Acute day units in non-residential settings for people in mental health crisis: the AD-CARE mixed-methods study

Reflection 1: my involvement in the AD-CARE research project

After I was appointed, I was immediately included in the very first meeting at the university to discuss the study. This was both exciting for me and formative for the study because I was able to add my penny’s worth to the perception of ADUs by service users. It was clear to me that the service user voice was very much part of the study. From the outset I felt included at every stage of the process and David Osborn, the team lead, ensured service users understood the jargon and the issues. We were also asked after every point made by other team members what our views were on the matter. Clearly the fact that ADUs were part of the treatment on offer for service users was a key reason for the study.

As part of our active involvement we wrote reflections after each session. After the first meeting I reflected that community day centres and OT-run [occupational therapy-run] day care at the hospitals were part of my recovery, but had been closed. There had been an overall reduction in recovery possibilities and clinical support for the most vulnerable. Service users generally did not have a voice in changes or initiatives. Thus, studying ADUs in the midst of these closures and austerity took on an additional and more important role of preserving records whilst the service was still in place.

It was suggested by our McPin PPI lead, Vanessa Pinfold, that we all go to visit an ADU. The team met up by arrangement at [XXXX] in [XXXX] and we spent several hours discussing with staff what was on offer. This vibrant centre gave us a positive feel for ADUs. Ironically, before the study was completed this ADU was closed. However, the insights we gained there showed what an important role ADUs could play in mental health treatment and we could use it as a benchmark to the others we visited and studied.

Key tasks

The key tasks of the PPI group in terms of advisory input were as follows:

• In project management meetings, scrutinising progress, being curious about emerging findings and helping with problem-solving.

• Lived experience advisory group meetings – small sessions with only the lived experience advisory group present alongside programme manager. One session included training in coding qualitative data. In lived experience advisory group meetings, we helped develop interview guides, kept up to date with study progress and planned dissemination activities.

• Visiting acute day units, observing the environment and the activities provided and reflecting on these personally. We recommended that photographs were taken of the units to help get a sense of the space and environment.

One of the challenges for any research study is keeping the PPI group informed and engaged over several years. Some members are not able to attend all arranged meetings, and when the meetings are infrequent this can lead members to feel disconnected and lost.

Reflection 2: the challenges of being involved in the AD-CARE research project from a distance

I am a former mental health service user, and in the past I have used an acute day care hospital in one of the trusts that AD-CARE has been doing its research. I have been to most of the meetings; two I attended via telephone. One was a meeting where the team were visiting the acute day care hospital in the trust that covers the area that I live in. However, I do not have access to my own transport and I had already committed myself to other work each morning. It would have been quite stressful and pressured to get there, so the decision was taken that I could ring in after the group had their tour. I was grateful that there was the option to be part of this meeting via teleconference. However, I would have to say that the experience was quite stressful and difficult. It left me quite frustrated.

Towards the end of the project, logistics meant that a face-to-face PPI meeting took place in London a week after I had had a major operation. I had at first been upset at the timing of this as I live over 100 miles away from London. However, I was told that the team could ring me and I could contribute that way, which was a great relief. The documents were e-mailed to me the day before and I was able to prepare for the meeting as I usually would have. This experience was so much better than the previous one. I was able to hear what was being said and a big effort was being made to involve me.

Over the last few years the lived experience advisory panel have been invited to the research team meetings for an hour. Myself and another member have telephoned into these. It has been good to be part of them but there have been issues with hearing what was going on, not really knowing some of the people there, not being able to hear very well, making it contributing hard and frustrating; quite a difficult experience. On one occasion I rang off quite upset. I was later offered support by the PPI lead but these calls are examples of where technology has been a mixed blessing.

Inclusivity

Overall, we sought to employ strategies that were inclusive but also tailored to individual needs. We recognised that the involvement of all PPI members in project update meetings was useful but, as these meetings were short and fast-paced and included often quite technical language, support was required for some members to feel able to contribute. An important innovation in this study was PPI meetings at ADU sites, which helped all of the team more fully contextualise the ‘topic’ of study. We also recommended the photographs were taken of the sites, and this provided an important record to share with other team members.

Reflection 3: visiting units

It was interesting to visit an ADU. In all, I visited three units and it brought the study alive: meeting staff, seeing the activities on offer, meeting a few service users, looking at the physical environment of the unit including artwork and positive recovery messages.

