these studies have typically lacked appropriate control groups and have also failed to account for patient behaviours that might explain this elevation in risk. In the current study, we found little evidence of referral bias in relation to ethnicity or diagnosis; after adjusting for a range of behavioural precursors and demographic/clinical factors, these factors were not associated with transfer to a PICU or seclusion. One exception to this is that patients with bipolar disorder were twice as likely to be transferred to a PICU than patients with schizophrenia, even after adjusting for behaviours typically associated with this diagnosis (i.e. mania, agitation and irritability). Presumably this simply reflects a degree of disturbance that our measures were not able to detect. However, further work to determine the reasons why clinical teams consider patients with bipolar disorder to be particularly difficult to manage in general adult wards might potentially bear fruit if it leads to strategies that can help to avoid PICU transfer.

One interesting finding was that even after accounting for a range of confounders, sex differences in the risk of transfer to a PICU and seclusion were still apparent, with men being more likely to be transferred to a PICU, whereas women demonstrated no difference in the use of seclusion that could be not be accounted for by the inclusion of other variables such as ward and time period. Overall, the influence of ward was substantial in the case of both transfer to a PICU and seclusion, supporting a key assumption of this project, which is that major differences in the use of coercive practices can coexist in apparently similar services and units.

Our findings emphasise the importance of adjusting for a wide range of demographic/clinical factors and behavioural precursors when conducting any non-randomised analyses examining the effects of transfer to a PICU and seclusion on outcomes. The work presented in this chapter, therefore, has important implications for the analyses performed in Chapter 3. Thus, having identified a range of factors that clearly distinguish treated patients (i.e. those receiving PICU care or seclusion) from untreated patients, it is essential to adjust for these variables when examining the effect of PICU care and seclusion on adverse incidents, length of stay and costs.

Conclusions

The findings presented in this chapter indicate that there are a number of demographic and clinical factors that distinguish between patients who are subject to PICU transfer/seclusion and those who are not. Moreover, some of these factors (notably, patient sex, age and time since admission) remain consistently associated with treatment receipt after adjusting for a range of behavioural precursors. Patient sex was found to modify the effect of other demographic/clinical factors and behavioural precursors on treatment receipt, and was also significantly associated with both PICU and seclusion duration. These findings highlight the importance of patient sex when examining predictors and outcomes of these treatments.

Identification of treated and untreated groups

Full details of the methods used to identify treated (cases) and untreated (controls) groups are provided in Chapter 2. The PICU case group comprised all transfers from general adult wards to non-forensic PICU wards during the study period (n = 986); PICU controls were randomly selected person–date combinations taken from the set of person–day combinations on general adult wards where PICU transfer did not occur and where the person had not just been transferred back from a PICU (n = 994). Similarly, the seclusion case group comprised all episodes of seclusion occurring on non-forensic PICU wards (n = 990); seclusion controls were randomly selected dates corresponding to periods of treatment in non-forensic PICU wards where seclusion was not in progress, started or ended (n = 1032).

Extraction of covariates

Demographic and clinical data for all cases and controls (PICU and seclusion) were extracted from CRIS using the procedure described in Chapter 2. These data were used to derive the following variables: age, sex, ethnicity, diagnosis, MHA section (at midnight on the day of sampling) and time since admission for the current SLaM inpatient episode. Chapter 2 additionally provides details of the two-stage process used to identify behavioural precursors of PICU transfer and seclusion from clinical ‘Events’ recorded prior to treatment (0, –1 and –2 days before the date of PICU transfer or date of seclusion) and the procedure used to extract, code and clean the final pre-PICU and pre-seclusion data sets, comprising behavioural precursor data for all cases and controls.

Extraction of violence and aggression outcomes subsequent to transfer/non-transfer

Extraction of length of stay subsequent to transfer/non-transfer

In order to examine the effect of PICU transfer and seclusion on length of stay, we excluded that part of the admission occurring prior to PICU transfer/seclusion, defining length of stay for our purposes as the difference between the sample date (the date of transfer to PICU/seclusion for treated patients and the randomly selected date for controls) and the end of the admission.

Extraction of economic outcome data subsequent to transfer/non-transfer

Separately, we obtained SLaM internal financial data for the financial years covered by the study, comprising costs of each separate SLaM team or ward for each financial year. These data were examined carefully, locating budgets shared between services and services funded by more than one budget. Having grouped services and budgets when necessary, we extracted and totalled all service use data for the team(s) or ward(s) corresponding to each budget: at the ward level, we extracted inpatient nights and for teams, which were community based, we extracted face-to-face contacts. These data were used to create a unit cost for each service, along with a currency (inpatient night or face-to-face contact). The data were fully cleaned for inpatient services, general adult community mental health teams (CMHTs) and home treatment teams, and therefore encompassed the overwhelming majority of spending for adult mental health services. In the data extraction, we extracted total costs under each of these headings for 7, 30 and 365 days subsequent to the sampling date. We separated out PICU days and also separated PICU days into secluded and non-secluded, allowing for the possibility of sensitivity analyses modelling different relative costs of secluded and non-secluded time.

The primary measures for the economic analysis were therefore:

1. total health-care costs, including non-PICU inpatient care, seclusion and non-seclusion inpatient care within a PICU, community care provided by CMHTs, Mood, Anxiety & Personality outpatient and day services, psychotherapy care and home treatment care

2. health outcomes, including total 7-day general aggression and violent incidents, and 30-day total serious incidents

3. cost-effectiveness analysis of PICU transfer and of seclusion.

The total health-care costs were analysed for different time periods of 7, 30 and 365 days. The cost-effectiveness analysis considered only the weekly and monthly total costs, as these could be linked to the corresponding health outcomes.

Statistical analyses

Effect of psychiatric intensive care unit transfer on adverse incidents

Analysis of propensity scores

The distribution of propensity scores was markedly unequal between treated and untreated observations (Figure 3). Not one untreated observation fell in the interval 0.1–0.9. Even though the sampling frequency for untreated observations was 0.0017, equivalent to approximately 1/588, it seemed not only that there was no possibility of deriving a causal effect from the data set, but also that further samples would not necessarily find a usable number of comparison observations.

FIGURE 3.

Box plots of propensity scores for PICU: treated vs. untreated. Added lines at 0.1 and 0.9 represent the limits established in Crump et al. 34 as equivalent to the common support condition in standard matching and regression designs.

Effect of seclusion on adverse incidents

Analysis of propensity scores

The distribution of propensity scores was unequal between treated and untreated observations (Figure 4). However, 26 untreated observations fell in the interval 0.1 to 0.9. The sampling frequency for untreated observations was 0.016, equivalent to approximately 1/63, suggesting (by multiplication) the existence of 1638 candidate untreated observations within the entire sampling frame. However, the number of usable untreated observations was too low to perform adequately powered analyses.

FIGURE 4.

Box plot of propensity scores for seclusion: treated vs. untreated. Added lines at 0.1 and 0.9 represent the limits established in Crump et al. 34 as equivalent to the common support condition in standard matching and regression designs.

Effect of psychiatric intensive care unit on length of stay

The extent to which those treated with PICU transfer and those untreated differed in length of stay between the sample date and the end of admission was estimated using unadjusted and adjusted random-effects linear regression, with SEs estimated using 1000 bootstrap samples. In the unadjusted analysis, PICU status was significantly associated with higher length of stay [b = 22.05, 95% confidence interval (CI) 6.35 to 37.75; p < 0.001].

Effect of seclusion on length of stay

The extent to which those who were placed in seclusion and those who were not differed in length of stay was estimated as for PICU transfer above. There was no significant unadjusted association between seclusion status and subsequent length of stay (b = 0.19, 95% CI –13.43 to 13.82; p = 0.98).

Psychiatric intensive care unit economic analyses

Seclusion economic analyses

Strengths and limitations

Notwithstanding the very obvious limitations of our findings, our methods were scrupulous. Without our sampling design and our associated strategy for measuring time-varying covariates, it would not have been possible to calculate propensity scores and, without those, we would not have been aware of the extent to which regression-based estimates would have been biased.

Implications and conclusions

The main implications are for further research. Our findings in relation to seclusion indicate that a modified design applied to our existing sampling frame – one that is not, for pragmatic reasons, limited to using the sample for investigation of predictors and investigation of effects – could yield a usable estimate of effect, bearing in mind the usual caveat regarding unmeasured confounding. It would be necessary to select a group of controls whose probability of treatment was similar to the small group identified in our analysis as having a conditional probability of treatment greater than 0.1, perhaps using more or less sophisticated text-mining approaches. In the case of PICU transfer, the most appropriate approach would be to study a number of provider organisations, including wards that have easy and frequent access to a PICU and wards that do not. Assuming a lack of systematic differences in behaviour between PICU- and non-PICU-using wards, this approach might be expected to yield a higher number of non-PICU untreated controls than was the case in the present study. Once plausible effect estimates have been calculated based on non-randomised data, it will be possible to judge whether further non-randomised or randomised studies are needed.

Aim

The purpose of this study was to examine the use and acceptability to staff of a range of containment methods currently utilised in acute psychiatric wards, as well as exploration of speed of initiation of manual restraint across service configuration, dependent on access to PICUs and seclusion.

Participants

Eight hospitals providing inpatient acute psychiatric care participated in the study. The hospitals were identified in a purposeful sample to include two of each of the following: (1) no seclusion and restricted PICU access, (2) no seclusion and full PICU access, (3) seclusion available and restricted PICU access and (4) seclusion available and full PICU access. In order to ensure greater national representativeness, half of the sample was drawn from hospitals in the north west of England and half from hospitals in Greater London. The study participants were acute ward staff members (qualified nurses, n = 130; health-care assistants, n = 69; others, n = 7) who were drawn from the eight hospitals included in the study. All eligible members of staff on duty during frequent researcher visits to the study wards were invited to participate, of whom 206 from 18 wards took part.

Measures

Procedure

Ethics approval was granted by the King’s College Ethics Committee, with NHS research and development approval obtained at each participating trust. After a complete description of the study, written informed consent was obtained. Staff members completed the study at their hospital site, on the ward on which they worked. After completion, participants were asked not to discuss the tasks with other staff members in order to prevent contamination. Testing took part in a quiet room; participants were asked to complete paper versions of both the demographic questionnaire and the ACMQv2, and the MAPAT was administered on a laptop computer. The participants were debriefed and thanked for their time.

Analysis

Spearman’s rank-order correlations were calculated to determine the relationship between service configuration and the items from the demographic questionnaire. Significant associations between service configuration and demographic variables were further examined using logistic regression modelling with seclusion provision and PICU provision.

Spearman’s rank-order correlations were calculated to determine the relationship between service configuration and individual items of the ACMQv2 with significant associations further tested using logistic regression.

Reactions during use of the MAPAT had a bimodal distribution and scores were categorised to match their distribution as follows: (1) ≤ 224 seconds, (2) 225–250 seconds and (3) ≥ 251 seconds. In time frame 1, a patient displays signs of agitation by pacing, fidgeting and becoming agitated when a nurse attempts to verbally de-escalate. In time frame 2, the patient displays similar agitation and is verbally abusive and threatening to the nurse. In time frame 3, the patient hits a piece of furniture and shoves a chair out of the way while approaching the nurse, finally attempting strangulation. Spearman’s rank-order correlations were run to determine the relationship between MAPAT time-to-restraint and other questionnaires. Chi-squared tests were performed to explore the relationships between MAPAT score and use of containment method. Using ordinal regression, MAPAT score was modelled using seclusion provision and PICU provision as predictors.

