Study design
This is a retrospective observational cohort study for service evaluation based on case notes, incident registers and legal registers. This descriptive study is intended only as a ‘proof of concept’ study to establish some essential elements for a planned prospective study.
Data Sources
In the Republic of Ireland, the Mental Health Act 2001 requires statutory registers to be kept concerning the use of seclusion, manual restraint and mechanical restraint (41). The authorisations, timing, nature and context of each episode thereof must be recorded on statutory forms for collation and onward reporting to the statutory regulator, the Mental Health Commission (40). There is also a legal requirement that all medication should be recorded when prescribed and administered, with further safeguards concerning consent and documentation under the legislation. These records are subject to inspection by the Inspectorate of Mental Health Services, established by the Mental Health Act 2001 (41). These records formed the basis for the identification of episodes of the use of restrictive and intrusive practices and the ‘context and decision pathways’ that guided them. In addition, the daily combined continuous multi-disciplinary clinical notes (including prescription and administration records for medication) provided a much more detailed and nuanced account of the context, antecedents, behaviours and consequences of such events.
Setting
The Central Mental Hospital is the only forensic psychiatric hospital for the Republic of Ireland, serving a population of approximately 4.8 million. The hospital is divided into units (wards) according to the need for therapeutic security, ranging from acute high security to medium and low security, forming a stratified series of levels of therapeutic security along a recovery pathway from admission to pre-discharge (42,43). All ward-based staff in the hospital are trained in de-escalation and approved methods for prevention and management of violence and aggression (PMVA).
The Central Mental Hospital Dundrum, as with all psychiatric hospitals in Ireland is subject to review by external bodies – the Mental Health Commission (40,41), a National body, which reviews the standards of healthcare provided in all Irish mental health settings and the Committee for Prevention of Torture (CPT) of the Council of Europe. Within the hospital clinicians review the use of restrictive practices regularly as peer review. This takes place in monthly Seclusion Monitoring and Restraint Group (SMARG) meetings. This is the case internationally in most secure forensic hospitals. Clinicians and the hospital are regularly asked to justify the use of restrictive practices and to demonstrate that they were both necessary and proportionate. The aim of the development of the DRILL tool was to support the clinicians in meeting this need .
Sample
All male patients are admitted to a 12-bed unit with a high ward-based staff-to-patient ratio (two to one in whole time equivalents). This male admission unit was built in the 1980s and modernised in 2007. Patients have single en suite bedrooms and are locked in their rooms from 9pm to 8am. There are two seclusion rooms, a small timeout area, a day room and two courtyards while patients have access to a dining room at meal times and for other activities. Patients are not permitted to return to their bedrooms during the day and are expected to attend 25 hours of structured activity each week, which includes psychological and occupational therapies and psycho-education (44).
Ward-based staff are male and female. The high levels of relational security reflect the recent histories of serious violence by patients in the community prior to admission and the continuing levels of risk and positive symptoms amongst the patient groups placed there (43). Two multi-disciplinary teams, each led by a consultant forensic psychiatrist, care for the patients in the male admission unit. All nursing staff had been trained in the therapeutic management of violence and aggression. For the six-month period between 1st January 2011 and 30th June 2011 a total of 16 patients were admitted to this unit. In total, 28 patients were in scope. This paper presents our preliminary findings for patients on the male admission unit (Table 1). Relational therapeutic security levels (ward based staff to patient ratio) were consistent throughout the study period (Table 2). In this service evaluation cohort study, all eligible in-patients were included. None were excluded and none were lost to follow-up. All 28 patients studied were followed until transfer to a less acute unit, hospital discharge, or until the end of the study period.
Variables
Need for therapeutic security, risk of violence.
The DUNDRUM-1 assessment (38,39,45,46,47) was used to measure need for the appropriate level of therapeutic security. The DUNDRUM-1 is a static measure completed prior to admission. An 11 item score is calculated which includes need for suicide prevention while a 9 item score includes only items relevant to violence. Each item is calibrated in units of meaningful change from 0 to 4, so that an item scored 0 indicates no need for admission, 1 indicates a need for an open ward placement, 2 indicates low security, 3 medium security and 4 high security. A mean item score is calculated by adding all items and dividing by the number of items. A mean DUNDRUM-1 item score of 3 to 4 indicates need for high security, 2 to 3 indicates medium security, 1 to 2 indicates low security and a lower mean item score is in keeping with admission to an open ward. Cronbach’s alpha for internal consistency in this sample was 0.668, reflecting the relatively narrow range of scores for patients admitted.
The Historical, Clinical, Risk Management-20 (HCR-20) (48) is completed soon after admission and at six monthly intervals. Data were taken from those assessments completed prior to the study period for those already in hospital at the commencement date and within two weeks of admission for those admitted during the study period. Cronbach’s alpha for the HCR-20-C items – a measure of current dynamic risk of violence – was 0.642.