The reason I got involved in the ADU study is because I wanted to see how ADUs provided alternative crisis support for people. I felt strongly uncomfortable with some units being attached to acute inpatient hospital. This can create a feeling of being associated with being sectioned, as some patients were probably compulsory sectioned on the ward. I found that the locations did vary – stand-alone units, co-location on hospital site, some in cities with good transport links, others where clients have to rely on hospital transport to attend. Inside, these units also varied. One had a joint staff service user communal area; in another the staff office space was separate. There were units with clear disability access, including wide corridors, and others were in old repurposed buildings, sharing rooms with other services. I found a strong sense of the creative in the units, maybe too much emphasis on artwork for my liking in one [see Figure 1 ]. I found it was really important part of our work, visiting the units. I had strong reactions and it helped me reflect on my current views. I had last seen one in 2008 – I needed to update my knowledge, and I feel I have.

FIGURE 1.

Photographs of artwork displayed in one of the participating ADUs.

Reflection 4: challenge of working within academic environment

Whilst overall my experience has been a positive one, an area of unforeseen difficulty has been where I belong within the teams. I am employed by the McPin Foundation as a sessional peer researcher but have worked predominantly within the UCL research team. My experience has shifted in this time; at the start of the project I worked closely with the full-time peer researcher and whilst we were doing interviews it was easy to maintain good contact. When we shifted to the analysis this was not so easy as I worked remotely. Then the full-time peer researcher left to do other work and I also joined other projects – maintaining contact and ‘in’ the research became more difficult. I also felt a bit cut adrift because I wasn’t quite sure which team I should identify with. The impact of this was that I did not have a sense of belonging to any specific team. This may have been mitigated by having more clarity around my role in the beginning. This was a new way of working and was an unintended consequence; with reflection, an overview of the team structure, demarcation of roles and more regular meetings, perhaps by Zoom or Skype if this was not practical, would be helpful adjustments to make.

Reflection 5: peer researcher significant moments

Right from the moment of interview I have felt that my skills as a researcher as well as my lived experience have been equally valued. I was very excited to be able to put both sets of skills to good use and I have not been disappointed. At no point have I felt like the proverbial ‘token presence’. When I compare my experience of using my lived experience of mental health difficulties within care settings with my experience within a research setting – I would suggest that research teams are way ahead of the game. I wonder if this is because the ethos of social science is to value the individual’s experience as important data for research and the stringent ethical considerations that go alongside this. Perhaps an area for future research? I have felt privileged to be interviewing my peers and the opportunity (and responsibility) of attempting to represent their voice accurately in this research. There is something unique about the relationship a peer researcher can offer an interviewee, which can enhance the data that is gathered; the knowing nod of a shared understanding really does go a long way. As I have visited ADUs, and hearing the vast array of positive experiences from my peers, I am struck once again by (1) the need for an approach to mental health care that has ‘relationships’ at its centre; and (2) the need for those with lived experience of mental health illness to be the ones asking the questions. This has been a valuable experience and one where I have been able to develop my skills as a researcher, and to have such an investment in the whole process has been a great opportunity.

Design

This was a mapping exercise to identify existing ADUs for mental health in England, followed by a questionnaire survey of ADU managers.

Objectives

• To produce a mapping of existing ADUs for mental health in England.

• To collect data via a cross-sectional questionnaire on the characteristics and service functioning of ADUs.

• To ascertain whether or not there are distinct ADU models, and, if so, to characterise them.

Setting and participants

All NHS mental health trusts in England were contacted to ascertain whether or not they provided an ADU service. In addition, voluntary sector non-residential crisis day services were sought. NHS trusts that provided at least one ADU, and any relevant voluntary sector services, were included in the study.

ADUs were defined for this survey as non-residential services offering daytime treatment and care to adults experiencing a mental health crisis who would otherwise be considered for acute psychiatric hospital admission, or other alternatives to admission (including crisis resolution services). Services were excluded that:

• provide rehabilitation, rather than acute care

• work only with groups of service users who would not be considered for acute psychiatric hospital admission

• work primarily with populations other than people with mental illness, such as older adults or people with dementia, learning difficulties or primary drug or alcohol dependence disorders

• do not accept referrals for service users currently living at home (i.e. exclusively provide ‘step-down’ care from hospital)

• routinely work with service users for longer than 3 months (i.e. longer-term support rather than acute care)

• do not accept referrals from the local CRTs.

Non-NHS services that met the above criteria were included in the survey. Day services providing dedicated acute care meeting the above criteria within a broader service were also included.

Measures

We used a survey developed for the AD-CARE study (the full survey is available in Report Supplementary Material 1). The survey was provided online via the secure UCL Opinio website, and could be completed over the telephone with a researcher. The survey included questions covering the following areas: location and access, throughput and service user characteristics, integration with other services, service organisation and staffing, interventions provided, service user experience, service development, and outcomes data.

Procedure

All 58 NHS mental health trusts in England were contacted (August 2016) to enable ADUs to be located. The following methods of contact were used:

• All England NHS mental health trust websites were screened.