Demographic information

Table 8 summarises the demographic features of the sample (counts and percentages).

The absence of a seclusion room was associated with younger staff (r_S = –0.16, n = 204; p = 0.021) and fewer years’ experience of working in psychiatry (r_S = –0.14, n = 204; p = 0.040). When further tested in a logistic regression equation with seclusion as the dependent variable and controlling for PICU access, neither of these two variables remained significant.

The absence of an on-site PICU was associated with higher numbers of female staff (r_S = 0.15, n = 204; p = 0.028) and greater exposure to mild physical violence (r_S = –0.14, n = 206; p = 0.047). When tested in a logistic regression controlling for seclusion availability, only sex (p = 0.034) remained significant.

Attitude to Containment Methods Questionnaire

The means and SDs of each item from the ACMQv2 are illustrated in Table 9. The containment methods have been ranked in order of acceptability, from most acceptable to least acceptable.

Psychiatric intensive care unit, intermittent observations and PRN medication received the highest approval ratings, whereas mechanical restraint and net beds received the lowest. Approval scores with the greatest variability were open-area seclusion, mechanical restraint and seclusion.

Access to a seclusion room was associated with greater acceptability of seclusion as a method of containment (r_S = 0.25, n = 198; p < 0.001) and lower acceptability of open-area seclusion (r_S = –0.23, n = 199; p = 0.001). When tested in a logistic regression controlling for PICU access, seclusion acceptability remained significantly associated with seclusion availability (p < 0.001); however, open-area seclusion acceptability was no longer significant.

Participants were asked to identify whether or not they had ever used any of the 11 methods of containment indicated by a response of ‘yes’ or ‘no’. Frequency (and percentage) of total responses can be seen in Table 9. Intermittent observation, constant observation and manual restraint were the methods that had been used by most members of staff, whereas mechanical restraint and net beds had been used the least.

The availability of a seclusion room was associated with a greater reported use of seclusion (r_S = 0.548, n = 196; p < 0.001) and time out (r_S = 0.152, n = 200; p = 0.032), and a lesser use of open-area seclusion (r_S = –0.181, n = 201; p = 0.010). When entered into a logistic regression controlling for PICU access, greater reported use of seclusion (p < 0.001) and less open-area seclusion use (p = 0.001) remained significant, whereas reported time-out use did not (p = 0.715).

The availability of an on-site PICU was not statistically associated with any containment method acceptability score. The availability of an on-site PICU was associated with less reported use of open-area seclusion (r_S = –0.154, n = 201; p = 0.029). This usage association remained significant (p = 0.048) when tested in a logistic regression equation controlling for seclusion availability.

Moylan Progression of Aggression Tool

Using Spearman’s rank-order correlation, MAPAT time to restraint was not associated with demographic information or details of current post. Table 10 shows the frequency (and percentage) of responses for each of the three time frames during use of the MAPAT across seclusion and PICU provision.

Moylan Progression of Aggression Tool timings were inversely associated with seclusion availability (r_S = –0.258, n = 186; p < 0.001) but were not associated with PICU availability. Using logistic regression with seclusion availability as the dependent variable, controlling for PICU availability, MAPAT times remained highly significant (p < 0.001). When there was no seclusion room available, staff took longer and allowed a greater degree of escalation before initiating restraint, as judged by their MAPAT scores.

Moylan Progression of Aggression Tool scores were also explored in relation to ACMQv2 scores. The MAPAT timings were positively associated with participants’ judgements of mechanical restraint acceptability (r_S = 0.190, n = 179; p = 0.011) and net bed acceptability (r_S = 0.168, n = 177; p = 0.025). A longer time before restraint was initiated was associated with greater acceptability of these containment methods. MAPAT scores were not associated with the reported use of any of the containment methods on the ACMQv2.

Service configuration is associated with acceptability of some containment methods

The acceptability and use of seclusion is related to its access. No such associations were found between PICU access and its acceptability and use. Seclusion use has been shown to increase when a seclusion room is directly available to the ward,2 suggesting that with first-hand experience of seclusion room use, staff members are more likely to approve of it as a method of containment. Seclusion is regarded as more effective in aiding with emergency situations than other methods of containment,46 and those with access to a seclusion room reported that without the use of this room, the unit could not operate effectively. 55 It is also possible that some form of cognitive dissonance56 process might underlie this association, with nurses exposed to, and therefore involved in, seclusion use, shifting their beliefs to fall in line with their behaviour. One study suggested that the availability of a seclusion room made staff believe that they were providing more effective care, with the use allowing staff to become more accustomed to it, leading them to rate seclusion as less intrusive to patients, compared with staff who had never secluded a patient on the same site. 57 Alternatively, it is possible that members of staff with strong feelings against seclusion room use perhaps avoid working at wards with direct access to one. All things considered, it is possible that seclusion rooms are being overused at sites with direct access to one, with evidence that some staff members conform to the use of seclusion rooms when seclusion rooms are available, feeling discriminated against if alternative methods are suggested. 58 The concern that seclusion could be abused, for example by being overused when available or used as a substitute when staffing levels are decreased, has been expressed by some authors. 51,59 This has important implications, as patients rate seclusion as unacceptable when compared with many other methods of containment. 4

For those without direct access to seclusion, staff members are more likely to approve of open-area seclusion and this method of containment was more commonly used by staff on wards without on-site seclusion and a PICU. Open-area seclusion is more often referred to as ‘nursing in a side room’ or as the use of an ‘extra care area’. The ACMQv2 defines seclusion as ‘a patient being isolated in a locked room’ and open-area seclusion as ‘a member of staff stays in the locked room with the patient’. Both seclusion and open-area seclusion fit under the umbrella term of seclusion in 2002 English guidance40 and it is possible that wards without defined seclusion rooms are simply substituting this for a different type of seclusion.

Seclusion provision, and not psychiatric intensive care unit provision, is associated with time to restraint in response to aggressive behaviour while using the MAPAT

Time to restraint in response to aggressive behaviour while using the MAPAT was strongly and significantly associated with seclusion provision but not PICU provision and, in places without seclusion, there was a longer time lapse before staff initiated restraint. In units without seclusion, staff members tolerated higher levels of aggression before choosing to restrain while using the MAPAT. Previous studies have suggested that staff without access to seclusion rate their methods of containment as less effective in resolving emergency situations. 46 This lack of confidence could explain delayed time to restraint during the MAPAT. There was no association between PICU access and MAPAT score, despite the fact that on-site PICU availability leads to increased use. 10 This may be because, unlike seclusion, which can be utilised immediately, a transfer to a PICU takes some time to organise and commonly occurs after the immediate crisis is over.

Other findings

There was no association between MAPAT time to restraint and manual restraint acceptability or between MAPAT time to restraint and use of manual restraint. Therefore, it was not the acceptability of restraint that was driving the difference in MAPAT scores, but perhaps more likely a rational calculation about managing outcomes, on which seclusion availability seems to be an influence.

Although mechanical restraint and net beds remain the two containment methods with lowest acceptability ratings, greater acceptability was associated with longer time to restraint during the MAPAT. Although the reasons for this are unclear, it is possible that staff members who are less judgemental of these methods of containment tolerate more extreme patient behaviours and, thus, react more slowly.

The current study found that initiation of restraint was not associated with exposure to either mild or severe physical violence. This conflicts with previous research by Moylan and Cullinan49 using the MAPAT, in which staff members who had suffered from injury at work took longer to initiate restraint than those with no history of injury. The authors suggested that it was fear itself that delayed the restraint process. Moylan and Cullinan’s study considered associations between injury and serious injury, in which type of injury was clearly defined (e.g. evidence of fracture). 49 Our study did not ask participants to be so detailed with their exposure to physical violence and was more subjective in comparison. Therefore, the different methods of investigation in these studies may account for the differences in findings.

Clinical implications

This study raises important questions about clinical practice, particularly the links between the availability, approval of and use of seclusion, coupled with the faster use of manual restraint as judged by the MAPAT. However, none of these findings constitute evidence that seclusion can be safely abandoned. Faster restraint may, in fact, be safer for staff and patients. Secluding a patient might be safer than not doing so (although findings elsewhere in this report suggest that this is not the case).

Limitations

The sample was representative of two urban regions in England (London and the North West). Not all staff participated in the study, with 9.71% of staff not completing the MAPAT. Some degree of response bias may be a possibility. Participants may have previously worked at hospitals with or without seclusion/PICU availability and this may have impacted on the results. ACMQv2 scores are valid and have previously been confirmed to be related to usage; however, generic acceptability ratings ignore potential variation by specific types of behaviour such as aggression, self-harm or mania. Different scenarios may influence judgments of acceptability. Although the MAPAT has been rigorously developed, the extent to which MAPAT scores correlate with actual restraint use in practice is not known, and nor are there any criteria for judging what score represents the optimum or best for a safe outcome. As such, the validity of the MAPAT is unclear.

Aim

Comparing units with differing access to a PICU and seclusion we (1) describe how the same patient high-risk behaviours are managed; and (2) explore and identify differing thresholds for the commencement of severe containment methods (manual restraint, rapid tranquillisation, seclusion and PICU transfer).

Objectives

To compare the management of disturbed behaviours in four different service configurations: (1) no seclusion and restricted PICU access; (2) no seclusion and full PICU access; (3) seclusion available and restricted PICU access; and (4) seclusion available and full PICU access.

By seclusion available we mean that there is a defined seclusion room available to acute wards, which is on the same hospital site as the acute ward. By restricted PICU access we mean that a PICU may be available to the acute wards, but is on another site and/or in a different organisation and/or privately provided.

Specific hypotheses explored

The absence of seclusion and/or restricted access to PICU care are managed through one or more of the following four mechanisms.

1. Substitution: seclusion for PICU (or vice versa) or other alternatives such as medication (regular in higher dose, more intensely sedating medications and/or routes of faster or more effective absorption, intramuscular/intravenous) and/or long duration of manual restraint, time out and open-area seclusion (confinement of a disturbed patient in a separate area together with nursing staff) of some form.

2. Early intervention: faster response to escalation, earlier physical intervention, leading to safer management and more benign outcomes, possibly accompanied by temporarily increased nurse staffing levels or the use of intensive forms of constant special observation (such as two staff to one patient).

3. Therapeutic intensity: deployment of a range of psychotherapeutic and behavioural interventions to manage, ameliorate and reduce the frequency of risky patient behaviours.

4. Non-standard transfer: move of the patient to another ward providing more secure or intensive care in some fashion, perhaps in a private hospital or in a neighbouring trust, or via a secure rehabilitation ward or a forensic ward of some type.

Design

Qualitative data collection was undertaken through interviews of nurses on the management of disturbed behaviour. The interviews used standardised vignettes to elicit what typically happened on the interviewee’s ward in response to particular types of disturbed behaviour. Interviews were conducted within the four different service configurations, systematically coded and subjected to quantitative analysis. The design therefore includes elements of both qualitative and quantitative design, straddling the border between the two.

Sample and sample size

Eight hospitals providing inpatient acute psychiatric care were identified in a purposeful sample, two in each of the following categories: no seclusion and restricted PICU access; no seclusion and full PICU access; seclusion available and restricted PICU access; and seclusion available and full PICU access. In order to ensure a less London-centric sample and greater national representativeness, half of the sample was drawn from hospitals in the north west of England, and half from hospitals in and around London. All NHS mental health hospitals in these areas were identified and one hospital in each category was randomly selected for inclusion (with replacement for any refusals). Thus, in the North West, four randomly selected hospitals (one in each category) were included in the study, and this was the same in London.