The mean scores for these instruments may be taken as a measure of the ‘ambient’ risk and seriousness of the risk over the medium term. Similarly, ward atmosphere was measured using the EssenCES (49). The Forensic Satisfaction Scale (50l) and Suicide Risk Assessment and Management Manual (S-RAMM) (51,52) were collected but completed by too few patients to be analysed in this part of the study.
Antecedent factors - milieu
The DASA (33) is an assessment tool primarily based on observation. It was completed daily by nursing staff. Seven items are rated absent or present and include: impulsivity, unwillingness to follow directions, irritability, sensitivity to perceived provocation, easily angered when requests denied, negative attitudes and verbal threats. The DASA score allows an individual management of risk as well as cumulatively providing a contextual assessment of the ward environment. The mean DASA score on any given day may be taken as a measure of the ambient risk or level of disturbance on the ward on a given day. The patient’s primary nurse routinely calculated this each day at 18.00 hours. Cronbach’s alpha was 0.942. Reliable Change Index (53) was 0.56 (range 0-7) indicating that for an individual patient, a unit change was reliable. In an explanatory system consisting of fixed risk factors, stable dynamic risk factors, acute dynamic risk factors and triggers, these imminent items are acute dynamic risk factors (54,55) though they may also be regarded as causal.
We used the DASA for the previous day as an independent predictor and as a co-variate in models of behaviour and interventions on the following day.
Staffing as context
The use of non-permanent staff is recognised to lead to inconsistency and a higher rate of violent incidents (56,57). We recorded the ratio of actual staff to patients for each day, using only those staff allocated to work consistently on that particular unit for the study period. Staff brought in from other units on a casual basis to fill gaps in the rotas were categorised as non-regular staff. The service does not use agency staff. The hospital did not use unqualified nursing staff at the time of this study. This data will be described in a subsequent paper.
Dundrum Restriction Intrusion Liberty Ladders (DRILL)
Development process
This framework was derived following a preliminary review of the literature and clinical practice. A series of rating ‘ladders’ was then drafted by HGK using a conceptual mapping process (Figure 1). We used a modified iterative Delphi process facilitated by HGK to identify and define the elements and magnitudes of restrictive and intrusive interventions. A first draft was circulated amongst experienced forensic psychiatrists with an interest in intensive care (PO’C, DM, RM), nurses with a wide range of experience were consulted, including accredited trainers in PMVA (DT, PB, PMcK, JT) and allied health professionals and each draft was discussed and critically reviewed amongst clinicians drawn from psychiatry, mental health nursing, social work, occupational therapy and clinical psychology. The fourth draft was considered ready for application in this preliminary study. Statements were developed to describe each adverse incident and interventions that were in turn grouped in a hierarchical manner to construct a framework of ‘ladders’. This process was repeated and refined iteratively by experienced clinicians from medicine, nursing, psychology, social work and occupational therapy.
Context decision pathways
Context, antecedents, behaviour, interventions, consequences were conceptualised as a series of events organised in temporal sequence - a context decision path - so that causes, interactions and effects can be considered.
The Delphi process generated a taxonomy of behaviours, interventions and consequences. In the interests of dimensional congruence each ‘ladder’ was rated from 0 to 5. Each ladder was stratified into five main graduations tethered to definitions with sub-divisions for descriptive options.
For behaviours, these could broadly be summarised as: no behaviour, threats, minimal acts, moderate, serious or severe acts. Each gradation was however tethered to clear operational definitions.
For interventions, these gradations corresponded to: no intervention, anticipation, reaction, intrusion, restriction, constraint where zero represents no obvious restriction or intrusion over the ordinary status of in-patient and five represents the most serious intervention legally permitted.
Each behavioural ladder and each intervention ladder should be rated for any single incident dependent on the circumstances, severity and the duration of the context decision path. In a day, multiple episodes can be rated and then summated, or the highest scoring episode can be rated for the day. In this study, the highest scoring episode was taken.
DRILL Behaviour Ladders
Assessment scales were composed for common adverse behaviours (Five ladders: violence, self-harm, risk to others, absconding and non-compliance with treatment; Cronbach’s alpha = 0.720, increasing to 0.740 when self-harm is omitted). For dimensional congruence, the scores for the five ladders were summated and divided by five to yield a mean score with range 0 to 5. The Reliable Change Index for the mean score was 0.36. Each of these behaviours was considered to hold the possibility of leading to restrictive interventions. Each of these adverse incidents equated to a single incident for that type of behaviour. However, the DRILL Behaviour ladders can also occur together in a context decision pathway and may correlate with each other. For example a patient may be non-compliant with medication and may then attack staff trying to administer medication. Such an action - while a single incident - would be recorded under both headings above.
