Use of community treatment orders and their outcomes: an observational study

Patterns of use

Unlike compulsory inpatient admission, there has been relatively little research on patterns of use for CTOs. In England, routine administrative data sources suggest that the use varies between demographic groups: for example, in 2016–17 rates of CTO use for males (11.4 per 100,000 population) were almost twice the rate for females (6.6 per 100,000 population), and the rate for black and black British people (60.1 per 100,000 population) was almost nine times the rate for those of white people (6.8 per 100,000 population). 24 Strong patterning on the basis of ethnic characteristics has also recently been found in Western Australia. 26 Few studies have, however, moved beyond the univariate demographic profiles given in routine statistics or considered how these relate to other patient characteristics, such as clinical need. One recent exception in terms of the latter is work in Victoria, Australia, where CTO patients were found to have greater clinical need on the basis of Health of the Nation Outcome Scales (HoNOS). 27

Routine administrative data also suggest that significant spatial variation in CTO use, either between geographical areas and/or service providers. The former is perhaps seen most clearly at a national scale with Australia, New Zealand and some parts of the USA having much higher rates (40–60 per 100,000 of the population) than Canada and New York (around 2 per 100,000 population),28 with England being low to moderate at about 10 per 100,000. 29 There are, however, variations within national jurisdictions, as Australian data emphasises, with estimated rates of CTO use varying between 30 per 100,000 in Tasmania to almost 100 per 100,000 in Victoria. 30 In England, routine statistics suggest significant differences between service providers,25 a point emphasised by recent research exploring these data. 29,31 Again, however, little attempt has been made to relate spatial variations in CTO use to patient demographic or clinical characteristics, or, indeed, to the characteristics of the areas or service providers themselves.

Association with patient outcomes

Much of the existing literature on CTOs has focused on associations with patient outcomes. Following the traditional hierarchy of evidence, the results of the three randomised controlled trials (RCTs) conducted to date (comprising a total sample of 749 patients) have come to assume special prominence. 5

The first two of these trials were conducted in the late 1990s in the USA: one in New York32 and the other in North Carolina. 33,34 Both of the US trials evaluated court-imposed ‘outpatient commitment’ and both excluded patients with a history of violence. Neither trial reported significant differences over 12 months in health service or patient-related outcomes, including re-admission to hospital, compliance with medication or social functioning. 5 Results from the North Carolina trial were more positive, but mostly for a specific patient subgroup: hospital re-admissions and total hospital bed-days were reduced for psychotic patients on longer-term CTOs (≥ 180 days) who received intensive community treatments. 33 Patients on CTOs also showed a reduced rate of violent behaviour and were less likely to be victims of crime over 12 months in the North Carolina study. 34

A third trial has been conducted more recently based on CTOs in England, the Oxford Community Treatment Order Evaluation Trial (OCTET), comprising 333 patients followed up over 36 months. 14 The original main study arising from this trial found that CTOs did not demonstrate improved outcomes on re-admission rates and time spent in hospital over 12 months. 35 A number of additional studies arising from the trial have now been published, including two studies based on longer-term follow-up data and another based on a subgroup analysis. 36–38 These found no evidence of benefits accruing over 36 to 48 months or on a larger number of secondary outcomes, including symptoms, social functioning, substance misuse and a broad range of social/welfare outcomes. In addition, recent work based on CTO recalls over time has been used to counter the suggestion that the original study found negative outcomes because clinicians were not yet familiar enough with CTO regimes for them to be used effectively. 39

The evidence from RCTs has, therefore, demonstrated relatively few benefits for patients. As several commentators have noted, however, RCTs of CTOs face particular ethical and practical challenges that limit their utility. In critically appraising the two US trials, Kisely et al. 3 highlighted small sample sizes, the court-imposed nature of compulsory community treatment and the exclusion of patients with a history of violence.

Although the least prone of the three trials to bias,5 the OCTET trial in England also experienced several methodological limitations, including the involuntary nature of the control condition, high rates of non-participation and patients moving between trial arms after randomisation. 40–42 To comply with ethics standards, the control condition involved Section 17 leave, which is itself a form of compulsory care. In the light of these difficulties, it has been suggested that RCTs may be less suited to evaluating complex mental health interventions such as CTOs. 43

A number of observational studies of CTOs have also been conducted. After excluding uncontrolled studies, this observational research is characterised by matched cohort designs and before-and-after studies, with outcomes generally (but not invariably) assessed over 2 years or less. The results of these studies are more positive overall, with several showing benefits for CTO patients.

Early work in Australia centred on large-scale, population-based data from the states of Victoria and Western Australia. Although the Victorian work showed some positive effects in terms of reduced mortality for those receiving CTOs (a 14% reduction in the probability of non-injury related death and a 24% reduction per day on CTOs for injury related death), findings for key measures such as re-admissions and length of stay were more mixed. 44–46 West Australian work, meanwhile, reported a reduction in bed-days among those placed on CTOs (19.16 days less per episode of hospitalisation), as well as reduced mortality. 47–49 More recent work from New South Wales has suggested a delay in rehospitalisation for CTO patients [incidence rate ratio = 1.47, 95% confidence interval (CI) 1.36 to 1.58]50 whereas the latest work from Victoria has found reduced hospital stays [10.4 fewer days per inpatient episode (INPATEP)], lower rate of mortality (9% less for any cause and 20% less for non-injury) and marginally better quality-of-life scores. 27,51

In North America, a number of studies have used long-term routine administrative data from the New York Office of Mental Health. These have shown a reduced likelihood of re-admission plus increased use of community mental health services for those receiving CTOs, especially when CTOs were implemented for longer (≥ 6 months). 52 Such population studies have also shown more favourable forensic outcomes in terms of fewer arrests. 53 Early work comparing a Canadian province (Nova Scotia) without CTOs to an Australian state (Western Australia) with CTOs showed reduced inpatient stays in the latter,54 and more recent Canadian findings based on a province (Quebec) with CTOs suggest that CTO patients experienced better physical health and reduced mortality. 55

By contrast, there has been only one observational study of CTO outcomes in the UK, an uncontrolled before-and-after study based in Scotland. 56 This absence is particularly surprising given the rich source of administrative data available, as highlighted above in Patterns of use.

Evidence relating to the outcomes of CTOs is, therefore, complex. Discrepant findings may highlight the importance of local context. Together with evidence of variation in use, such findings suggest potential differences between people and places in both the operation and the effects of CTOs.

Ethical and governance approvals

Ethics approval for the study was obtained from Warwick Medical School Biomedical Research Ethics Committee on 12 August 2015 (reference REGO-2015-1623). The data used in the study contained confidential patient-level information and the research team were granted access to anonymised data by the NHS Digital (NHSD) Data Access and Advisory Group. All data analyses took place through the Secure Research Service (SRS) for accredited researchers provided by the Office for National Statistics (ONS) at their safe setting in Titchfield. The Economic and Social Research Council (ESRC)’s Administrative Data Research Network (ADRN) oversaw and facilitated these arrangements.