Our intention was to draw a random sample of 30 qualified nurses from acute wards in each category of hospital to participate in the threshold interview. This figure was a compromise between achieving a degree of representativeness of the nursing workforce at each hospital and, thus, being able to accurately describe their patient management systems and techniques, and being able to comprehensively manage the data analysis of a large number of interviews centrally.

These plans changed during execution in two respects. Development of the interview indicated that patient sex might influence the pathway and some randomly selected hospitals had separate male and female PICUs available on different sites. We therefore limited the sample to PICUs taking male patients and the sample of interviewees to all acute wards at the study site with male patients. The sample therefore included mixed-sex wards but not female-only wards. Second, and partly as a consequence, the eligible number of qualified nurses was smaller than anticipated at some sites and, instead of randomly selecting participants, we invited all eligible staff to participate.

Description of the Threshold and Escalation Pathway Interview and how it was utilised

The Threshold and Escalation Pathway Interview (TEPI) is a fully structured interview schedule that seeks to elicit from the interviewee the typical escalation pathways of interventions used to control and manage disturbed behaviour on the ward where they work. As such, it presented the interviewee with a number of behavioural scenarios and asked them to respond by describing in detail what typically happens and in what order, when this behaviour occurred on their ward. Three cases, each exhibiting a number of challenging behaviours, were presented one by one to the interviewee.

1. Mike (mania): medication refusal, public nakedness, giving away expensive property, shouting, attempting to abscond, sexual assault.

2. Steve (self-harm): suicidal ideation, scratching self, cutting self, attempt to hang self.

3. Alan (aggression): verbal abuse, property damage, wielding a club, assaulting a fellow patient, assaulting a fellow patient very seriously.

Each behaviour was presented independently to the interviewee for their description of responses, that is, free standing and independent of previously presented disturbed behaviours. The TEPI also contained additional questions to explore the possible impact of staffing levels, sex of the patient, vulnerability of the victim, intoxication, what explanation the patient might give, patient history of violence, etc. The interview was digitally recorded on the wards that were studied. Handwritten notes and check boxes were also completed by the interviewer as part of the process.

The TEPI is designed to be administered to qualified psychiatric nurses only and requires several hours of detailed study by potential interviewers, coupled with discussion and training. An online training module was designed as part of the research. Conducting a TEPI takes approximately 90 minutes.

The full TEPI and the instructions for its use are provided in Appendices 1 and 2, respectively.

Development of the Threshold and Escalation Pathway Interview

The development of the TEPI was a lengthy, sequential and iterative process. The initial ideas for the cases and their potential stages of increasingly difficult behaviour were developed via small group work by the mental health nursing research team under the direction of the principal investigator. These were subsequently synthesised and changed into a viable schedule, with some cases being dropped and others proposed, by the core SPICES collaborators. These then went back to the wider research team who split into five groups of three people (one note-taker, one subject, one interviewer) and the interviews were tried. A significant number of lessons were gained from this and the interview was revised. At this stage it was sent out to all the SPICES collaborators for comments and feedback and was taken to consultation with SUGAR (Service User and Carer Group Advising on Research), a service user and carer group,63 both leading to further amendments. Four trial interviews were then conducted with experienced clinical nurses by the principal investigator. These were recorded and carefully reviewed and, after each interview, changes were made to the TEPI schedule. Following these pilot interviews, the schedule was again presented to the main mental health nursing research team, to a group of prevention and management of aggression trainers and was reviewed by the SPICES qualitative research expert (AQ). These led to the final revision and the form of the TEPI used during the study: the interview schedule and its accompanying instructions that formed the basis for training interviewers. This TEPI development process took 8 months in total.

The issue that was the key focus throughout the development process was that of standardisation. It became apparent that many things might influence the direction of the escalation pathway in actual practice: the age, sex and ethnicity of the patient; nurses’ previous knowledge of the patient; their legal status; their history of violence; what the patient says about their behaviour; what other patients in the vicinity do; the patient’s degree of resistance to the interventions the staff deploy; the numbers of staff on duty; the general frequency of difficult behaviours on the wards, etc. All of these had to be standardised during the interview so that the influence of seclusion and PICU availability could be made visible. Additional care needed to be taken with the presentation of the problem behaviours and instructions produced on how to respond to a number of different but common ways that interviewees took the interview in directions not required by the research. Terminological differences had to be clarified, and at one point it became clear that many containment names in common use did not have any common meaning, necessitating extra questions in the TEPI to elicit the precise actual meaning of the words used by interviewees to describe interventions.

Data quality

The TEPI is a long and complex interview. It was conducted by a number of different interviewers (eight). Although all were trained in the same way to conduct the TEPI, each brought their individual experience of inpatient psychiatry, clinical work and their academic background to the way in which they conducted the interview. These factors in combination led to a number of identifiable problematic issues when the recorded interviews were being processed for analysis.

The most prominent difficulties were associated with changes to the way in which the interview was administered, either by adding questions or leaving them out. On the one hand, prompts were given as to possible interventions, such as ‘would you increase his level of observations?’, ‘would you restrain him?’ and ‘would you call the police at this point?’, even though interviewers had been asked not to do this. Such prompting seemed to be particularly prevalent in interviews conducted by clinically experienced nurses. When entering codes for analysis, the default position was to ignore any step that was prompted by the interviewer; however, these were sometimes included when they had already been given as steps in a previously covered similar pathway. On the other hand, interviewers were required to ask what would happen if an already discussed behaviour was to be repeated during the same shift, in the following shift and throughout the day, but we found that these three levels of repeats were often collapsed or asked idiosyncratically or not at all. These omissions seemed to be dependent on the perceived level of co-operation of the interview respondent.

More minor difficulties included large variation in the detail included when exploring different steps included in the pathway, such as who carries out the management method, what disciplines, how many staff are included, what exactly do they do and a full and detailed description. This was largely dependent on what the interview respondent was asked. The missing information was often relatively unimportant and not central to the analysis, but was problematic in the area of manual restraint practice. Planned explorations of resistance to the use of a low stimulus environment and how Mike’s treatment might have been different if he had a known history of violent behaviour were also not always discussed.

In rare cases, and contrary to their instructions, the interviewer would challenge the response of the respondent. Some examples included ‘do you really have time to do this?’ and ‘is that what would really happen, though?’. These responses were coded according to the initial response before the challenge.

It was typical for interviewees to reply in the first person singular, for example ‘I would – ‘. This often produced a response based on what they would do or what they believed should happen. The interviewers were asked to prompt the participant in the direction of what would typically happen on this ward. Although this did happen in most of the described cases, the use of this prompt was not absolutely consistent.

On occasion, interviewers introduced variations in delivery of the behaviours and some would extend the scenario or indulge in additional questioning, as would be more typical in a semistructured interview. For example, in the case of Mike, when he tries to abscond, one interviewer asked ‘what would happen if he actually escaped?’ and, in other cases, additional questions such as ‘and what if PICU [transfer] didn’t happen?’ and ‘but what if they didn’t come and you were alone?’. In such cases it felt as if the interviewer was probably curious, rather than intending to change the outcome of the scenario. The produced material was not included in the coding or analysis.

On occasion the interviewee would introduce more extreme behaviours such as ‘if he got violent we would – ‘. Usually (but not always) the interviewer would prompt the respondent back to the correct behaviour. Examples that were not controlled included ‘if Mike escapes we would . . .’ and ‘if he started assaulting people we would . . .’. In these examples and others like them the suggested pathways were simply ignored in the coding process.

These problems with data collection had several consequences. The use of the information on repeated behaviours by patients (in other words, what would typically happen on the ward if the patient performed the behaviour again and again) was avoided as much as possible. These data were considered to be less reliable than the first-line responses. A detailed evaluation of different manual restraint technique usage, particularly the issue of long duration restraint, could not be properly compared across categories of hospital, as in many cases details were sparse.

Some areas for which it was hoped to have further information from the interviews in addition to the main research questions about the impact of seclusion and PICU availability were not covered consistently enough to be usable.

Data analysis

The initial plan was that the interviews would be professionally transcribed and imported into qualitative data analysis software. However, the interviews were structured in such a way that this proved to be unnecessary and coding could be conducted directly from the digital audio recordings.

The interviews were thematically analysed for intervention themes (the treatment and management methods described by respondents). This generally resulted in a two-level hierarchy of codes, with the highest level being the most general and the lower having greater specificity. For example, the higher-level code ‘observation’ included a multiplicity of lower-level codes covering intermittent to continuous observation utilising various numbers of staff and differing time intervals. De-escalation, manual restraint and several other common intervention methods were hierarchically broken down in the same way. The codes and definitions utilised were initially derived from those used during the Sequences of Conflict and Containment Events Study (CONSEQ). 3 Additional codes were proposed by researchers coding the data based on the content of the interviews. These were reviewed with the principal investigator and the wider mental health nursing research team, who agreed a final definition for each code that was then deployed. The creation of additional codes ceased owing to saturation around half-way through the data entry process.

The coder would listen to the digital recording of the interview while examining the written notes of the interview taken by the interviewer. From this they would draw up on paper the escalation pathway as depicted by the interviewee. Codes were then entered directly into SPSS version 22 (IBM, Armonk, NY, USA) with their relative order in response to a specific challenging behaviour scenario and whether or not they were specified by the interviewee as a response to the first time the patient exhibited that behaviour or in response to repeats of the same behaviour.

It was initially planned that the escalation pathway of intervention responses to a particular challenging behaviour at any one hospital site could be summarised onto some form of network diagram or flow chart, so that these could be directly compared. Although this was attempted with the data from the first full hospital to be entered, it proved to be impossible to be confident that this was truly representative, given the huge variation in pathways between interviewees. A variety of other means of representing the data were tried in order to extract a visible pattern that could potentially be compared across hospitals, including heat maps, line graphs of mean order by intervention, sequence index plots and simple intervention frequency bar charts. This last simple method was more useful, but the variability highlighted that statistical testing would be required to identify which differences between hospital types were meaningful.

The frequency of interventions was statistically tested across the hospital categories using chi-squared tests. The frequencies were tabulated so that the differences could be seen in conjunction with the results of the statistical tests. When appropriate, multivariate models were applied using logistic regression. More complex multivariate modelling was not conducted in the light of both the small numbers of interviewees at each category of hospital and the difficulties already described in the standardisation of the data collection. Microsoft Excel, SPSS and Stata were all used during the analysis.

Nurses completing the Threshold and Escalation Pathway Interview

Eighty-one TEPI interviews were completed at the eight selected hospitals. The numbers of eligible wards at each hospital site for sampling of staff ranged from 1 to 4 (mean 2.25, SD 1.16). The numbers of nurses recruited at each hospital ranged from 1 to 16 (mean 10.12, SD 6.77). Not all of the nurses who participated in the TEPI also completed the quantitative measures; however, these details were available for 75 of the 81 interviewees. Of those 75, the modal age group was aged 40–9 years, 43% were male, 45% were white British, 65% were married or cohabiting, and around half had been in their current post for > 3 years. Table 11 indicates the numbers from each of the four categories of hospital.

There were a number of reasons why the sample size was smaller than initially planned. At some sites there were simply insufficient numbers of staff meeting the inclusion criteria. In addition, on some wards, support from the ward manager was less enthusiastic and recruitment was poor. It is worth bearing in mind that the participation of staff was a significant burden on wards, as key staff members were taken away from work for a period of up to 90 minutes. We extended our data collection period several times and made repeated visits to the study hospitals in order to increase recruitment. Although this incrementally increased the numbers, eventually it was clear that no further recruits could be obtained.