The behaviour ladders are summated because we conceptualised all of these behaviours as being on a continuity with and additive with other threatening or harmful behaviours such as actual violence, or self-harm. In order to prevent imminent violence, skilled clinicians continuously assessed all of these behaviours and used clinical judgement to assess the imminence and seriousness of threatened harm or actual violence. This is done in a planned, synchronised and progressive way, guided also by the DASA ratings in order to anticipate and to plan proportionate interventions, based on a knowledge of the individual patient’s risk assessment, history of violence, mental state, vulnerabilities and advance preferences.
DRILL Intervention Ladders
A set of interventions was formed by the same process into eight ‘ladders’: de-escalation, observations, personal searches, extra medication, situational coercion, manual restraint, seclusion, mechanical restraint. Each ladder was again composed of verbal definitions of ordinal ratings and for dimensional congruence each was scored from zero to five. Cronbach’s alpha calculated for seven items (omitting mechanical restraint which was not used) = 0.876, and for the mean of the summated item scores (range 0 to 5), Reliable Change Index = 0.50.
In Ireland, seclusion is defined as the supervised confinement of the person alone in a room where staff observe continuously from the outside. Seclusion is the most restrictive of the five gradations on restrictions of the use of space (58). Physical or manual restraint refers to the immobilisation of the person by two or more staff (i.e., manual control and restraint). In this study, mechanical restraint did not extend to the use of straightjackets, Pinel restraints or similar devices, practices or equipment since these were never used. A rating for mechanical restraint was included to allow for international comparisons. Manual restraint, seclusion and mechanical restraint can only be initiated by a registered mental health practitioner and must be authorised by a medical practitioner and consultant psychiatrist at fixed, frequent intervals with a view to early termination.
De-escalation was considered to encapsulate a range of interventions from normal communication to conflict resolution. Special nursing observations are used extensively both in the UK (59) Ireland and internationally. This procedure resulted in a restriction of privacy and an intrusion on the patient’s personal space. Likewise, the use of extra medication above and beyond the prescribed regular daily treatment would compromise the patient’s bodily integrity either by the route of administration, the extra dosage or the level of force used. Forensic mental health also includes a number of security-orientated tasks such as personal searches and room searches. While consensual when passing through airports, the loss of dignity and loss of implied consent in hospital, along with removal of possessions or the removal of day clothes constituted a restrictive, intrusive or coercive practice.
A last set of interventions defined a set of sanctions that are considered legitimate in providing a safe environment: detention under mental health legislation, the use of behavioural treatment programmes and the presence of more than one member of staff during negotiations.
The intervention ladders are summated because all of these practices were conceptualised as being on a continuity with and additive with legally regulated practices such as seclusion or restraint. In order to prevent imminent violence, skilled clinicians use anticipation and clinical judgement to deploy not one but a range of the interventions described and quantified here. This is done in a planned, synchronised and progressive way, proportionate to the seriousness and imminence of the threat (8,25,29). The vulnerabilities and advance preferences of the individual can also be taken into account. In this way not only can violence and harm be prevented, but the use of the most intrusive and restrictive interventions such as seclusion and restraint can be minimised.
DRILL Consequences ladders
The context decision path ends with three ladders describing the consequences of the behaviour and interventions and is intended to help structure a ‘debriefing’. In the spirit of an interactive intervention, the debriefing should seek to explore the subjective experiences of the patient, both reinforcing and aversive, as well as the staff perceptions of alienation if any. This stage can be seen as completing the analysis of a context decision path, by exploring the various facets of the events from varying points of view.
DRILL toolkit
This series of rating 'ladders' was structured into an audit toolkit that allowed a qualitatively derived quantitative analysis of the use of restrictive and intrusive interventions. This toolkit (DRILL: Dundrum Restrictive-Intrusion of Liberties Ladders) was structured into a handbook and is available on www.tara.tcd.ie (60).
Context decision path
A context decision path organises the ladders into a chronological sequence. This allowed analysis of any series of events and associated interventions over a meaningful period of time such as a nursing shift, a 24 hour cycle of shifts or for the duration of an adverse episode and related interventions such as seclusion. Each context decision path commences with the antecedent risk (DASA) for the day in question or the previous day and includes the ratings for subsequent behaviours and interventions. Each ends when the last intervention is concluded. For the purposes of this study, each set of observations ends each 24 hours.
In so far as possible, the daily assessment of risk using the DASA tool, or similar instrument, should define the starting point for a context decision path. This allows clinicians to avoid the uninformative formulation of 'unprovoked incident', concentrating instead on the analysis of antecedent context and mental state, the behaviour itself and the consequences for all concerned, whether reinforcing, alienating or neutral.