• Local communications teams, patient advice and liaison service teams, research and development teams, trust headquarters, local acute care leads or other appropriate clinical staff were contacted by telephone and e-mail.

• Relevant professional organisations and networks (e.g. the Royal College of Psychiatrists’ Acute Care Network and Mind’s Acute Care Campaign) were contacted on Twitter (Twitter, Inc., San Francisco, CA, USA; www.twitter.com) and by e-mail and telephone.

• The CRT managers of all teams that had been identified as having an ADU in the 2012 CRT Optimisation and Relapse Prevention (CORE) study survey8 were contacted.

• Online searches were conducted for any voluntary sector services that met the inclusion criteria.

Where ADUs were identified, study researchers contacted the service managers by telephone to explain the survey, answer any questions and obtain e-mail/postal addresses to enable them to send out information sheets.

Managers (or alternative ADU clinicians with appropriate knowledge and experience) were asked to complete a questionnaire either online or over the telephone with a researcher. Participants were assigned a unique, anonymised study ID. All data were entered into the questionnaire in Opinio and then extracted into Microsoft Excel^® version 16 (Microsoft Corporation, Redmond, WA, USA) and SPSS version 24 (IBM Corporation, Armonk, NY, USA) for data analysis. Data collection was carried out from September to November 2016.

Those who did not respond were followed up by study researchers by telephone and e-mail. Any service that declined to complete the survey was not contacted further.

A brief follow-up survey was conducted 1 year after initial data collection (October 2017) to ascertain whether any ADUs had opened or closed. Services identified in the original mapping exercise were contacted by telephone and e-mail to check that they were still operating and to identify any changes to services.

Analysis

As outlined above, the survey had two main aims: (1) to establish a typology of ADU models and (2) to describe current practice in ADUs.

A cluster analysis was carried out to address aim 1. A cluster analysis is a way of grouping units so that those units more similar to each other appear in the same cluster. The process aims to minimise variability within clusters and maximise variability between clusters. There were four stages in this process:

1. Potential grouping variables were identified. These were collated from the questions in the survey, with some grouping variables obtained by the amalgamation of multiple survey questions covering the same topic.

2. The expert working group ranked the list of potential grouping variables, ordering them by most to least important in distinguishing different types of ADUs.

3. The five highest-ranked grouping variables were included in a cluster analysis (where a grouping variable was considered to have poor-quality data available from the survey, it was discarded, and the next highest-ranking variable was used instead). Five is considered an appropriate number of grouping variables to include in this type of analysis.

4. The cluster analysis was refined, with different models run to establish the most appropriate number and composition of groups.

The resulting variables were used in a cluster analysis using SPSS software.

To address aim 2, descriptive data were collated for each survey question, including range, mean and median scores.

Cluster analysis

Eight members of a multidisciplinary expert working group ranked 14 variables in the survey identified as relevant to ADU type (Table 1). Variables were ranked by participants, using 1 to indicate the most important and 14 to indicate the least important, meaning that those variables with the lowest total scores were considered (by consensus) to be the most important in distinguishing services from each other.

Two-step cluster analysis was used to enable the inclusion of categorical and continuous variables, and for the automatic determination of the optimal number of clusters.

Prevalence of acute day unit services

Forty-five individual ADU services meeting our criteria were identified across England. Of these 45 ADUs, 27 (60%) were provided solely by NHS trusts (17 trusts, 29% of the 58 mental health trusts in England), eight were jointly provided NHS and voluntary sector services (17%) and 10 were voluntary sector services (23%).

The geographical locations of the services identified are shown in Figure 2.

FIGURE 2.

Map of UK ADU services. Map data © 2019 Google (Google Inc., Mountain View, CA, USA).

Map data ©2019 Google

In total, 37 of the 45 identified ADU services completed the initial 2016 survey (two declined and six did not respond to multiple requests for information), representing a response rate of 82%. The two ADUs that declined were from the same trust, but the six that did not respond were from different trusts and voluntary organisations. Twenty-two NHS services and 15 joint or voluntary services responded to the survey.

Results are reported separately for NHS services (referred to as ‘NHS-ADUs’) and for joint and voluntary services (referred to as joint/voluntary). As not every survey respondent answered every question, the denominator is given when reporting each result.

Main findings

The mapping exercise, which identified 45 ADUs in England, demonstrates that ADUs are not an established part of mental health service provision in most areas. The cluster analysis found evidence of two types of service model: (1) NHS services (n = 27) and (2) voluntary sector services (including jointly run NHS and voluntary sector services) (n = 17). Considering the geographical distribution of services (see Figure 2), it is clear that large parts of the population have no access to any kind of acute day service as defined by this survey. Although the evidence base for ADUs is small, there have been positive findings in previous studies (i.e. greater symptom improvement and service user satisfaction than for inpatient wards13), so it is surprising that ADUs are not more widespread.