Nurse staffing and bed numbers compared

All interview participants were asked the following: ‘On your ward, how many staff are typically on duty mid-morning on a usual mid-week shift? Numbers of qualified nurses? Numbers of health-care assistants? Numbers of anyone else? What is the number of beds on your ward?’.

The purpose of collecting these data was to ascertain whether or not there was any indication that hospitals without seclusion or easy access to a PICU managed this situation by having higher numbers of staff available.

Figures were converted into staff numbers per 20 beds, to standardise for the different numbers of beds (and therefore patients) on the study wards. Replies to these questions were provided by 56 interviewees. The mean number of staff on duty per 20 beds on a mid-week morning was 5.00 (SD 0.89). Preliminary t-tests were applied to assess whether or not there was any indication of a difference by seclusion or PICU availability. These tests were not significant for all staff on duty or for qualified nurses or health-care assistants (unqualified nurses/nursing assistants) alone. Further statistical exploration controlling for clustering by ward was, therefore, deemed unnecessary.

There was no indication that wards differed in their staffing numbers by seclusion or PICU availability; therefore, staffing number differences are unlikely to account for any differences found in the escalation pathways reported.

Frequency or recency of adverse incidents on the study wards compared

At the end of the interview, participants were asked to report on how recently their ward had experienced any of the incidents covered in the preceding interview. For each incident type, participants were asked how often this typically occurred, if ever, and specifically if it had occurred in the past week, month, 6 months or year. Participants were asked both about one-off occurrences and about repeated incidents involving the same patient.

Table 12 shows part of the TEPI used to score such behaviours.

Spearman correlations were conducted between the above data items and the presence of a seclusion room or on-site PICU, thus allowing for the ordinal nature of the data. Positive correlations indicate less frequent incidents with either seclusion or PICU provision.

Seclusion was associated with less one-off self-harm by scratching with a sharp object (r = 0.341; p < 0.001; n = 66). There were no other significant findings.

An on-site PICU was associated with less repeated, recent verbal abuse (r = 0.376; p = 0.002; n = 63), less repeated damage to objects (r = 0.354; p = 0.005; n = 61), less one-off brandishing of a blunt weapon (r = 0.300; p = 0.015; n = 65), less one-off physical assault of a fellow patient (r = 0.324; p = 0.006; n = 69), less one-off threats of suicide (r = 0.303; p = 0.011; n = 69) and less one-off serious attempted suicide (r = 0.522; p < 0.001; n = 66), as well as less repeated attempted suicide (r = 0.286; p = 0.023; n = 63) and less repeated loud shouting (r = 0.269; p = 0.027; n = 67).

These findings do not support the hypothesis that hospitals without seclusion or a PICU are more quiescent places and experience fewer risky behaviours on the part of patients. On the contrary, there is support for the interpretation that having an on-site PICU (but not access to a seclusion room) results in significantly calmer and safer acute wards.

Relative frequency of the higher-level codes

Table 13 shows the frequency of the higher-level codes in the data set, separated into first-line responses and responses to repeated behaviour. These frequencies cannot represent the frequency of use of these interventions in clinical practice, as they were cited in response to the fixed vignettes and protocols of the TEPI. However, they are also likely to be related to frequency of use in practice, and gross differences in frequency may carry some implications. In particular, the low citation of calm-down methods, use of activity rooms and garden areas and, particularly, psychological therapy might indicate interventions that could be beneficially increased in clinical nursing practice.

In first line and repeats, de-escalation and observation are the most common interventions mentioned by nurses. The repeat interventions lay a greater emphasis on talking to others, discussing medication and reviewing the case. They also place a greater emphasis on increasing medication, either oral offers or enforced injections, and are more likely to include interventions to transfer the difficult patient away from the ward, either to a PICU or to some other type of ward at a different hospital.

There is no escalation pathway

There was little evidence of any uniform escalation pathway at any one hospital. Instead, there was an extraordinary range of variation. This is illustrated in Table 14, which shows the responses of 15 qualified nurses at a single hospital in response to the following scenario as presented during the TEPI:

• BA2 behaviour 2: Alan turns over furniture, throwing objects around the ward (not deliberately at others), picking up a chair and hitting the wall with it, on top of shouting and being verbally abusive.

Notably, there is not a single intervention on which everyone agrees. Considering that staff members were being asked not what their own responses would be, but what would typically happen on their ward, this level of variation is very high.

This level of variation was the same for every scenario in the TEPI. The researcher conducting the coding was clear that the vast majority of this variation was because of actual differences in the way interviewees considered the escalation pathway to operate on their ward, rather than because of differences between interviewers or the same interviewer on different occasions, in the way the interview was conducted. This confidence was based on the consistency and high degree of structure that was a feature of the application of the TEPI.

Pathway length

The length of the escalation pathways in general was explored and is depicted in Table 15. The mean length of an escalation pathway was 5.99 (SD 3.47) interventions and the modal number of interventions was 4.

Tests were conducted for differences by seclusion and PICU availability. Pathways were significantly shorter in hospitals without access to seclusion (mean 5.61 vs. 6.47, t = 4.26; p < 0.001) and longer in hospitals without on-site access to a PICU (mean 6.35 vs. 5.47, t = 4.31; p < 0.001). Multiple regression of pathway length with both of these variables and an interaction term showed that having access to seclusion was significantly associated with a 1.57 (SE 0.26) longer pathway and access to both seclusion and an on-site PICU was associated with a 1.39 (SE 0.41) shorter pathway. PICU access by itself was not significant in the multiple regression equation.

These differences in pathway length complicate comparisons between the hospital types, as differences in frequencies of interventions might be partly a product of greater or lesser pathway length. Stronger comparisons are therefore likely to be those that compare the proportions of particular types of interventions within the differing types of hospitals (i.e. by chi-squared tests) rather than examination of the relative frequency of interventions by interviewee (i.e. by t-test).

Comparisons by proportion of use of interventions

This analysis compares, for example, the percentage of de-escalation as a proportion of all interventions cited by those in no-seclusion hospitals, with the same percentage in seclusion hospitals, using a chi-squared test. This method removes the influence of pathway length that complicates other types of analysis. Two levels of analysis were conducted, one at the level of the whole interview for interventions used when the described behaviours first occurred (i.e. without the interventions used on repetition of the problem behaviour) and the second level of analysis was by behaviour type (i.e. aggression separately, self-harm separately, mania separately). All degrees of freedom equal one in these chi-squared tests. At each level of analysis, significant variables were entered into a final logistic regression model controlling for the alternate condition (seclusion or PICU availability). Table 16 provides the frequencies of use by seclusion availability and Table 17 provides the same information by PICU availability. Statistically significant differences are detailed in the text.

All problem behaviours

Exploration of thresholds for the use of certain interventions

Seclusion

Figures 5 and 6 show the proportion of nurses who mention seclusion as one of the interventions in relation to the scenarios of aggression. It can be seen that as the scenarios get more serious, the proportion of nurses reporting that seclusion is used goes up. Nevertheless, it still remains a small proportion of the total numbers interviewed. There does not appear to be any specific behavioural threshold, that is, seclusion suddenly being used at a particular level of severity. Instead there seems to be a gradual rise in the propensity to cite the use of seclusion with increasing severity, with the differences caused by seclusion room availability becoming greater at higher levels of severity. The absence of a PICU also seems to slightly raise the frequency of reports of seclusion use. Similar charts for the self-harm questions (Figures 7 and 8) show that seclusion was hardly specified as an intervention for these questions, regardless of level of severity. S3 (deep cuts) was the only self-harm behaviour that a few nurses reported as likely to lead to seclusion on their ward and, even then, possibly only because the instrument of self-harm could be used as a weapon against staff attempting to manage the situation.

FIGURE 5.

Proportion of nurses citing seclusion as an intervention, by level of aggression and seclusion availability. A1, verbal abuse; A2, property damage; A3, wielding a club; A4, assaulting a fellow patient; and A5, assaulting a fellow patient very seriously.

FIGURE 6.

Proportion of nurses citing seclusion as an intervention, by level of aggression and PICU availability. A1, verbal abuse; A2, property damage; A3, wielding a club; A4, assaulting a fellow patient; and A5, assaulting a fellow patient very seriously.

FIGURE 7.

Proportion of nurses citing seclusion as an intervention, by level of self-harm and seclusion availability. S1, expressing suicidal ideation; S2, scratching self; S3, cutting self; and S4, attempt to hang self.

FIGURE 8.

Proportion of nurses citing seclusion as an intervention, by level of self-harm and PICU availability. S1, expressing suicidal ideation; S2, scratching self; S3, cutting self; and S4, attempt to hang self.

Psychiatric intensive care unit transfer

Figures 9 and 10 show the proportion of nurses who reported a PICU transfer in the escalation pathway for aggressive behaviour, broken down by the specific severity of aggressive behaviour (TEPI questions A1–A5) and by seclusion or PICU availability. It can be seen from Figure 10 that PICU transfer is more likely to be reported at hospitals with a PICU on site and that the difference increases with severity. However, the general level of difference is not as large as for the reporting of seclusion described in the preceding section. There is a notable dip in rates for question A4 (assault with punches and kicks) relative to A3 (wielding an object as weapon) suggesting that in terms of the interventions seen as necessary, A3 is perceived to be a more serious incident than A4.

FIGURE 9.

Proportion of nurses citing PICU transfer as an intervention, by level of aggression and seclusion availability. A1, verbal abuse; A2, property damage; A3, wielding a club; and A4, assaulting a fellow patient; A5, assaulting a fellow patient very seriously.

FIGURE 10.

Proportion of nurses citing PICU transfer as an intervention, by level of aggression and PICU availability. A1, verbal abuse; A2, property damage; A3, wielding a club; A4, assaulting a fellow patient; and A5, assaulting a fellow patient very seriously.

By comparison, Figures 11 and 12 make clear that PICU transfer is seldom seen as a relevant intervention for suicidal behaviour, even when that behaviour is severe. However, even at these low levels of reporting, a difference can be seen by whether or not a PICU is available on site. It is possible that the restriction of the TEPI to a consideration of male patients only may have influenced these particular findings.

FIGURE 11.

Proportion of nurses citing PICU transfer as an intervention, by level of self-harm and seclusion availability. S1, expressing suicidal ideation; S2, scratching self; S3, cutting self; and S4, attempt to hang self.

FIGURE 12.

Proportion of nurses citing PICU transfer as an intervention, by level of self-harm and PICU availability. S1, expressing suicidal ideation; S2, scratching self; S3, cutting self; and S4, attempt to hang self.

Manual restraint

Figures 13 and 14 show the proportions of nurses who reported manual restraint in the escalation pathway for aggressive behaviour, broken down by the specific severity of aggressive behaviour (TEPI questions A1–A5) and by seclusion or PICU availability. Figures 15 and 16 present the parallel information for self-harm. Use of manual restraint for aggression does not appear to differ by seclusion or PICU availability, although there is a notable dip in rates for question A3 (wielding a weapon) overall. The picture for self-harm is, again, identical regardless of seclusion and PICU availability, except for one question. For S4, which describes a very serious but prevented suicide attempt by hanging, manual restraint is reported more frequently when seclusion is not available.

FIGURE 13.

Proportion of nurses citing manual restraint as an intervention, by level of aggression and seclusion availability. A1, verbal abuse; A2, property damage; A3, wielding a club; A4, assaulting a fellow patient; and A5, assaulting a fellow patient very seriously.

FIGURE 14.