Brief overview of the Mental Health Services Data Set

The MHSDS is a mandatory data return for all providers of NHS-funded mental health services. It is a patient-level, output-based, secondary uses data set that aims to deliver robust, comprehensive, nationally consistent and comparable person-based information on children, young people and adults who are in contact with mental health services. It supports a variety of secondary use functions, such as commissioning, clinical audit, research, service-planning, inspection and regulation, monitoring government policies and legislation, local and national performance management and benchmarking, and national reporting and analysis.

The MHSDS provides data on a wide range of secondary and tertiary mental health services, including those provided in hospitals, outpatient clinics and in the community (where the majority of people in contact with these services are treated). It brings together key information from adult and children’s mental health, learning disabilities or autism spectrum disorder, Children and Young People’s Improving Access to Psychological Therapies programme (CYP-IAPT) and early intervention care pathways that have been captured on clinical systems as part of patient care.

Although still not problem free, systems for MHSDS data collection, collation and curatorship have improved dramatically since the data set’s original inception. This has been reflected most recently in the decision for MHSDS returns to supplant KP90 returns as the official national statistics source for uses of the MHA. 24

The MHSDS is a large and complex data set. This study used annual ‘sensitive’ extracts from MHSDS for 2011–12, 2012–13, 2013–14 and 2014–15, with the period between 1 April 2011 and 31 March 2015 being referred to, hereafter, as ‘the study period’. Sensitive extracts cover uses of the MHA 1983 and include periods of time when service users were subject to a section of the MHA, on a CTO, or recalled to hospital while on a CTO.

Figure 1 provides a schematic representation of the structure of the MHSDS based on one hypothetical patient’s case history.

FIGURE 1.

Schematic representation of the structure of the MHSDS based on one hypothetical patient’s case history. a, Within MHMDS, the acronym ‘SCT’, supervised community treatment, is used to refer to CTO episodes. DAYATT, number of attendances at day treatment facilities; MHAEVENT, Mental Health Act event episodes; PROVSPELL, Hospital Provider Spell episodes.

As can be seen, annual extracts (downloads to prespecified formats) are structured around care spells – periods of complete continuous care delivered to patients by a care provider. Care spells consist of episodes and events. Episodes are elements of care delivered over extended periods of time. Events are elements of care delivered at specific moments in time. Care spells also have records associated with them that provide key ‘background’ information about the patient receiving a care spell (e.g. sociodemographic characteristics, area of residence) as well as details of the organisation providing it.

Every care spell has a unique identifier (ID) that is present on all episodes, events and records relating to it. Although most patients have only one care spell in any particular year, care spells (and, therefore, care spell IDs) can extend over > 1 year.

Each patient has a unique person ID that is present on all records, episodes and events relating to them. Person IDs can, therefore, be used to link records, episodes and events for patients across all years and all care spells.

Periods of time during which patients are subject to a section of the MHA, admitted to hospital or placed on a CTO constitute episodes. These episodes formed the main data source for the present study. Within MHMDS, the first two are referred to as Mental Health Act event (MHAEVENT) and Hospital Provider Spell (PROVSPELL) episodes. PROVSPELL episodes start at the time a patient is admitted to hospital and end at the time they are discharged from hospital. Time spent receiving treatment under a CTO is referred to in these records as supervised community treatment (SCT) episodes, SCT being the name of the overall legal framework allowing the use of CTOs in England. For the sake of consistency, we use the name CTO for these episodes in this report (see Figure 1).

Since 2011, the dates on which specific care spells start and end, together with the episodes and events that comprise them, have been included. These dates can, therefore, be used together with the unique patient ID code to determine the duration and timing of CTO use and hospital admissions for individual patients.

The patient ID code also facilitates the linkage of MHSDS data with external data sources. In the present study, this was used to link patient records to the ONS Mortality Database to ascertain patient deaths during the study period. Access to, and linkage of, these data sets was facilitated by the ESRC’s ADRN.

Local area of residence

Lower-layer super output areas (LSOAs) are a set of stable geographical areas developed to facilitate estimation and dissemination of census data and neighbourhood statistics by England’s national statistics office. 61 Originally developed for the 2001 Census, modifications were made to their boundaries in the light of population change for the 2011 Census. More recent 2011 LSOA boundaries have only been used in MHSDS data since 2017. Accordingly, this study’s data were based on 2001 LSOAs. In 2001, there were 34,378 LSOAs across England (n = 32,482) and Wales (n = 1896), with a LSOA typically including 672 households and 1614 residents. Equivalent spatial units were also developed in Scotland and Northern Ireland.

For those patients in the ‘ever on a CTO’ group, the LSOA given in the record for the year of the care spell relating to their first SCT episode was used. For patients in the ‘never on a CTO’ group, the LSOA given in the record for the year of the care spell at the end of their first eligible MHAEVENT episode was used.

Using this approach, LSOA identifiers were missing for 2415 patients. Further analysis found: 1154 of these had one LSOA code in records in other years and these were used, 478 had more than one LSOA code in records in other years and the earliest available was used, and 783 had no LSOA codes in records in any other year. Although the majority of these 783 respondents were ‘never on a CTO’ (95.1%), not all of them were, and given their small number as a proportion of the sample overall, they were removed. This left a study sample of 72,171 patients from 23,481 LSOAs in 2001.

Although the vast majority of the included LSOAs or equivalents were in England (n = 23,285, 99.2%), a small number were in Wales (n = 168, 0.7%), Scotland (n = 18, 0.08%) and Northern Ireland (n = 6, 0.03%). The number of patients distributed across English and Welsh LSOAs or Scottish and Northern Irish equivalents were 71,740 (99.4%), 309 (0.4%), 18 (0.02%) and 6 (< 0.01%) respectively. In addition, a very small number of patients (n = 98, 0.14%) were recorded with ‘pseudo-LSOAs’ indicating other jurisdictions of residence rather than local areas (i.e. ‘Northern Ireland’, Channel Islands, Isle of Man and ‘Scotland’). These patients, and their ‘pseudo-LSOAs’ (n = 4, 0.02%), were retained.

Given the sample size and the fine scale of geographical disaggregation, a high percentage of LSOAs contained only one patient. To ensure stable models, it was, therefore, necessary to use middle-layer super output areas (MSOAs), rather than LSOAs. MSOAs are aggregations of LSOAs; in 2001 there were 7194 MSOAs across England (n = 6781) and Wales (n = 413). To facilitate the use of data reflecting local area characteristics contemporaneous with the period covered by the MHSDS data, it was decided that 2011 MSOAs should be used rather than 2001 MSOAs. In 2011, there were 7201 MSOAs across England (n = 6791) and Wales (n = 410), with a MSOA typically including 3245 households and 7787 residents.

Using lookup tables provided by the UK Data Service,62 a two-step process was followed. First, 2001 LSOAs were linked to 2011 LSOAs. Second, 2011 LSOAs were then aggregated to 2011 MSOAs. Following this work, the cohort of 72,171 patients were linked to a total of 6841 2011 MSOAs, 6708 of which were in England and 132 in Wales, with the number of patients in the two countries remaining the same as before (71,740 and 309 respectively). All those patients (n = 122) residing in other constituent parts of the UK (i.e. Scotland, Northern Ireland, Channel Islands and Isle of Man) were assigned to a single pseudo-MSOA, as LSOA to MSOA lookup tables are specific to England and Wales.