Data sources and measurement
The registers and records for each patient who spent any time on any of these units were systematically collated and used to rate the outcome measures. All data were collected by two experienced clinicians, entered first into Excel and then in SPSS-25 (61) for analysis. The recording of adverse incidents was elicited from the hospitals’ statutory clinical practice forms for seclusion, mechanical restraint and physical restraint. These incidents were cross-referenced to the hospital’s incident forms for health and safety. The incidents were also cross-referenced with a review of the daily entries in the multidisciplinary clinical notes, medication prescription charts and daily reports to nursing administration for all 181 days. These data were finally cross-referenced with data submitted to the Mental Health Commission and with the register of admissions and discharges for that period.
Two SPSS databases were collated. One Excel database was established to record daily DASA scores by patient, generating a line of data for each day for each patient present on the unit that day (181 X 12 = 2172 patient-days). A second database recorded the mean measures for all patients present for each day, generating 181 lines of data, one for each day. The second data base will be analysed for milieu / contextual effects in a subsequent paper.
Statistics
All calculations were carried out using SPSS-25 (61). Cronbach’s alpha statistic was used to assess internal consistency (62). The Reliable Change Index (RCI) (53) was calculated for the mean item score for DRILL-Behaviour (sum of five items divided by five to yield a range from 0 to 5) and for mean item score for DRILL-Interventions (sum of seven items excluding mechanical restraint, divided by seven to yield a range from 0 to 5) using Cronbach’s alpha (53,62) as a guide to the reliability of unit changes in each measure. If the RCI is less than one unit of change, then at the individual level, a unit of change is significant.
Inter-rater reliability could not be assessed in a meaningful way by rating the same material twice. The internal consistency of the newly developed scales was calculated using Cronbach’s alpha statistic.
Spearman’s rank correlation test was used for preliminary exploratory tests of hypotheses.
The receiver operating characteristic area under the curve (ROC-AUC) was not used as a measure of predictive power when antecedent events were compared with subsequent events because of the repeated measures on the same patients inherent in this design and this clinical context.
In a model in which there is extensive (daily) repeated measures on the same subjects, we opted not to use data smoothing or seasonal adjustments (63) because the time period of six months and limited number of subjects involved made seasonal adjustment potentially misleading. Similarly the exclusion of outliers would be likely to lose important data since it is normal for a small proportion of patients to account for the majority of incidents. Since this was not an A-B study of a change in practice, a corrected regression analysis (25) was also not appropriate. Recent progress in correcting for repeated measures in similar settings and paradigms to this (26,27,28) indicated the use of General Estimating Equations. Repeated measures in the same individuals were studied using General Estimating Equations (GEE) in SPSS-25 (61). A custom model was used in all cases, with normal distribution and identity as the link function. Case number was used to identify the subject variable, day number (1 to 181) as within-subject variable. A main effects model was used for model building using the independent factor then adding covariates, with intercept not included in the model. The scale parameter estimate was maximum likelihood estimation. Model effects analysis was type III and 95% confident intervals. Wald X2 statistics were calculated. Corrected Quasi Likelihood under Independence Model Criterion (QICC) was used to test goodness of fit, with information criteria in ‘smaller-is-better’ form. Competing models were compared on this basis. Marginal means for the independent determinant were calculated.
In a preliminary scoping exercise using 181 days of data for a six bed male intensive care unit, various levels of autoregression were compared with an independent structure for the working correlation matrix. In each model, ‘independence’ yielded a lower QICC than any level of autoregression.
Study size
A goal of this study was to allow calculation of effect sizes that would inform the calculation of power for future prospective studies. The data accumulating for 28 patients in 12 beds over 181 days allowed tests of correlation. Correcting for repeated daily measures within individuals (mean days at risk 6.46) also allowed calculation of effect sizes.
Qualitative variables
Demographic details were recorded for diagnosis, age, length of stay and index offence. All patients were detained under mental health legislation. Other variables included co-morbidity with either personality disorder or substance misuse as neither of these diagnoses makes up part of the legal criteria for detention under Irish mental health legislation. A wealth of qualitative material was accumulated from the Delphi process and this is summarised informally.
Staffing levels
The standard staffing levels for this ward during the period studied wee eight ward-based nurses per day and three ward- based nurses at night and the team is expected to absorb the first special observation within these numbers. Nursing staff work 13-hour and 11- hour shifts on a ‘two days on, two days’ off roster. Of the 57 episodes of special observations at 1:1 nursing, 39 required additional staff. Table 2 shows the actual staffing levels over the course of the study period. The hospital utilises overtime from within the service to manage gaps in the roster but 66% of the nursing team were regularly allocated to the unit for that period with a mean of 5.57 regular staff.