Proportion of nurses citing manual restraint as an intervention, by level of aggression and PICU availability. A1, verbal abuse; A2, property damage; A3, wielding a club; A4, assaulting a fellow patient; and A5, assaulting a fellow patient very seriously.

FIGURE 15.

Proportion of nurses citing manual restraint as an intervention, by level of self-harm and seclusion availability. S1, expressing suicidal ideation; S2, scratching self; S3, cutting self; and S4, attempt to hang self.

FIGURE 16.

Proportion of nurses citing manual restraint as an intervention, by level of self-harm and PICU availability. S1, expressing suicidal ideation; S2, scratching self; S3, cutting self; and S4, attempt to hang self.

Observation

Figures 17 and 18 show the proportion of nurses who reported observation in the escalation pathway for aggressive behaviour, broken down by the specific severity of aggressive behaviour (TEPI questions A1–A5) and by seclusion or PICU availability. Figures 19 and 20 present the parallel information for self-harm. In general, the relevance of observation dwindles as aggression severity increases. However, this decrease is sharper in hospitals with seclusion availability and sharper still when an on-site PICU is available. For self-harm, observation is reported at similar frequencies for all severities of behaviour; the only difference emerging is for S3 (deep cuts), when observation dips in frequency when seclusion is available.

FIGURE 17.

Proportion of nurses citing observation as an intervention, by level of aggression and seclusion availability. A1, verbal abuse; A2, property damage; A3, wielding a club; A4, assaulting a fellow patient; and A5, assaulting a fellow patient very seriously.

FIGURE 18.

Proportion of nurses citing observation as an intervention, by level of aggression and PICU availability. A1, verbal abuse; A2, property damage; A3, wielding a club; A4, assaulting a fellow patient; and A5, assaulting a fellow patient very seriously.

FIGURE 19.

Proportion of nurses citing observation as an intervention, by level of self-harm and seclusion availability. S1, expressing suicidal ideation; S2, scratching self; S3, cutting self; and S4, attempt to hang self.

FIGURE 20.

Proportion of nurses citing observation as an intervention, by level of self-harm and PICU availability. S1, expressing suicidal ideation; S2, scratching self; S3, cutting self; and S4, attempt to hang self.

Coerced intramuscular medication

Figures 21 and 22 show the proportion of nurses who reported coerced intramuscular medication in the escalation pathway for aggressive behaviour, broken down by the specific severity of aggressive behaviour (TEPI questions A1–A5) and by seclusion or PICU availability. Figures 23 and 24 present the parallel information for self-harm. Overall, there is a slight decline in the perceived relevance of coerced intramuscular medication as aggression severity increases. Its use is reported more frequently for aggression across the spectrum of severity when there is no seclusion available. Coerced intramuscular medication is hardly seen as appropriate in the management of self-harm and was seldom reported for any level of severity of the behaviour.

FIGURE 21.

Proportion of nurses citing coerced intramuscular medication as an intervention, by level of aggression and seclusion availability. A1, verbal abuse; A2, property damage; A3, wielding a club; A4, assaulting a fellow patient; and A5, assaulting a fellow patient very seriously.

FIGURE 22.

Proportion of nurses citing coerced intramuscular medication as an intervention, by level of aggression and PICU availability. A1, verbal abuse; A2, property damage; A3, wielding a club; A4, assaulting a fellow patient; and A5, assaulting a fellow patient very seriously.

FIGURE 23.

Proportion of nurses citing coerced intramuscular medication as an intervention, by level of self-harm and seclusion availability. S1, expressing suicidal ideation; S2, scratching self; S3, cutting self; and S4, attempt to hang self.

FIGURE 24.

Proportion of nurses citing coerced intramuscular medication as an intervention, by level of self-harm and PICU availability. S1, expressing suicidal ideation; S2, scratching self; S3, cutting self; and S4, attempt to hang self.

Exploration of lower-level intervention codes

Examination and exploration of the more fine-grained codes was undertaken to further illuminate those differences already discovered. Not all higher-level codes had such subcodes, but those with many were observation, de-escalation and manual restraint.

Breakdown of different types of observation

Figure 25 provides a bar chart of the frequencies of the different types of observation identified within the interviewee responses and Table 18 provides the codes and their definitions.

FIGURE 25.

Relative frequencies of observation.

TABLE 18 - Codes and definitions of observation

Note

5-minute observation was enquired into, but was not reported.

Code	Definition
OBS	Unclassified
OBS-VIG	Vigilance, all staff on duty asked to be aware and keep their eyes open for the patient concerned, without them being placed on formal observation, including monitoring the outcome of treatment changes, give space, do nothing
OBS-I	Intermittent observation at unspecified intervals
OBS-I-10	Intermittent observation at 10-minute intervals
OBS-I-15	Intermittent observation at 15-minute intervals
OBS-I-30	Intermittent observation at 30-minute intervals
OBS-C-E	Constant observation within eyesight
OBS-C-A	Constant observation within arm’s length
OBS-C-A2	Constant observation within arm’s length, two staff, 2 : 1
OBS-C	Constant observation, unspecified distance
OBS-C-2	Constant observation by two staff, no distance specified
OBS-C-3	Three staff on constant observation, no distance specified
OBS-LSE	Room observation – patients voluntarily stay in unlocked room
OBS-P-C	Policy of hospital to put on constant observation
OBS-P-I	Policy of hospital to be on intermittent observation (e.g. 15-minute intervals)
OBS-RLC	Relocate patient to a different bedroom to allow better observation of patient
OBS-STL	Staff assigned to a location to prevent conflict such as absconsion (e.g. ward door or patient door after coerced intramuscular injection)

Intermittent observation clearly has a normative practice of around 15 minutes, this constituting the first data on observation intervals used during this practice. There is no clear uniformity or agreement around constant observation practice, with many different varieties being reported. Observation via team vigilance has not been previously described and does not form a part of hospitals’ observation policies. Its efficacy and impact are, therefore, unknown.

In the analyses that follow, using the same proportion-of-reports approach as previously used, it should be noted that some types of observation were mentioned in very low numbers. Such small differences in rates can be statistically significant without necessarily being of great clinical relevance. When the numbers are small, this has been indicated with the results.

Seclusion

When there is no access to seclusion, there is more generic constant observation (χ² = 5.06; p = 0.024), constant observation by two staff (χ² = 6.51; p = 0.011), constant observation by three staff (χ² = 4.55; p = 0.033), constant observation in a low-stimulus environment (χ² = 26.69; p < 0.001) and staff located to facilitate observation (χ² = 13.13; p < 0.001). There is less use of 10-minute intermittent observation (small numbers, χ² = 5.25; p = 0.022), constant observation within eyesight (χ² = 5.81; p = 0.016), constant observation within arm’s length, two staff (small numbers, χ² = 5.25; p = 0.022), intermittent observation by policy (χ² = 65.24; p < 0.001) and relocating bedroom to facilitate observation (small numbers, χ² = 9.21; p = 0.002). Using multivariate logistic regression, controlling for PICU availability, the only items still significant were constant observation in a low-stimulus environment, staff located to facilitate observation, constant observation within eyesight and intermittent observation by policy. A number of smaller items were dropped from the analysis because they predicted seclusion availability perfectly and could not, therefore, be included using this statistical method. The absence of seclusion therefore means that there is a greater emphasis on constant observation in a low-stimulus environment and staff being located to facilitate observation, and a lesser emphasis on constant observation within eyesight and intermittent observation by policy.

Psychiatric intensive care unit

When there is no access to a PICU on site, there is more constant observation by two staff (χ² = 4.28; p = 0.039), constant observation by three staff (χ² = 5.25; p = 0.022), constant observation within eyesight (χ² = 46.37; p < 0.001), observation in a low-stimulus environment (χ² = 6.15; p = 0.013) and constant observation by policy (χ² = 6.31; p = 0.012). There is less use of generic observation (χ² = 11.47; p = 0.001), 15-minute intermittent observation (χ² = 4.25; p = 0.039) and generic constant observation (χ² = 42.51; p < 0.001). Use of multivariate logistic regression controlling for seclusion availability only leaves constant observation within eyesight, generic observation and constant observation as significant. The absence of an on-site PICU therefore means that there is a greater emphasis on constant observation within eyesight and less use of generic forms of observation.

Summation

The absence of seclusion leads to a change of emphasis in observation use towards the constant observation of a patient in a separate room by themselves, accompanied by one or more staff members or with a staff member stationed at the door of the room. The absence of an on-site PICU leads to greater use of constant observation within eyesight.

Breakdown of different types of de-escalation

Figure 26 provides a bar chart of the frequencies of the different types of de-escalatory actions identified within the interviewee responses and Table 19 provides the codes and their definitions.

FIGURE 26.

Relative frequencies of de-escalation actions.

TABLE 19 - Codes and definitions of de-escalation

Code	Definition
DESC	Unclassifiable de-escalation
DESC-R	Ask the reason for the behaviour, explore why the patient is doing what they are doing, what they are experiencing
DESC-CH	Consequences to patient’s health, problems with non-compliance
DESC-O	Orientate, give information about the ward, what is going on, who people are and why they are here
DESC-F	Give feedback about the impact of the behaviour on others, patients and staff to reflect on behaviour
DESC-D	Distract, talk about completely separate and different issues, make conversation
DESC-AS	Ask to stop and give/drop object (if appropriate) or similar
DESC-TS	Tell to stop, instruct, command, say behaviour is not acceptable
DESC-C	Inform of consequences to themselves (e.g. vulnerability to others’ actions)
DESC-SOL	Solutions, find out what has helped before, what can be done, what options are available, how can staff help
DESC-MEDIC	Get the doctor to talk to them
DESC-B	Set boundaries, explain boundaries, expectations
DESC-M	Encourage to take medication, providing information about it and the patient’s care plan
DESC-DB	Debrief patient about what has just happened
DESC-P	Talk about a plan that has been made and what is going to happen
DESC-FAM	Family speak to patient
DESC-REF	Ask other people to speak to patient (e.g. another health professional or expert in a non-health field)
DESC-S	Give reassurance, support, offer to help the patient and there to talk if they want

Discounting those responses that were unclassifiable, the top six most frequently cited forms of de-escalation were (1) asking the reason for the behaviour (17%); (2) asking to stop (14%); (3) giving support and reassurance (14%); (4) telling to stop (11%); (5) encouraging to take medication (6%); and (6) talking about what is going to happen (6%). All other de-escalatory actions were mentioned at much lower frequencies (≤ 4%).

Differences in de-escalation by seclusion and PICU availability were tested using the difference-in-proportion-of-reports approach, followed by multivariate logistic regression including those items found to be significant, as before.

Seclusion

When there is no access to seclusion, de-escalation focuses more frequently on giving reassurance and support (χ² = 12.44; p < 0.001), finding solutions (χ² = 7.63; p = 0.006), telling to stop (χ² = 25.7; p < 0.001) and asking to stop (χ² = 13.48; p < 0.001), and focuses less frequently on giving feedback on the impact of the patient’s behaviour on others (χ² = 12.11; p = 0.001), informing of the consequences for themselves (χ² = 5.15; p = 0.023), orientating and giving information (χ² = 5.15; p = 0.023), talking about or making plans (χ² = 7.51; p = 0.006), discussing medication (χ² = 5.58; p = 0.018) or unclassifiable ‘de-escalation’ (χ² = 19.24; p < 0.001). Using multivariate logistic regression, controlling for PICU access, all of these variables remain significant except for discussing medication, making plans and informing of consequences. Orientation was excluded in this analysis due to collinearity (i.e. it was a perfect predictor and could not therefore be included). The absence of seclusion therefore means that there is a greater emphasis on reassurance, solutions, asking and telling to stop, and less emphasis on generic de-escalation or elaboration of the patient’s impact on others.