Provider trust

The provider organisation code was present for all care spells for all patients because it is automatically derived at the time of data submission. The same process for selecting provider trusts was used as for selecting local area of residence. Hence, for those patients in the ‘ever on a CTO’ group, the provider given in the record for the year of the care spell relating to their first CTO episode was used. For patients in the ‘never on a CTO’ group, the provider given in the record for the year of the care spell at the end of their first eligible MHAEVENT episode was used. In total, this resulted in 77 provider trusts, 67 of which were NHS providers and 10 of which were independent sector providers. Six of the NHS provider trusts recorded in the data were no longer active, and data relating to these were assigned to the 61 active providers based on organisational change information available on these trusts’ websites.

Variation in the use of community treatment orders between patients, places and service providers

The primary outcome variable for the analysis of the use of CTOs was based on the binary categorisation of patients as ‘ever on a CTO’ or ‘never on a CTO’, as defined earlier. A secondary outcome in this part of the analysis was the time in days that ‘ever on a CTO’ patients had been subject to CTOs over the entire study period.

Association between the use of community treatment orders and patient outcomes

The association between the use of CTOs and patient outcomes was assessed according to three outcomes: re-admission, total time in admission and mortality.

Re-admission was defined in two ways: first, by categorising patients according to whether or not they were admitted after becoming eligible for a CTO (whether or not placed on one) and, second, the time between becoming eligible for a CTO (whether or not placed on one) and the start of the first, if any, subsequent period of hospital admission (i.e. time to re-admission). Total time in admission was defined as the total period of time that a patient spent in hospital between becoming eligible for a CTO (whether or not placed on one) and the end of the study period (or the date of death if they had died during the study period). Mortality was defined on an all-cause basis, with patients categorised as either having died during the study period or not.

For data quality reasons, we could not distinguish the cause of admission for the two admission-based patient outcomes just described. Although CTO episodes have a variable giving the reason for the termination of a CTO, this had high levels of missing data. Separate episodes recording supervised community treatment recalls were similarly affected by high rates of erroneous or missing data. Accordingly, we could not differentiate between recall, revocation and other types of admission.

Patient-level variables: sociodemographic characteristics

Data on the sociodemographic characteristics of patients were obtained from the same records referred to in the section on geographical and spatial identifiers. Thus, for patients in the ‘ever on a CTO’ group, the record for the year of the care spell relating to their first SCT episode was used, whereas for patients in the ‘never on a CTO’ group, the record for the year of the care spell at the end of their first eligible MHAEVENT episode was used. Three sociodemographic characteristics were considered: gender, age and ethnicity. Information for age was complete for all 72,171 patients, whereas for gender and ethnicity the level of missing data was low: gender was missing for 395 patients (0.5%) and ethnicity was missing for 880 patients (1.2%). Although several other sociodemographic individual-level variables were present (e.g. employment status, accommodation status), they were not included because of high levels of missing data.

Patient-level variable: clinical characteristics

Data on the clinical characteristics of patients were obtained from an additional type of episode recorded in MHMDS – Payment by Results (PbR) Care Cluster episodes. These episodes reflect each period that a patient is assigned to a PbR Care Cluster (derived from clinician assessment using the HoNOS for PbR) assessment. 64

To avoid misclassification (e.g. due to clinical presentations changing over time and/or with age), the highest care cluster recorded in the years up to and including each patient’s ‘index year’ was used, (i.e. for patients in the ‘ever on a CTO’ group, this was the year of the care spell relating to their first SCT episode, whereas for patients in the ‘never on a CTO’ group, it was the year of the care spell at the end of their first eligible MHAEVENT episode). Care cluster information was less complete than age, sex and ethnicity and was missing for 5211 patients (7.2%).

Patients were grouped as follows according to their highest care cluster code: group 1 (non-psychotic): codes 0 to 9; group 2 (less severe psychotic): codes 10, 11, 14 and 15; group 3 (severe psychotic): codes 12, 13, 16 and 17; group 4 (cognitive impairment): codes 18 to 21; and group 5 (missing): no codes present.

Local area-level variable: neighbourhood deprivation

Neighbourhood deprivation was measured using the English Indices of Deprivation 2015. 65 These are based on 37 separate indicators at 2011 LSOA level and are used to calculate the IMD 2015. These indicators reflect seven distinct domains of deprivation: income; employment; health and disability; education, skills and training; crime; housing and services; and living environment. Only the overall summary IMD 2015 score was used here. To produce IMD 2015 scores for 2011 MSOAs, IMD 2015 scores for 2011 LSOAs were population weighted according to 2011 LSOA population counts based on the lookup tables referred to earlier and averages calculated across the LSOAs within each MSOA. Those patients living in Wales, Scotland, Northern Ireland, the Channel Islands and the Isle of Man (n = 431; 0.6%) were assigned a missing value for IMD 2015. A categorical variable for the IMD score for each MSOA was then created by dividing the score into quintiles, with the least deprived quintile serving as the reference category.

Multilevel models

Multilevel models are widely used for analysing hierarchically clustered data. 66 They provide a statistical framework where the complex structure of real-world settings in which patients live and use mental health services can be considered of substantive interest (i.e. they permit models that mirror the structure of the data). Importantly, they allow variation in the response variable to be apportioned to the appropriate level as well as providing measures of association with characteristics at each level.

As outlined earlier, there are three distinct levels at which variation might occur: between patients, between the local areas in which patients live (MSOAs) and between the settings in which mental health services are delivered to patients (provider trusts). These levels do not, however, nest neatly within each other. Although patients nest within MSOAs, MSOAs do not nest within provider trusts, because patients residing in the same MSOA may receive services from different provider trusts. This produces what is known as a cross-classified data structure. 67 This is illustrated schematically in Figure 2. Such structures can be handled through the use of cross-classified multilevel models (CCMMs).

FIGURE 2.

An illustration of cross-classified multilevel data. Reproduced from Weich et al. 68 This is an open access article distributed in accordance with the Creative Commons Attribution 4.0 Unported (CC BY 4.0) license, which permits others to copy, redistribute, remix, transform and build upon this work for any purpose, provided the original work is properly cited, a link to the licence is given, and indication of whether changes were made. See: https://creativecommons.org/licenses/by/4.0/.

Reproduced from Weich et al. 68 This is an open access article distributed in accordance with the Creative Commons Attribution 4.0 Unported (CC BY 4.0) license, which permits others to copy, redistribute, remix, transform and build upon this work for any purpose, provided the original work is properly cited, a link to the licence is given, and indication of whether changes were made. See: https://creativecommons.org/licenses/by/4.0/.

Cross-classified multilevel models cannot, however, be estimated for all types of response variable. Accordingly, conventional two-level hierarchical MMs based on either patients within MSOAs or patients within provider trusts were used as well as CCMMs based on all three levels.