Psychiatric intensive care unit

When there is no on-site PICU, de-escalation focuses more frequently on unclassifiable ‘de-escalation’ (χ² = 6.53; p = 0.011), debriefing patients about what has just happened (χ² = 7.27; p = 0.007), distraction (χ² = 9.88; p = 0.002), informing about medication or encouraging to take it (χ² = 10.60; p = 0.001) and giving feedback on the impact of the patient’s behaviour on others (χ² = 16.54, p < 0.001). De-escalation focuses less frequently on telling patients to stop (χ² = 9.61; p = 0.002), asking patients to stop (χ² = 6.13; p = 0.013) and asking for/exploring the reason for the patient’s behaviour (χ² = 16.66; p < 0.001). Using multivariate logistic regression, controlling for seclusion access, all of these items remain significant. The absence of an on-site PICU, therefore, means that there is a greater emphasis on generic de-escalation, debriefing, distraction and encouraging medication, and less emphasis on asking or telling the patient to stop or exploring the reasons for their behaviour.

Summation

These findings resist a neat summary. It could be argued that the absence of seclusion leads to greater use of softer forms of de-escalation. Associations with PICU care availability are more opaque.

Breakdown of different types of manual restraint

A large number of codes were created to deal with information on manual restraint. These codes detailed the numbers of staff involved, the position of the patient and the location of the patient. However, the vast majority of the data collected during the interview were generic, providing no specific information about these aspects of the manual restraint process. It was particularly disappointing that a question about long-duration restraint was missed out so frequently that no useful information was collected. The questions in the TEPI were perhaps too detailed for the scenario being discussed, requiring a great deal of supposition and guesswork on the part of the interviewee, or possibly requests for clarifications from the interviewer that could not be provided. Alternatively, the issue of prone versus supine restraint was a matter of controversy at the time the data were collected, thus being an even more sensitive issue than usual. These may offer reasons as to why both parties seemed to avoid detailed discussion.

Some interviewers probed the participant for details, whereas others just left it at a generic description: ‘we would restrain him’. In some cases, the participant was asked about positions only (on the floor, sitting down, etc.) and not numbers. In others, they were just asked about the number of people involved in the restraint and not the position. So, in short, often ‘restraint’ was not investigated enough to include the correct level of detail. In many cases, as soon as the first restraint was described (usually in the case of Mike refusing his medication – behaviour 1, at some level of repeat or within the ‘if history’ prompt) this exact description was referred to from then on. For example, the participant would describe restraint for the first time. If restraint was then mentioned a second time at any point after, the interviewer would say ‘the same process as described first?’ and the participant would confirm ‘yes’. This would continue throughout the interview. This might have eliminated any variation in restraint between types of behaviour (e.g. mania vs. aggression). It also meant that if the first described form of restraint lacked in detail, this would follow through for every other behaviour thereafter. Owing to the difficulty in collecting these data comprehensively and consistently, the findings below should be taken with great caution.

It was necessary to collapse the many detailed lower-level codes to gain any useful information, and the results of this exercise can be found in Figure 27. This makes the high number of unspecified reports relative to others clearly visible. Setting these to one side, the most frequent patient position for restraint was lying, presumably on the floor as manual restraint on the bed was counted separately. There was insufficient information to say whether this was prone or supine. The next most frequently reported type of restraint was on the bed. Standing, sitting or kneeling positions were very infrequently reported.

FIGURE 27.

Frequencies of different types of manual restraint.

Seclusion

When there was no seclusion, manual restraint was less likely to be sitting (χ² = 12.92; p < 0.001). This result remained significant when PICU availability was taken into account using multivariate logistic regression, although the numbers were very small.

Psychiatric intensive care unit

When there was no on-site PICU, restraint was more likely to be standing (χ² = 11.99; p = 0.001) or on the bed (χ² = 10.86; p = 0.001), and less likely to be sitting (χ² = 20.98; p < 0.001) or lying (χ² = 4.95; p = 0.026). Using multivariate logistic regression controlling for seclusion availability, standing and on the bed remained significant, whereas sitting was dropped from the analysis as it predicted PICU availability perfectly. The absence of an on-site PICU meant that restraint was more likely to be standing or on the bed.

Summation

Detailed data on manual restraint were sparse and to some degree unreliable. The most frequent position was lying, probably on the floor. However, restraint on a bed was nearly as frequent. Differences by seclusion or PICU availability were not susceptible to easy interpretation.

Repeated behaviours

The TEPI was constructed so that, for each behavioural scenario, the interviewer was to collect information about what would happen if the behaviour was repeated during the shift, in the next shift or for the whole day. Interviewers found it difficult to collect this information consistently and in a logical way. It was not always easy to say the behaviour was repeated if the previous account had ended in seclusion, PICU transfer or a period of close observation. The different time intervals were also impossible to separate. Furthermore, in the flow of this lengthy and difficult interview, interviewers sometimes omitted questions about repeats completely for some or many questions. Nevertheless, there was a substantial quantity of repeat information in the data set and this was therefore, with some caution, submitted to an analysis, using the proportion-of-reports approach. These frequency data are presented in Tables 20 and 21, with the statistically significant results in the text that follows.

Summary of findings

Hospitals and their staff do not have any common practice threshold and escalation pathways. Hospitals without a seclusion room available managed disturbed behaviour through:

• less use of seclusion, although small numbers of seclusions still occur outside policy and utilising rooms not built for the purpose

• greater use of coerced intramuscular medication

• less use of calm-down methods and show of force

• observation use by the constant observation of the patient in a separate room by themselves, accompanied by one or more staff members, or with a staff member stationed at the door of the room

• indicative evidence of more use of PRN medication and calls to the police and fewer reviews of care.

Hospitals without an on-site PICU managed disturbed behaviour through:

• less use of PICU transfer; transfers to PICUs at other sites did occur, but they were less frequent

• greater use of seclusion

• greater use of de-escalation

• greater use of observation, specifically observation within eyesight

• greater use of show of force

• greater use of calm-down methods

• indicative evidence of less use of psychological interventions and reviews of care.

These differences by seclusion and PICU availability are greatest for aggression, less prominent for self-harm and unclear for mania. The differences are also greater for the more severe expressions of these behaviours, although seclusion was not used for self-harm and PICU transfer was seldom reported. Coerced intramuscular medication was also not seen as relevant in the management of self-harm.

The normative interval for intermittent observation is every 15 minutes and a practice of team vigilance outside policy was described. The second most frequent manual restraint position was on a bed.

No escalation pathway

Although the absence of a common escalation pathway has been previously reported by a single hospital qualitative study in the USA,64 not to find one was surprising and also unwelcome, as this element of the overall research was predicated on finding and describing such pathways. It is possible that the interview process was flawed in some way and failed to detect pathways that actually existed, perhaps because interviewees failed to recollect or accurately describe common practice. Certainly, orthodox social researchers have succeeded in generating similar ‘ethnographic decision trees’ through quasi-naturalistic methods. 65 Alternatively, it could be that, as the TEPI development process revealed, managing crisis situations was dependent on so many moment-to-moment changes of behaviour and response by the individual patient, balanced with their known history and existing relationship with the nurse, that in fact a common pathway practised by a group of staff in one location is not possible. A third possibility is that such responses are first and foremost pragmatic moment-to-moment social judgements through which staff members actually create a meaningful structure to their actions and the way that they are sequenced. In this case, if escalation pathways do exist in any documentary or attestable form, they may not in any simple sense determine what staff members do; rather, they are oriented to by staff members as they either ‘do’ following an escalation pathway or attempt to perceive such structure in actions that they observe. Ethnomethodology66 and post-Wittgensteinian social theory67 may be relevant approaches to understanding this.

Nevertheless, despite methodological difficulties, there were differences in our sample dependent on the availability of seclusion and PICU care at hospitals, and we were able to describe these based on statistical testing. Lessons about practice can be drawn even without a common escalation pathway, for there were many intervention strategies, particularly types of de-escalation, that were mentioned only rarely by few staff. It could be that these methods were less mentioned because they were less useful, or it could be that nursing staff could gain through the expansion of their range of strategies for intervention, giving them a greater number of interventions to choose from to match the patient and situation.

Seclusion still used in the absence of a dedicated room and policy

Clearly this is not desirable. The absence of a specially constructed robust room means that a disturbed patient could easily break fixtures and fittings, harming themselves or staff members who go in to manage the situation. The use of seclusion without a policy means a deficit of recording, reporting and an absence of clinical oversight, mandated clinical reviews at intervals and managerial supervision of practice and rates of use. Although this practice was seldom mentioned in our data, it is not really possible to extrapolate from that how frequently it is occurring. Interviewees may have had some embarrassment or sensitivity about it and been disinclined to mention it for these reasons. This should, therefore, be an issue of concern at any acute psychiatric hospital without a dedicated seclusion room.

The existence of this practice, not previously documented, does explain data previously collected during the City-128 study, showing a rate of reported seclusion at hospitals without a seclusion room. 2 This was previously attributed to reporting error; however, now it is clear that the data were, in fact, likely to be close to accurate and this suggests that the rate of seclusion in English acute psychiatric wards that do not have access to a seclusion room is once every 20 days on a 20-bedded ward. This figure may seem high, but it is based on 6 months’ worth of end-of-shift reports (> 21,000 reports) submitted by 67 randomly selected acute wards where the ward managers reported no access to a seclusion room and using the same definition of seclusion used in this study.

Nursing in a side room as an alternative to seclusion

Hospitals without seclusion rooms also managed disturbed patients by two practices commonly referred to as ‘nursing the patient in a side room’: one in which the nurses sat outside the door of the room (often, but not always, the patient’s bedroom) and the other in which nurses (sometimes two or three) accompanied the patient into the room.

The first of these practices, stationing a nurse near the door either to disincline or to prevent the patient from leaving, or to observe them through the open door, is seen by some as being, in effect, seclusion. It would come under the definition of seclusion used in some policy documents, such as the UK MHA code of practice (2015, paragraph 26.103). 40 Whether or not it is useful to do so remains open to question as, in current practice, it can mean that the patient is restricted to their own bedroom, with access to all their own clothing and other items, and their own en-suite bathroom (if provided), compared with isolation from staff and patients in a separate robust seclusion room with a heavy locked door, observation panel, robust unbreakable furnishings, etc. There is a great deal of difference between these two practices and counting them together for the purpose of official statistics would equate both and fail to recognise the less severe and more acceptable end of practice. Exactly how seclusion is defined will also influence international comparison statistics and assessing where we are in the UK vis-à-vis other countries’ reductions in restrictive practices. At present there are no national statistics in the UK on seclusion rates, unlike many other countries where seclusion requires a formal report to the central government. 68 Clarity of definition and terminology in this and related areas is highly desirable. Good terminology and definitions have the capacity to exert beneficial influence. Poorly constructed or vague resolutions of the difficulties of describing different interventions may warp practice or the reporting of practice in unpredictable ways, at worst leading to a less desirable range of options for patients.

The second of these interventions, in which staff accompany the patient into the room, has been less well described in UK psychiatric nursing practice. The room could be the patient’s bedroom, but could also be a specialist de-escalation room of some sort, or those hospitals that do have a seclusion room can use it in this way. Some UK hospitals do have special extra care or intensive care areas where staff can accompany patients, locking the door behind them so that all are isolated together. This practice is referred to in Norway as ‘open-area seclusion’,69 and is the only form of seclusion used there. In Norway, the time with the patient is well used in a therapeutic way, to debrief the patient, work with their understanding of what has happened and what has to change in order for them to rejoin the rest of the patients. This may be an opportunity for nurses to forge and develop therapeutic interventions with patients. However, what happens during this time in the UK is unknown and the association of observation with untrained staff70,71 suggests that the practice here is likely to be more restrictive than therapeutic. This practice is also clearly more risky for the staff. Instead of the angry, agitated, deluded and hallucinating patient being on the other side of a hefty seclusion room door, nurses accompany the patient, potentially in sufficient numbers to instantly manually restrain them if their violence recommences. Any misjudgement about the timing of interventions could, in these circumstances, lead to injured nurses. In addition, the presence of the nurses implies a greater use of manual restraint, which carries risks both for the patient and the staff. 72 Finally, the use of staff members, if more than one, is bound to be costly relative to seclusion.

In previous research by our group, stationing a nurse at the doorway and a third option that did not differ by seclusion availability in this study (asking the patient to stay voluntarily in a side room for a period of time) were collectively referred to as ‘time out’. Our previous research had no special category for staff accompanying the patient in the side room, so this may have been categorised as constant observation or time out or both in that research and related reports. Bearing this in mind, our previous work has sought to examine the acceptability of different containment practices and their efficacy. Time out is seen as one of the most acceptable forms of containment by patients and staff,4 student nurses51 and the general public,54 all of these studies also demonstrating that it was rated very differently to seclusion, which was seen as one of the least acceptable containment methods currently in use in the UK. Several studies found, however, that time out was associated with less severe disturbed behaviour,2,3 although the use of time out following a violent assault appeared to be equally effective in preventing a repetition. 3

More evaluation studies are clearly required in this area and it has been demonstrated that it is possible to conduct randomised trials comparing different forms of containment. 73 More evaluation information has the potential to quickly shape practice into a safer and more effective form.

Coerced intramuscular medication as an alternative to seclusion

Nurses from hospitals without access to seclusion were more likely to report the use of coerced intramuscular medication (rapid tranquillisation) in their escalation pathways. The two are not mutually incompatible interventions and it is known that a small number of patients are given intramuscular medication as well as being secluded (5%), and more accept oral PRN medication during the seclusion process (24%). 74 The use of seclusion is also statistically associated with the use of coerced intramuscular medication. 2 The relative efficacy of coerced intramuscular medication and seclusion is not known, by any outcome indicator. In UK acute psychiatry, the vast majority of seclusions are terminated within a few hours. 3 However, it is a negative experience for patients, who feel a combination of anger, upset, abandonment, fear, humiliation and worthlessness, with the experience often being interpreted as punishment. 1 A similar range of negative emotions are reported by those subject to coerced intramuscular medication: angry, helpless, frightened,75 sadness, violation, panic and fear of becoming ill or dying as a result of the injection. 76 Any parenteral medication of sedating drugs carries some risk and the contribution of rapid tranquillisation to the (thankfully) rare deaths during manual restraint is not known. 72

Large-scale surveys of UK acute patients have shown slightly higher ratings of acceptability for seclusion than coerced intramuscular medication; however, the difference is not large and masks a wide variety of different opinions between patients. 4 Those patients who had experienced coerced intramuscular medication were less likely to consider it acceptable, whereas the experience of seclusion made no difference to acceptability ratings. A study with a smaller sample conducted in the Netherlands showed equal numbers with preference for seclusion or medication, with one-quarter of patients having no preference either way. 77 Acute care nurses in the UK, on the other hand, give significantly higher acceptability ratings to coerced intramuscular medication than seclusion,4 whereas nursing and non-nursing students rated their acceptability as equal. 54 Using some of the same data from the City-128 study, Dack et al. 45 examined the relationship between acceptability ratings and actual rates of usage on the wards. Greater use of seclusion was associated with higher acceptability ratings of seclusion by all patients (i.e. including, but not exclusively, those who had experienced it); however, greater use of coerced intramuscular medication was associated with lower acceptability ratings for coerced intramuscular medication itself and nearly every other form of containment, including seclusion. The authors suggested that this might be interpreted as the use of coerced medication having a ‘toxic’ effect on patient attitudes.

These different survey results on relative acceptability come out slightly in favour of seclusion rather than coerced intramuscular medication, albeit with significant variation in individual patient preferences. It is relatively easy to recommend that patient preferences be requested and acted on; however, in busy acute wards, with constantly changing staff and patients and unpredictable events in unpredictable circumstances, where, as we have seen, there are not even any consistent escalation pathways, getting such a system into practice would represent a significant challenge. When there is no seclusion room available, the choice cannot be offered.

Seclusion as an alternative to psychiatric intensive care unit care

The data from this study support the idea that when seclusion is available and an on-site PICU is not, seclusion is utilised as an alternative for the management of disturbed behaviour. This clarifies a relationship that is difficult to investigate because of the way seclusion is provided and care managed in some places. For example, in a significant number of places the available seclusion room is located in the PICU. In some of those hospitals, the patient must be transferred to the PICU in order to use the seclusion room, whereas in others, staff from the acute ward bring the patient into seclusion and stay outside the room to monitor and provide care, without any such transfer taking place. Cross-sectional studies looking at statistical associations are thus confounded by service provision and its associated policies.

With seclusion, once the person has calmed, they return to mix with the other patients on the acute ward where the disturbance took place. Such episodic seclusion seems likely to be a cheaper option than PICU care; however, it means that the risk of further conflict events remains on the acute ward, where staffing levels are lower and the environment is less robust. Moreover, the remaining patients on the acute ward must then accommodate their behaviour towards that potentially risky individual and may be more fearful as a consequence. 78

De-escalation as an alternative to psychiatric intensive care unit care

Greater use of de-escalation was a notable component of care pathways when no on-site PICU was available. Any judgement of the value of this depends on the efficacy of de-escalation, about which little is currently known. The components of de-escalation have only recently been brought together and described,79 and have been coded into a model80 that forms part of the Safewards interventions known to be effective in reducing rates of conflict and containment. 39 This account of de-escalation is further expanded by the de-escalatory interventions identified in this study. Around half of all incidents to which a rapid response team were called in one hospital were resolved without the use of restraint, although this may have been as much a product of show of force as de-escalation. 81 An analysis of the CONSEQ data,3 obtained during the first 2 weeks of a hospital admission for 522 randomly sampled patients, showed that 60% of uses of de-escalation were successful, although de-escalation was less likely to be successful with patients who had a history of violence. 82 Research into de-escalation is at an early stage; current evidence suggests that it can be effective and therefore could to some degree act as a substitute for PICU care.

Observation within eyesight as an alternative to psychiatric intensive care unit care

There is a great paucity of information or evidence on the efficacy of this type of observation. The analysis of thresholds shows that this use of within-eyesight observation as a substitute for PICU care is specific to aggressive behaviour. Placing a member of staff within eyesight does mean that the patient concerned can be promptly distracted from activities likely to trigger their irritation or aggression, that other patients can be warned away or diverted from putting themselves at risk, and that de-escalation and other interventions such as PRN medication are offered early and at times when they are most likely to be effective. When aggression does occur, it means that someone is present to call for assistance immediately and then intervene so as to prevent harm. However, within-eyesight observation might also place the observer at some risk, as it can be highly irritating to the patient to be followed around by another person and therefore actually evoke the behaviour that the team wishes to prevent. In some hospitals (known to the authors but not in the TEPI sample) this risk means that all such observation is done on a two-to-one basis (two nurses to one patient being observed) and therefore is an expensive option, as well as potentially being even more irritating to the patient who is subject to it. The only evidence that has a bearing on the efficacy of observation for aggression is a natural experiment in which observation was successfully substituted for seclusion in one PICU. 83 This does not really take us much further, and more research is urgently required on this topic.

Calm-down methods, show of force and seclusion/psychiatric intensive care unit availability

It was unclear why show of force and calm-down methods showed contrasting associations with PICU and seclusion availability. Their positive association with the provision of seclusion might indicate that they form part of a regular process of getting into seclusion, in other wards that show of force is used to persuade the patient to voluntarily enter the seclusion room and calm-down methods are offered to the patient once they are in seclusion. However, in relation to PICU availability, calm-down methods and show of force are associated with its absence, whereas it can be imagined that show of force might be needed to compel transfer. It is possible that as PICU transfers take time, the crisis is over before the transfer takes place and such interventions are not required. Nevertheless, the association of show of force with the absence of an on-site PICU remains without an explanation. Findings from the City-128 study offer no help here, as no relationship was found between seclusion/PICU availability and rates of show of force in that study. 84

Reviews of care and seclusion/psychiatric intensive care unit availability

Fewer reviews of care were reported when seclusion or a PICU were not available. This is probably because hospital policies for seclusion and PICU transfer generally specify that formal reviews should take place; in the case of seclusion, these may be repeated at fixed intervals once seclusion has started. Those places without access to one or the other might benefit from instituting formal reviews of care as part of the management of disturbed behaviour, unless those reviews have poor utility and fulfil only a bureaucratic function. Such reviews do have the potential for the joint exploration of psychological and social causes of the patient’s difficult behaviour and thus have the potential to form part of a therapeutic process. This is yet another topic in the area of disturbed behaviour management that requires further evaluative research.

Normative 15-minute intervals for intermittent observation

This study provides the first empirical data on the typically implemented intervals for intermittent observation in the UK. A previously unpublished review of the policies of 31 mental health trusts by D Stewart and L Bowers (King’s College London, 2012) reported that:

The minimum time period for checking patients was either 5, 10 or 15 minutes. Eight Trusts did not specify a minimum time, while a further two advised that checks should be entirely random. The maximum time interval included 10, 15, 30 and 60 minutes, although 30 minutes was the most frequent recommendation (n = 16). Seven policies did not specify a maximum interval for checking patients. The highest number of intervals specified in a single policy document was six, while two policies specified only one permissible interval (15 minutes in both cases). However, most Trusts (n = 21) did not provide guidance for the total number of possible time intervals for intermittent observation. Seven policies advised that checks on patients should vary within the specified interval. For example, where checks were prescribed to take place every 15 minutes, this should be conducted at different times within this period.

The issue of intervals is important because intermittent observation is one of several important mechanisms on acute psychiatric wards for the prevention of suicide, in that it has a demonstrable degree of efficacy, as it has been shown to be critical in stopping serious suicide attempts in progress. 85 Intermittent observation has been criticised, as some patients still successfully take their own lives in the intervals between checks;86 however, to deduce that it should therefore not be used is a logical fallacy. Some patients who are prescribed antidepressants still take their own lives, but this is not an argument to cease such prescriptions. Intermittent observation is one of a wide range of suicide prevention precautions taken by wards and their staff, and has a useful place when matched to the assessed risks for patients. However, clearly the interval length is going to have implications for its potential efficacy and it might be considered a relatively easy improvement to decrease the normative time interval to 10 minutes when intermittent observation is used.

Team vigilance as an observation method

Team vigilance may be an interesting addition to practice, involving the whole team on an informal basis and not carrying any bureaucratic burden of monitoring and reporting. To be more widely used it might be of benefit to consider in what circumstances it could be used and to manage what level and what types of risk. These questions being answered, it still needs to be made clear how the team of nurses are to put ‘vigilance’ into effect; in other words, what does it actually mean? Without such specification, alerting all of the staff members to be aware and vigilant regarding a specific patient may turn out to mean very little or even nothing, or might mean very different things to different members of staff. On the other side of such an equation, formalisation and regulation may actually make staff less likely to opt for using this as one tool to keep their ward safer for patients.