The most basic MM, whether hierarchical or cross-classified, is the null or empty model. This model estimates the total variance in the response variable without the inclusion of any explanatory variables. This variance is apportioned to each level of the model such that it equals the sum of the variances at each level. Further models, sometimes known as random-intercepts models, can then be developed that attempt to account for this variation by including explanatory variables at each level in the model. These types of models acknowledge that the average relationship (i.e. the regression intercept) between an outcome and any one explanatory variable may differ across different provider trusts and local areas. A final type of model, random-slope or fully random models, can also be estimated in which the effect of one or more explanatory variables is allowed to vary across higher-level settings. These types of models allow the gradient relationship (i.e. the regression slope) between the outcome and any one explanatory variable to vary across different provider trusts or areas.

Variation in the use of community treatment orders between patients, places and service providers

Associations between the use of community treatment orders and patient outcomes

To examine associations between the use of CTOs and patient outcomes, a subset of the full study sample was created consisting of a treatment group (‘CTO’) selected from those ‘ever on a CTO’, and a matched control group (‘non-CTO’), selected from those ‘never on a CTO’.

The treatment group consisted of patients who had been subject to a single CTO episode that had started and finished during the study period. These patients were selected as they had complete information for the start and finish dates of CTO episodes. The matched control group consisted of patients who had at least one MHAEVENT episode coded with an eligible section that had finished during the study period but who had no CTO episodes during the study period.

To prevent bias arising from secular changes in clinical practice, control group patients were frequency-matched to treatment group patients on the basis of the time at which both types of patient were discharged from hospital [i.e. the start date of CTO episodes for CTO (treatment group) patients and the end date of MHAEVENT episodes for non-CTO (control group) patients]. In this way, these dates represent the ‘index event’ for patients in each arm of the study.

Mental Health Act event episodes for control group patients were randomised and a window of ± 42 days around the CTO start dates was applied. Three MHAEVENT episode matches per CTO episode were allowed. Where there was more than one match for a control group patient, one was chosen at random. A wider range of matching variables would have led to considerable difficulties in achieving adequate matches, resulting in a much smaller sample size and reduced external validity. The window of ± 42 days was applied as it provided the best trade-off in terms of closeness in time to the index event and sample size requirements. Using this window enabled us to achieve the approximate 3 : 1 ratio between non-CTO and CTO patients in the final sample.

This matched subsample was used for the analysis of the three patient outcomes defined earlier (re-admission, total time in admission and mortality) as well as the analysis of the health-care costs associated with CTOs. The following sections provide further detail about each of these outcomes and outline the way in which each was analysed. Once again, these methods centre on the use of MMs, although the same set of covariates was used as in the work analysing the use of CTOs (i.e. gender, age, ethnicity, care cluster group and local area deprivation), the modelling strategy was slightly different. First, an additional covariate was included indicating patients’ treatment status (i.e. CTO or non-CTO). Second, rather than estimate ‘null’ models with no covariates, models with only the treatment status covariate were estimated. In this way, the models provided an estimate of the effect of CTOs on patient outcomes before and after adjusting for the effect of other covariates.

Variation in the associations between the use of community treatment orders and patient outcomes

In order to examine variation in the association between the use of CTOs and patient outcomes, models were estimated in which the covariate for treatment status was allowed to vary across provider trusts, producing what are known as random-slopes or fully random models.

In these models, the effect for treatment status on patient outcome is not, therefore, fixed (i.e. uniform) across all provider trusts, but can be different in different trusts. Crucially, this difference is not estimated through the inclusion of a set of separate ‘dummy’ variables, one for each trust, but through the addition of an extra variance and covariance term at provider trust level. This formulation brings a number of benefits, most particularly the ability for information about the association between CTO use and patient outcomes to be shared so that the estimates of variation ‘borrow strength’ from where there is the greatest amount of information (i.e. the provider trusts with the largest sample size).

Association between community treatment orders and health-care costs

The matched subsample was also used to examine the associations between the use of CTOs and health-care costs for patients. As outlined earlier (see Association between the use of community treatment orders and patient outcomes), the treatment group comprised those patients who had been subject to a single CTO episode that had started and finished during the study period, whereas the control group had at least one MHAEVENT episode with an eligible section, but had no CTO episodes.

The key cost driver associated with this population group is mental health hospitalisation. 35 Consequently, the analysis of costs associated with the use of CTOs focused on the burden of inpatient admission. To estimate the costs associated with CTO vs non-CTO we measured the health-care resource used by both groups, and then attached unit costs to calculate total cost, before assessing difference in costs between the two groups. The primary resource use in this study was days in inpatient admission, whereas the unit cost for each admission day was sourced from the Personal and Social Services Research Unit (PSSRU) Costs of health and social care 2017. 77

The PSSRU unit cost used was sourced from the NHS reference cost data for ‘per bed-day’ mental health care clusters. 77 This unit cost was used because costing was reliant on the output of the prior total admission data analysis. Costs were estimated using the output of the hierarchical MMs with the negative binomial link model used for the total admission analysis, and consequently the results on which the costs were estimated are at an aggregated level. Given this, the average PSSRU unit cost was chosen (Table 2).

There is additional burden associated with delivering CTOs including increased visits by community MHTs, mental health review tribunals (MHRTs), increased chance of SOAD assessment; and clinician time completing capacity assessment and consent to treat forms. In consultation with two experienced psychiatrists who implement CTOs (SW and PK), the incremental burden of each of these activities relating to CTO patients was also estimated. This was an iterative process to identify all of the elements of burden associated with the implementation of CTOs and the resource use associated with each. PSSRU Unit costs77 were attached to these to create a fixed cost associated with the implementation of a CTO.

Three analyses were conducted: (1) the primary analyses focused solely on the costs associated with time in inpatient admission between CTO and non-CTO patients, (2) a secondary analysis included the incremental burden of CTO use in addition to inpatient admission data to assess the difference in cost when accounting for the extra burden of CTO implementation and (3) a final sensitivity analysis considered only severe psychotic patients.

To examine difference in cost of inpatient admission between CTO and non-CTO groups, results from the time in admission models described earlier were used (see Total time in admission). Estimates from these models were exponentiated to give mean daily rates of admission. These were multiplied by a period of risk (365 days) to give the total number of days in admission per year. Unit costs were attached to derive cost estimates for patients in both CTO and non-CTO groups. Results are presented as mean costs with associated 95% CIs.

‘Ever on a CTO’ versus ‘never on a CTO’

Initial comparison of NHS provider trusts revealed large variation in the likelihood of eligible patients being on a CTO, which was due to the presence of two outlier trusts. Although the likelihood of being ‘ever on a CTO’ ranged from 0% to 37.3% across 59 of the 61 NHS provider trusts (with an average of 21.4%), the proportion of eligible patients who were subject to CTOs was 82.8% and 98.6% in the other two trusts. Under the terms of use agreed with the NHS Data Access and Advisory Group, we are not able to name these, or any other, service providers.

The results of CCMMs based on patients, MSOAs and provider trusts for the binary response of ‘ever on a CTO’ (1) and ‘never on a CTO’ (0) are shown in Table 4.

The null model, Table 4, shows that although most variation in the use of CTOs occurred between patients (70.2%), there was a significant amount of variation between MSOAs (5.8%) and between provider trusts (24.0%). When patient- and MSOA-level covariates were included in the model, the amount of variation between MSOAs and provider trusts was reduced slightly to 4.5% and 22.4%, respectively.

Analysis of the provider trust-level residuals from this second model confirmed the existence of two dominant outliers where the use of CTOs was considerably more common than that for other provider trusts, and substantially beyond that predicted by the types of patients and places (MSOAs) that they served (Figure 4). In these two trusts, the prevalence of ‘ever on a CTO’ predicted by the model that took account of patient and MSOA-level covariates was 97.7% and 71.1% respectively, whereas across the other trusts this ranged from 1.2% to 18.5%.

FIGURE 4.

Provider trust-level residuals for the model in Table 4 including all provider trusts.

A third model excluding these two provider trusts, showed much less variation between the remaining providers (4.2%), with the amount of variation between MSOAs being slightly greater (5.8%). The influence of the two outlier trusts was confirmed on examining the provider trust-level residuals from this third model (Figure 5). As can be seen, however, there were still a number of remaining providers at both tails of the distribution where levels of CTO use differed significantly from the overall average (i.e. where the CI around their point estimate does not straddle the dotted horizontal line is the overall average). In total, 15 provider trusts had rates of CTO use that were significantly higher than the overall average, and nine had rates that were significantly lower.

FIGURE 5.

Provider trust-level residuals for the model in Table 4 excluding two outlier trusts.

Excluding the two outlier provider trusts from the models did not have a marked effect on the estimates of the associations between CTO use and either patient or neighbourhood characteristics. As a result, tests for association between CTO use and patient characteristics were based on the results for all patients (i.e. Table 4). These results are shown graphically in Figure 6, where the vertical line for an OR of 1 relates to the reference category for each patient and local area characteristic.

FIGURE 6.

‘Ever on a CTO’ versus ‘never on a CTO’, by group category.

As can be seen, women had lower odds of being on CTOs than men (OR 0.70, 95% CrI 0.67 to 0.73). Patients aged ≥ 65 years were the least likely to be on CTOs (OR 0.61, 95% CrI 0.44 to 0.84) compared with those of other ages, all of whom were not significantly different from each other. Black patients had a significantly higher rate of CTO use than the white reference group (OR 1.43, 95% CrI 1.33 to 1.53) as did patients of mixed ethnicity (OR 1.27, 95% CrI 1.13 to 1.43). The largest associations were with care cluster grouping, with those in the ‘severe psychotic’ group (group 3) being almost four times more likely to be on a CTO than the ‘non-psychotic’ reference group (OR 3.76, 95% CrI 3.45 to 4.11).

At MSOA level, being on a CTO was associated with socioeconomic deprivation in a manner that suggested a dose–response relationship. ORs for being on a CTO rose to a peak of 1.13 (95% CrI 1.03 to 1.24) for those living in MSOAs in the top two (most deprived) quintiles compared with those living in the least deprived areas. Taking account of socioeconomic deprivation did not, however, substantially reduce the variation between MSOAs, and this was particularly true when the two outlier provider trusts were excluded (see Table 4).

Time subject to community treatment orders

An analysis of the total time, in days, that ‘ever on a CTO’ patients were subject to CTOs over the whole study period was also carried out. As the distribution of these times was substantially overdispersed owing to a small number of patients being on CTOs for much longer periods of time than the rest, a negative binomial link function was used. As outlined earlier (see Chapter 2, Strategies for analysis), CCMMs cannot be estimated for this link function as MCMC methods have not yet been developed. Accordingly, a series of two-level hierarchical MMs were estimated in which variation between provider trusts and between MSOAs was considered separately rather than simultaneously. Table 5 shows the results for models based on patients within provider trusts, whereas Table 6 shows the results for models based on patients within MSOAs.

All of the models in Tables 5 and 6 included an offset variable to take account of the differing amounts of time (in days) that ‘ever on a CTO’ patients were ‘at risk’ of being on a CTO. As a result, it is the proportion of an ‘ever on a CTO’ patient’s time at risk that is spent on CTOs that was modelled, rather than the absolute number of days they spent on CTOs. In this way, estimates from these models give the rate of CTO use per day (i.e. the likelihood, on any given day, that a patient would be on a CTO), with higher daily rates indicating a greater incidence of days on CTOs. When exponentiated, these estimates result in RRs, that have the same interpretation as ORs but for rates rather than odds.

As can be seen, estimates of between-provider and between-neighbourhood variation remained statistically significant after taking account of patient and neighbourhood characteristics, with variation in the daily rate of CTO use appearing to be greater between MSOAs (MRR 1.43, 95% CI 1.41 to 1.45) than between provider trusts (MRR 1.25, 95% CI 1.20 to 1.30). It is important to remember, though, that variances were estimated separately here, not simultaneously as in the CCMMs reported in the previous section.

Provider trust-level residuals for the model in Table 5 are shown in Figure 7. The two outlying trusts reported in the previous section were also unusual here, though less prominently so, appearing this time with the largest negative residuals (i.e. with their patients spending less time subject to CTOs, on average, than the mean for all trusts). There were also a number of other provider trusts at both tails of the distribution that were significantly different from the overall average (shown again by the horizontal line). In total, 13 provider trusts were significantly higher than the overall average and 12 were significantly lower.

FIGURE 7.

Provider trust-level residuals in Table 5 including all provider trusts.

As estimates of the effect of patient and area characteristics were similar when they were included in the model of patients within provider trusts and the model for patients within MSOAs, respectively, only the results from the model in Table 5 are shown graphically in Figure 8. The vertical line for a RR of 1 relates to the same reference categories for patient and local area characteristic as in Figure 6.

FIGURE 8.

Time subject to CTOs for ‘ever on a CTO’ patients, by group category.

As can be seen, women were subject to a lower rate of CTO use per day than men (RR 0.93, 95% CI 0.92 to 0.95) and people aged 36–64 and ≥ 65 years were subject to a higher rate than people of other ages (RR 1.20, 95% CI 1.05 to 1.37 and RR 1.23, 95% CI 1.08 to 1.41, respectively). Although black people experienced a higher rate of CTO use per day than any other ethnic group, the effect size was not large or significant (RR 1.02, 95% CI 1.00 to 1.05). In terms of care cluster group, the largest effect size was for those in the severe psychosis category (RR 1.22, 95% CI 1.18 to 1.26). All of the effect sizes for socioeconomic deprivation at MSOA-level were small and non-significant.

Comparing Figure 8 with Figure 6 we find a noticeable difference. The likelihood of being on a CTO is strongly patterned across a range of patient characteristics, with some groups of patients being much more likely than others to be on a CTO (see Figure 6). By contrast, the degree of patterning is much reduced for the time that CTO patients spend on CTOs, with no patient groups (as characterised by the covariates available in our data set) spending particularly longer or shorter periods of time on CTOs, although some are still statistically significantly different from others (see Figure 8).

Description of the matched subsample

As described earlier (see Chapter 2, Strategies for analysis), analyses of associations between the use of CTOs and patient outcomes were conducted using a matched subsample. This was drawn from the final study sample of patients in NHS provider trusts and comprised:

• Treatment group (‘CTO’) – patients who had been subject to a single CTO episode that had started and finished during the study period.