Manual restraint on the bed

The high number of manual restraints on the bed may be an issue of concern. Such restraints may arise when the patient is taken to a side room, for the sake of privacy, for the administration of medication by injection when some degree of force is being used. Alternatively, it may arise when the patient and the restrainers are seated on the edge of the bed and the patient’s arms are held. Perhaps most worrying would be the prone full-force restraint of a patient on the bed, as bedding, pillows and the soft nature of the mattress might all make it more difficult for the patient to breathe. Currently, manual restraint training in the UK is provided by a number of different organisations, none of which makes publicly available the range of techniques it teaches. It is, therefore, impossible to say whether or not restraint on a bed is covered during the training of staff. This is an issue of some concern. Further research on the actual practice of manual restraint (not just the teaching of it) is clearly required to guide staff.

The four initial hypotheses

These were that the absence of seclusion and/or restricted access to PICU care are managed through one or more of the following four mechanisms.

1. Substitution: seclusion for a PICU (or vice versa) or other alternatives such as medication (regular in higher-dose, more intensely sedating medications and/or routes of faster or more effective absorption, intramuscular/intravenous) and/or long-duration manual restraint, time out or open-area seclusion (confinement of a disturbed patient in a separate area together with nursing staff) of some form.

   • The process of substitution was supported by the data. In the absence of seclusion there was greater use of coerced intramuscular medication, PRN medication and nursing in a side room (open-area seclusion). In the absence of an on-site PICU there was greater use of seclusion, de-escalation and observation. No evidence was found for changes in the use of a low-stimulus environment (time out) other than nursing in a side room. Insufficient data were collected to provide an answer on the issue of long-duration manual restraint.

2. Early intervention: faster response to escalation, earlier physical intervention, leading to safer management and more benign outcomes, possibly accompanied by temporarily increased nurse staffing levels or the use of intensive forms of constant special observation (such as two staff to one patient).

   • This hypothesis was not supported. There were no differences found in rates of manual restraint use between the hospitals studied, although the failure to find such differences may be due to the precise content of the TEPI in relation to patient responses to staff actions.

3. Therapeutic intensity: deployment of a range of psychotherapeutic and behavioural interventions to manage, ameliorate and reduce the frequency of risky patient behaviours.

   • This was not supported. In fact, there was one indication that psychological interventions were more likely to be considered if a PICU was available on site. There were few reports at all of endeavours to psychologically understand disturbed behaviour or to respond to it in a creative or therapeutic way. As such, this study supports others demonstrating ‘therapeutic poverty’ in UK psychiatric nursing practice. 87

4. Non-standard transfer: move of the patient to another ward providing more secure or intensive care in some fashion, perhaps in a private hospital or in a neighbouring trust, or via a secure rehabilitation ward or a forensic ward of some type.

   • This was not supported. There were no differences in the rates of non-standard transfers reported as part of the escalation pathway, by whether seclusion or an on-site PICU were available.

Strengths and limitations

One of the strengths of this study is that data were collected directly from those people who were actively and directly in charge of taking decisions about the management of patient behaviour. Through the use of an interview format to obtain data, staff could be asked about the management of relatively rare events that would be difficult or impossible to collect information about in any other way, such as by observation. Standardised scenarios presented to the interviewees during a structured interview meant that differences in their responses were more likely to reflect differences in practices between the hospitals studied. For a qualitative study, this sample size was large, adding to the external validity of the results. Moreover, the sample was not limited to hospitals in London and was, therefore, more likely to be generalisable to acute psychiatric care in the UK as a whole. Differences between the different hospitals were not explicable by different rates of adverse incidents. In other words, hospitals that managed without seclusion of PICU care did not do so because they were caring for less disruptive, less ill and calmer patients. In fact, the indications were for the reverse in relation to PICU care, and it was the presence of an on-site PICU that was associated with calmer and more peaceful wards.

Nevertheless, there are several limitations to our findings. Only eight hospitals were part of the sample, moderating the generalisability of the findings, and the two research areas, London and the North West, may not be wholly comparable. Local policies for the use of seclusion and/or transfer to a PICU may have varied in important ways between the different hospitals, irrespective of the basic availability of each technique, giving rise to the results that have been obtained. The TEPI was a long and complex interview, and interviewers had difficulty in completing it precisely as asked on every occasion. The design of the TEPI itself was constrained by the many additional factors that could influence the escalation pathway, particularly the response of patients to each intervention attempted by the staff. Thus, many aspects of the scenarios had to be constrained to produce standardised situations that would allow us to identify differences between hospital practices. Not the least of these standardisations was the restriction of the scenarios and data collection to focus on male patients only. Although this was a practical necessity, it will have shaped the findings to some unknown degree. In addition, our data may be influenced by the desire of staff to show their wards in a good light; thus, they may have preferentially described ideal rather than actual practice on their wards. Differences in pathway length between the different types of hospitals remain unexplained, and it remains possible that the statistical differences we have described are caused by other local characteristics, rather than the availability or otherwise of seclusion or PICU care. Power and precision would have been reduced by our achieved sample size (n = 81), compared with that planned (n = 120).

Forthcoming

A further paper will be prepared and submitted based on Chapter 4.

Special probe on weapons

[Ask the following questions only if the previous example did not elicit a call for the police. If the interviewee has already said this, we can assume he or she will report the same call for the following two weapon examples.]

BA3A: What if anything would be different if he had a broken bottle in his hand that he was waving threateningly?

or

BA3B: What if anything would be different if he had a kitchen carving knife in his hand that he was waving threateningly?

[No diagram necessary unless participant describes an entirely different chain of events.]

BA4 Behaviour 4: Alan is attacking a fellow patient with punches and kicks, and the assault is still under way when the nurse arrives on the scene.

How would the rest of the staff on this ward manage Alan and his behaviour?

BA5 Behaviour 5: Alan is attacking a fellow patient very seriously by throwing him to the ground is kicking him with maximum available force.

How would the staff on this ward manage Alan and his behaviour?

[If at the end of this example, the interviewee has not yet talked of taking the patient to the floor, explore what happens when they are.]

Alan struggled so vigorously and has been so difficult to contain that the restraint team have taken him to the floor. What happens next? [Use over the page to make a record.]

Special probe on verbal thresholds and other issues

So, Alan has been verbally abusive, smashed furniture, and assaulted another patient. Previously we have said he is unable or unwilling to engage in discussion or negotiation about his behaviour.

BA5A: What if anything would be different in his management if he pleads self-defence on the grounds that he has paranoid beliefs about the person he attacked or hallucinatory experiences (e.g. he says ‘I heard him thinking bad things about me’, or ‘he was going to kill me when I turned my back’)?

[In questions BA4A to BA4F, allow the respondent to answer, then ask the following questions, circling their answer (i.e. whether staff would be ‘more likely’ or ‘less likely’ to respond in a certain way or be ‘just the same’). No diagram necessary unless participant describes an entirely different chain of events. Circle the interviewees’ choices so that we do not have to extract the data from each audio record!]

Would it be more likely or less likely that staff on this ward would invest time in talking to Alan before acting, or be just the same as when he could not communicate?

Would it be more likely or less likely, that staff on this ward would end up restraining Alan on the floor, or be just the same as when he could not communicate?

If restraint did occur, would it be more likely or less likely, that it would last for more than 10 minutes, or be just the same as when he could not communicate?

BA5B: What if anything would be different in his management if he apologises, saying he should not have done it, although the other patient gave him a funny look?

Would it be more likely or less likely, that staff on this ward would invest time in talking to Alan before acting, or be just the same as when he could not communicate?

Would it be more likely or less likely, that staff on this ward would end up restraining Alan on the floor, or be just the same as when he could not communicate?

If restraint did occur, would it be more likely or less likely, that it would last for more than 10 minutes, or be just the same as when he could not communicate?

BA5C: What if anything would be different in his management if he apologises, saying he should not have done it, although the other patient had been calling him names, threatening him, trying to get money and cigarettes from him, and being abusive (and there have been previous similar complaints from others about the patient who has been assaulted)?

Would it be more likely or less likely, that staff on this ward would invest time in talking to Alan before acting, or be just the same as when he could not communicate?

Would it be more likely or less likely, that staff on this ward would end up restraining Alan on the floor, or be just the same as when he could not communicate?

If restraint did occur, would it be more likely or less likely, that it would last for more than 10 minutes, or be just the same as when he could not communicate?

BA5D: What if anything would be different in his management if the person he has assaulted is a small and rather frail older person in their 60s who may have early dementia?

Would it be more likely or less likely, that staff on this ward would invest time in talking to Alan before acting, or be just the same as when the victim was a patient more similar in size and strength to himself?

Would it be more likely or less likely, that staff on this ward would end up restraining Alan on the floor, or be just the same as when the victim was a patient more similar in size and strength to himself?

If restraint did occur, would it be more likely or less likely, that it would last for more than 10 minutes, or be just the same as when the victim was a patient more similar in size and strength to himself?

BA5E: What if anything would be different in his management if the person Alan has assaulted is one of the ward nurses?

Would it be more likely or less likely, that staff on this ward would invest time in talking to Alan before acting, or be just the same as when the victim was a fellow patient?

Would it be more likely or less likely, that staff on this ward would end up restraining Alan on the floor, or be just the same as when the victim was a fellow patient?

If restraint did occur, would it be more likely or less likely, that it would last for more than 10 minutes, or be just the same as when the victim was a fellow patient?

BA5F: What if anything would be different in his management if it was noticed that Alan’s face was flushed, he was a bit unsteady on his feet and he smelt strongly of alcohol?

Would it be more likely or less likely, that staff on this ward would invest time in talking to Alan before acting, or be just the same as when he was sober?

Would it be more likely or less likely, that staff on this ward would end up restraining Alan on the floor, or be just the same as when he was sober?

If restraint did occur, would it be more likely or less likely, that it would last for more than 10 minutes, or be just the same as when he was sober?

Special probe on influence of known history

Can we just go back for a minute to the very first scenario with Alan? You’ll remember he was pacing up and down the ward restlessly, frowning, then becoming verbally abusive and shouting at other patients.

We have been talking about Alan as if he was a new patient that this was his first admission. What if this was actually his second admission, and the last time he was admitted he assaulted a member of staff quite badly, so that she was off work for a week with cuts and bruises?

Now think about Alan pacing up and down the ward restlessly, frowning, then becoming verbally abusive and shouting at other patients.

BA5G: How would the staff on this ward manage Alan and his behaviour?

[Be careful not to lead the interviewee; instead get them to expand on any differences in how they would respond and what they mean. Unpack the phrases they use so you get a good understanding of this.]

Would it be more likely or less likely, that staff on this ward would invest time in talking to Alan before acting, or be just the same as when he had no such history?

Would it be more likely or less likely, that staff on this ward would end up restraining Alan on the floor, or be just the same as when he had no such history?

If restraint did occur, would it be more likely or less likely, that it would last for more than 10 minutes, or be just the same as when he had no such history?

Thank you. We have come to the end of the questions about Alan.

BAE1: Do you want to say anything further about Alan and the management of his behaviour? Anything you can add about his management, or want to clarify about what you have already told us?

[Next question for staff from mixed-sex wards only.]

BAE2: If the case we have been talking about was a woman, would your answers have been any different? If so, how?

[Shouting/abusive, turning over furniture, breaking furniture and wielding chair leg as weapon, assaulting a fellow patient with punches and kicks, throwing patient to floor and kicking with maximum force.]

Is there anything else you would like to add to what you have said, or any additional comments you would like us to take note of?