• Matched control group (‘non-CTO’) – patients who had at least one MHAEVENT episode coded with an eligible section that had finished during the study period but who had no CTO episodes during the study period.

The final matched sample comprised 6329 treatment group (CTO) patients and 16,842 control group (non-CTO) patients. The sociodemographic, clinical and area of residence characteristics of the patients in these two groups are shown in Table 7. As with the main study sample, CTO patients were more likely than non-CTO patients to be male, aged < 65 years, of minority ethnic origin (particularly black), belong to the ‘severe psychotic’ care cluster group and live in more deprived areas (Table 7).

Figure 9 shows the distribution of the total time, in days, during the study period that the CTO patients were subject to CTOs. As before (see Figure 3), the distribution is considerably over dispersed owing to the tail of patients subject to CTOs for long periods of time, although the mean (197.7 compared with 498.2) and standard deviation (222.6 compared with 507.2) are much reduced.

FIGURE 9.

Distribution of the total time during the study period that CTO patients in the matched subsample (n = 6329) were subject to CTOs.

Three patient outcomes were studied: re-admission, total time in admission and mortality. Results for each outcome based on all NHS provider trusts now follow.

Re-admission

In the treatment group, 36.9% (n = 2335) of CTO patients were recorded as being admitted to hospital after their index event. The remaining 3994 (63.1%) CTO patients were not re-admitted to hospital during the study period and were, therefore, censored at the end of the study period (or, date of death, if applicable, whichever was earliest).

In the control group, 35.6% (n = 5999) of non-CTO patients were re-admitted during the study period and 64.4% (n = 10,843) were censored at the end of the study period (or date of death, if applicable, whichever was earliest).

Based on Kaplan–Meier estimates, the mean time to re-admission was 967.0 days (95% CI 951.4 to 982.6 days) for CTO patients and 957.9 days (95% CI 948.1 to 967.7 days) for non-CTO patients. Thus, although CTO patients had a slightly longer time on average to re-admission (9.1 days), the overlapping CIs suggested that there was little difference in the two groups in terms of average time to re-admission.

These results took no account, however, of patient or local area characteristics. To do this, a series of discrete-time models were estimated. To facilitate this approach, times to re-admission/censoring were grouped into 90-day periods, with time to censoring (i.e. for those not re-admitted to hospital during the study period) being defined as the interval between a patient’s index event from CTO start (CTO group) or discharge from hospital (non-CTO group) to the end of the study period or, for those who died during the study period, the date of death. A Kaplan–Meier survival plot based on these time periods is shown in Figure 10 and indicates a high degree of similarity in re-admission over time between the two groups.

FIGURE 10.

Survival (time to re-admission) curves for the two treatment groups based on 90-day periods.

As discrete-time models are particularly computationally intensive (i.e. take several days to run) when estimated as CCMMs using MCMC methods in MLwiN, a series of single-level discrete-time models were first estimated in SPSS. These showed that there was no significant interaction between treatment status and time period, (i.e. the hazard profile over time was similar for each treatment group). Interaction terms between treatment status and care cluster group were, however, statistically significant, and so these were included in a four-level CCMM (in MLwiN).

The results for discrete-time CCMMs are shown in Table 8. Taking the model with time periods and treatment status only (Table 8), results showed the rate of re-admission reduced over time. Because interaction terms between time period and treatment status were not required, this reduction is the same for each treatment group and is shown graphically in Figure 11.

FIGURE 11.

Rate of re-admission (log-odds) by 90-day period and treatment status.

The results from this model also showed that, when the multilevel structure was recognised, CTO patients were estimated as having a slightly overall greater rate of re-admission than non-CTO patients before taking account of patient and local area characteristics (OR 1.12, 95% CrI 1.06 to 1.18). Although this appears to contradict the results of the Kaplan–Meier analysis, differences can be expected given that the estimates of MMs adjust for autocorrelation and have a cluster-specific interpretation (i.e. they relate to the typical community) rather than a population-averaged interpretation (i.e. they relate to the average across all communities). 78

Taking the model with other covariates included (Table 8), we see first that the estimates for time periods were virtually the same, thus the hazard profile shown in Figure 11 remained unchanged after taking account of other patient and local area characteristics. Because interaction terms between treatment status and care cluster group were included in this model, the interpretation of the effect of treatment status on the rate of re-admission is not straightforward as it is different for different care cluster groups. Figure 12 uses the relevant terms to show these differential effects graphically. As can be seen, although there was a suggestion that the rate of re-admission was lower for CTO patients in the non-psychotic, less severe psychotic and severe psychotic care cluster groups than for the non-CTO patients in these groups, the effects were small (and non-significant). However, the rate of re-admission was significantly higher for CTO patients in the cognitive impairment care cluster group than for the non-CTO patients in that group.

FIGURE 12.

Rate of re-admission (probability) by treatment status and care cluster group.

In terms of other patient and local area characteristics, women had a slightly higher rate of re-admission than men (OR 1.08, 95% CrI 1.03 to 1.13) and Asian patients had a lower rate than white patients (OR 0.83, 95% CrI 0.76 to 0.91). There was no clear patterning on the basis of local-area deprivation, although the rate was significantly higher in the third and fifth quintiles.

There was significant variation between both MSOAs and provider trusts in the rate of re-admission, even after taking account of all patient and local area characteristics, with the former being slightly greater (MOR 1.21, 95% CrI 1.05 to 1.30 vs. MOR 1.18, 95% CrI 1.14 to 1.24 respectively).

Total time in admission

In the treatment group (CTO), 35.0% (n = 2216) of patients were recorded as having at least one hospital admission of known duration (i.e. the end date was present or could be imputed), with each of these patients having, on average, two admissions. In the control group (non-CTO), 33.7% (n = 5674) of patients were recorded as having at least one hospital admission of known duration (i.e. the end date was present or could be imputed), with each of these patients having, on average 2.2 admissions. A small number of patients (325 non-CTO patients and 119 CTO patients) had PROVSPELL episodes (i.e. admission end dates) that could not be imputed.

For all patients in each treatment group (i.e. not just those who had admissions), the average total time in admission across the study period was 54.4 days (standard deviation 144.3 days) for the CTO group, and 46.9 days (standard deviation 120.2 days) for the non-CTO group. CTO patients, therefore, spent 7.5 days longer in admission on average across the study period than non-CTO patients. It should be noted that the median value for the total time in admission was zero for both groups, as the majority of patients were not admitted to hospital during the study period.

Table 9 presents the results of two, two-level hierarchical MMs of patients within provider trusts based on all 61 NHS provider trusts for the total time in admission. As with time subject to CTOs in the previous section, total time in admission was substantially overdispersed with most patients having short periods of time in admission, whereas a small number had considerably longer periods. Accordingly, a negative binomial link function was used and hierarchical MMs rather than CCMMs were applied. An offset variable was, again, included to take account of time ‘at risk’. Thus, the outcome modelled was the rate of admission per day (i.e. the likelihood of being in hospital on any given day), with higher daily rates indicating a greater total time in admission. As before, when exponentiated, these estimates resulted in RRs.

As can be seen, CTO patients had a significantly higher daily rate of admission (RR 1.34, 95% CI 1.24 to 1.44) than non-CTO patients before taking account of patient and local area characteristics. Although this was reduced adjusting for these characteristics, their daily rate of admission remained significantly higher (RR 1.20, 95% CI 1.10 to 1.30) than those not subject to CTOs.

A model that included interactions between treatment status and care cluster group was also estimated. These additional terms were not statistically significant and did not improve the fit of the model and are not reported here. As a result, what follows is based on the model in Table 9, with treatment status, patient and MSOA covariates.

At patient-level, Asian patients (RR 0.82, 95% CI 0.72 to 0.94) had a lower daily rate of time in hospital than white patients, whereas patients in the less severe psychosis (RR 0.78, 95% CI 0.70 to 0.88) or cognitive impairment (RR 0.54, 95% CI 0.46 to 0.62) care cluster groups had lower daily rates than patients in the non-psychotic group. RRs for patients in the severe psychotic group, meanwhile, were not significantly different from the non-psychotic group (RR 0.96, 95% CI 0.87 to 1.08). There was no significant patterning on the basis of gender, age or local area deprivation.

There was significant variation in the daily rate of admission between all NHS provider trusts, even after taking account of all patient and local area characteristics (MRR 1.24, 95% CI 1.15 to 1.31), with the effect size being greater than any of these characteristics.

As this outcome was likely to have the greatest impact in terms of treatment costs, a sensitivity analysis based on the 59 provider trusts that remained after the two outliers were excluded was conducted. Results for this analysis are provided in Appendix 2. They did not show any marked differences from the results for all provider trusts.

The fixed part estimates of the model in Table 9, with treatment status, patient and MSOA covariates were used to obtain predictions of each patient’s daily rate of admission. Using these, the average number of days in hospital per year, based on all patients, was 27.1 days (95% CI 26.9 to 27.2 days) for those on CTOs and 20.5 days (95% CI 20.4 to 20.6 days) for those not on CTOs. The corresponding figures based on severe psychotic patients only, the care cluster group most likely to be on CTOs, were 29.7 days (95% CI 29.5 to 29.8 days) and 24.4 days (95% CI 24.3 to 24.5 days).

Mortality

In total, 98.4% (n = 22,799) of patients in the matched subsample could be linked to the ONS mortality database. Of these patients, 2143 (9.4%) were recorded as having died during the study period. The 1.6% (n = 372) of patients whose records could not be linked to mortality data were not concentrated in any particular provider trust or care cluster and did not appear to differ in these respects from those whose records were linked. Table 10 shows the distribution of deaths in the study period according to treatment status; as can be seen, 280 (4.5%) CTO patients died during the study period, whereas 1863 (11.2%) non-CTO patients died. Table 11 shows the distribution of cause of death by treatment status.

Owing to small numbers, the analysis of mortality was conducted on an all-cause basis. Results for CCMMs based on patients, MSOAs and provider trusts for the binary response of ‘died in study period’ (1) and ‘stayed alive in study period’ (0) are shown in Table 12. As can be seen, CTO patients were much less likely to die than non-CTO patients, particularly before (OR 0.37, 95% CrI 0.32 to 0.43) (Table 12), but also after taking account of patient and local characteristics including age, gender, ethnicity and care cluster group (OR 0.69, 95% CrI 0.60 to 0.81) (see Table 12, Model with treatment status, patient and MSOA covariates). A model with interaction effects between treatment status and care cluster group was also estimated but the additional terms were non-significant and did not improve the fit of the model (results not reported here). As a result, the remainder of the discussion relating to Table 12 is based on the results from the model with treatment status, patient and MSOA covariates.

As predicted, there was a clear association between age and likelihood of death, with the latter being highest among those aged ≥ 65 years (OR 13.65, 95% CrI 4.39 to 52.40). There was also a higher likelihood of death among patients with cognitive impairment (OR 2.05, 95% CrI 1.71 to 2.46). Women had a lower likelihood of death compared with men (OR 0.72, 95% CrI 0.65 to 0.69). Patients in all of the main ethnic minority groups had a lower likelihood of death compared with white patients. There was no significant association with care cluster group or local area deprivation.

There was substantial variation between provider trusts and, particularly, between MSOAs before taking account of patient and local area characteristics (MOR 1.57, 95% CrI 1.41 to 1.77, and MOR 2.02, 95% CrI 1.79 to 2.27, respectively). After taking account of these characteristics, however, this association was considerably reduced (MOR 1.18, 95% CrI 1.07 to 1.30, and MOR 1.16, 95% CrI 1.04 to 1.33, respectively), although it may be that this lack of variation is due to the comparatively small number of deaths in the subsample.

Sheffield event for mental health professionals

The first event took place in Sheffield on 2 May 2018, at Fulwood House, the headquarters and training base for Sheffield Health and Social Care NHS Foundation Trust. The event, which formed part of the monthly Regional Grand Round programme for South Yorkshire, was attended by a range of mental health clinicians, service managers and other health and social care professionals working in South Yorkshire and North Derbyshire. Many of those attending (including consultant psychiatrists) had extensive experience of caring for and treating patients subject to CTOs. Approximately 70 staff attended the half-day event that included presentations from four members of the research team, a plenary question and answer session, small group discussions, feedback and a final summary. The event was chaired by Scott Weich.

The discussions covered the following topics: the benefits of CTOs, possible disadvantages of CTOs, reasons for geographic variation in rates and duration of use, and alternative options for reducing compulsory detentions.

London event for service users and carers

The second event took place in London, at the Mental Health Foundation (MHF)’s main office on 19 July 2018. The format was based on learning from previous public and patient involvement events facilitated by the MHF in response to feedback from service users and carers about how to make such events accessible, and also in line with the INVOLVE standards 1–4 that consider accessibility (both physical access and the use of accessible language with timely and appropriate communications throughout the engagement process). The format included provision of summary materials in advance; a mid-morning start time to allow comfortable travel times; refreshment breaks and lunch provided; travel arranged on people’s behalf and out-of-pocket expenses settled on the day if requested. This sometimes required direct one-to-one support for participants, tailoring the practical support to each individual to ensure that they had support needs met both in advance and on the day. This was delivered by MHF staff with considerable experience of face-to-face work with those who have used mental health services but were less familiar with academic research.

The event was attended by 10 people who had either been subject to a CTO or provided informal care (as a family member or friend) to someone who had (two people had both of these experiences). People were recruited through existing MHF networks and most participants had previously participated in research as members of service user and carer reference groups (and knew each other either through this research or through their own support networks). Some were new to involvement in research. The majority (seven) were based in London with three people travelling from Hampshire. The group was balanced in terms of gender (six male, four female) and had a degree of ethnic diversity (seven white British, one Asian British, two black British).

The half-day event was chaired by David Crepaz-Keay and included presentations from three members of the research team and covered the following topics: the history and scope of CTOs; trends in rates of detention; an overview of the study and its predecessor; and the study’s initial findings, including rates and variations in the use of CTOs. The presentations were followed by whole group facilitated discussions based on three questions, namely:

1. What are your experiences of CTOs?

2. What do you think about the study findings?

3. What do you think might help to reduce rates of detention?

The agenda was approached with a degree of flexibility to ensure that contributors had an opportunity to articulate broader views before returning the focus directly to the research questions and interim findings. This meant that the discussion was wide ranging and prompted much thought that extended beyond the remit of the research but enhanced the experience for individual participants.

Benefits of community treatment orders

Disadvantages of community treatment orders

Some of the perceived disadvantages of CTOs reflected a different perspective on the advantages outlined above.

Ways to reduce compulsory admissions

The consultation event concluded with an invitation to participants to reflect on the wider issue of rising rates of compulsory admission under the MHA, and to consider actions to address this. Most professionals expressed concern at the continuing rise in rates of compulsory admissions and greater concern was expressed at the rising proportion of inpatients who were detained under section compared with informal patients. Many thought that this was making inpatient settings less therapeutic for all patients and more challenging for staff to manage.

Most professionals considered that compulsory detention was necessary under some circumstances but identified several ways that the need might be reduced.

One proposed approach was stronger integration of health and social care that would enable people to get rapid appropriate support in their own homes, either before admission became necessary or as part of a discharge process. Participants also reflected on the need to see care in holistic, rather than purely medical terms.

A number of people suggested that closer collaboration with local police could be valuable, so that mental health needs would be considered and addressed more quickly without the need for formal detention. In many places, there has been a policy change that ensured that those detained on Section 136 are assessed in health-based settings rather than at police stations. This may have contributed to more Section 136 detentions being converted to inpatient detentions.

Experiences of community treatment orders

One theme that emerged, and was striking, was that many participants were not entirely familiar with the rationale for, and workings of CTOs, even if they had been subject to them. When people had been subject to a CTO, they had often been confused about what they could and could not do as a result, and particularly what the consequences might be of discontinuing medication. It was also evident that most participants did not understand what they needed to do to be released from a CTO.

There was some difference in views between service users and carers about whether CTOs were a good or bad thing. Where carers thought that the medication was effective and that the person they cared for had a tendency to withdraw from medication after discharge from hospital, they thought that CTOs may encourage longer medication use that they saw as beneficial. Similarly, where people who had been detained patients saw CTOs as a way of spending less time in hospital, they favoured CTOs over extended detention.

Participants had a variety of experiences of CTOs. They were generally perceived as an increased form of control (i.e. coercion), which was definitely not welcome. Most people also thought that CTOs had an adverse impact on relationships with staff. Some people thought that good therapeutic relationships led to care and treatment that was agreed by both clinician and patient and that this agreement made the coercion associated with CTOs unnecessary. Other people described how relationships had been damaged by the threat of compulsion that CTOs carried.

One person gave a very detailed description of her experience of CTOs. She described them as counter-therapeutic (in her experience) and something that prevented her from managing her medication in the way she found most effective. The group broadly agreed that CTOs ‘assume’ that medication compliance is ‘right’ and did not seem to offer flexibility or choice that some people highlighted as important in getting the right treatment. Most participants thought that the availability, accessibility and quality of community support was much more important to recovery than CTOs.

Responses to research findings

The complicated nature of both the research and its findings was reflected in the participants’ reaction to the interim findings. Most of the service user or carer participants were not used to quantitative analysis and so the presentation and interpretation of interim results provided an opportunity to explore ways of making the complicated findings more accessible to wider audiences.

Participants were interested in the research findings and some were surprised by the range of differences in the use of CTOs, in particular by provider trusts that used them in nearly all discharge procedures. In these areas, people thought it showed that services were not responding to individual need. Others were less surprised and had experience of receiving different approaches from different psychiatrists and thought it was likely that psychiatrists’ views (and uses) of CTOs would vary.

Participants also raised a number of questions, including whether or not CTOs were more effective if they were used for longer, whether or not people on CTOs received more or better services, and whether or not CTOs cost money or took up time that could be used for other things.

Ways of reducing compulsory admission

Suggestions for reducing compulsory admissions

The following suggestions were offered as potential ways of reducing compulsory admission:

• More support for carers, who are having to spend more time on providing care at personal financial cost, and at a cost to their own mental and physical health, particularly as they got older.

• Quicker access to crisis services, so that people could get help when they needed it – before compulsory detention became necessary.

• More day facilities to reduce the isolation people experience during the day.

• More choices of and alternative approaches to medication, which might also encourage those who are put off from seeking help if they think medication is all they will be offered.

• Support with financial and housing difficulties, including problems with neighbours and experience of discrimination. These are often issues that might ‘tip people over the edge’ and into a crisis that may lead to compulsory admission.

Variation in the use of community treatment orders (aim 1)

Our first aim was to examine variation in the use of CTOs between patients, local areas and service providers. We began by estimating null or ‘empty’ MMs to assess the total degree of variation at each of these levels. Next, we estimated models with covariates to show how this variation related to patient and local area characteristics, and the effects of adjusting for these variables on higher-level variation in the use of CTOs. Provider-level residuals from these models were then examined to identify service providers where variation did not relate to these characteristics.

Association between community treatment orders and patient outcomes (aim 2)

The second aim of the study was to examine the association between the use of CTOs and patient outcomes. To achieve this, we estimated MMs with a single covariate denoting treatment status (CTO vs. non-CTO). We then estimated models with additional covariates denoting other patient and local area characteristics to establish the independent effect of treatment status. We also considered the degree of variation between local areas and provider trusts in each outcome before and after taking account of patient and local area characteristics. Three patient outcomes were studied: re-admission, total time in admission and mortality.

Variation in the association between community treatment orders and patient outcomes (aim 3)

The third aim of the study was to explore variation in the association between the use of CTOs and patient outcomes. We did this by estimating fully random MMs in which the effect for treatment status was not ‘fixed’ (i.e. the same) across all provider trusts but was ‘random’ (i.e. allowed to vary) at provider trust-level. Owing to computing and data restrictions, we only estimated these models for one patient outcome, total time in admission.

There was no evidence of significant variation between provider trusts in the effect of treatment status on total time in admission. Our results did, however, confirm the higher daily rate of admission for CTO patients than non-CTO patients.

Association between community treatment orders and health-care costs (aim 4)

Health-care costs for CTO patients were estimated to be greater than for non-CTO patients. For the most part, this was due to the longer period of time CTO patients spent in admission compared with non-CTO patients, although ‘one off’ administrative burdens associated with CTOs were also a factor. When both were taken into account, the estimated cost per year was £12,470.88 (95% CI £12,405.49 to £12,536.28) for CTO patients compared with £8272.51 (95% CI £8233.53 to £8311.48) for non-CTO patients. Thus, the mean difference was £2669.03 (95% CI £2592.89 to £2745.15). Although there were increased costs associated with CTO patients, there was a correspondingly lower rate of mortality than in non-CTO patients. Therefore, there is a possibility that CTOs could be cost-effective using conventional methods of economic evaluation that consider both cost and impact on lifetime health-related quality of life. However, we were unable to formally assess this owing to data limitations.