The difference between the NHS and joint/voluntary services is quite marked. NHS-ADUs are typically available 10 a.m.–4 p.m. on weekdays, with a wide range of interventions (including medication review/prescription/support, physical health, psychological therapies, and help with daily living activities), a multidisciplinary team including clinically qualified professionals, and service users attending for an average of 5 weeks. By contrast, joint/voluntary services tend to consist of supportive staff working in non-clinical capacities, who provide brief, one-off listening and signposting support to those in immediate crisis, often in the evening and the early hours of the morning. NHS-ADUs have less service user/carer involvement in paid roles, management, recruitment and training than the joint/voluntary services. In this regard, NHS-ADUs appear to involve service users and carers at similar levels to CRTs. 8 Although the practical offerings of the two types of service are quite different, the explicitly stated purpose of a large majority of both types is an alternative to inpatient admission. The joint/voluntary services are more often intended as an alternative to A&E, which may explain the difference in daily support offered. There are certainly opportunities for cross-sectoral learning here, and in particular the joint NHS/voluntary sector services could lead on sharing best practice between these different types of service.

It is notable that there are currently no national (or international) standards for how ADUs should be set up or function, and this perhaps explains the variation evident, for example, in the wide range of interventions offered. Unlike for CRTs, early intervention services and assertive outreach teams, no guidance was given in The Mental Health Policy Implementation Guide15 about the composition of NHS-ADUs, resulting in a certain amount of heterogeneity, and no standards or criteria are given by which to assess service functioning. Guidance on the place of ADUs within the acute care pathway is similarly lacking.

The findings of this study are in line with previous research about ADUs, both in England and internationally. For example, a previous survey of psychiatric day hospitals in England found heterogeneity of service provision,16 as did a survey of day hospitals for general psychiatric patients in Germany, England, Poland, Slovakia and Czechia,9 although both studies found that the majority of services aimed to provide an alternative to inpatient admission, similar to the current survey. One aspect we investigated in this survey, namely the involvement of service users and carers in the management and running of ADUs, is lacking in previous research, and there is little indication from international studies that this issue has been addressed elsewhere. It is also unclear from international research whether or not ADUs are provided by the voluntary sector in other countries, as this survey demonstrates they are in England.

NHS-ADUs and CRTs are similar in that both offer a range of interventions, delivered by multidisciplinary teams, as an alternative to admission. The key differences between the services are the location and the timing of contact. Because service users attend a single location during office hours, ADUs are able to offer a wider range of interventions, consistency in terms of the staff service users see, more contact time, and peer support. In comparison, by providing home visits and working shifts, CRT contact time is brief, there is little consistency in which staff member sees which service user, and there is no opportunity for peer support (all of which are well-documented complaints of CRT users17). Although CRTs offer more flexibility in timing and location of care, and the opportunity for the clinical team to observe a service user’s home environment (for people for whom loneliness, isolation and lack of activity are a problem, or whose home environment is problematic), ADU care potentially has added benefits than CRT-only care.

In addition to the differences between NHS-ADUs and CRTs, the two ADU models identified by this survey (NHS and joint/voluntary services) indicate further complexity in the acute care pathway. The different offerings of NHS and joint/voluntary services may explain the geographical overlap evident in Figure 2, with joint/voluntary services ‘filling the gaps’ that NHS-ADUs and CRTs do not cover by providing drop-in services out of office hours. Research into how NHS and voluntary sector services complement each other and work together is currently lacking, although a programme of work is under way to gain insight into this important area. 10

The follow-up survey suggests that NHS-ADU services occupy a precarious position. The closure of five NHS-ADUs in a relatively short time is striking. It implies an unstable environment in which non-mandated services may be seen as easily disposable when there is pressure on resources, despite research evidence suggesting that they can be effective. 12,13 At the same time, the piloting and planned opening of seven new NHS-ADUs in one trust suggests that the value of such units is recognised by some commissioners, which reflects the importance of providing choices for people in crisis. 14

Strengths and limitations

There are two main strengths of this survey. The first is the high response rate (81%), meaning that we can take the results to be broadly representative of existing ADUs in England. The second is the inclusion of all services, whether provided by NHS or by voluntary sector services, which gives a comprehensive picture of what is available, and where.

There are three key limitations. The first is that because ADUs are not mandated services, lacking a definitive name or model, identifying such services was challenging. Although we used a clear and specific definition of the type of team we were interested in, it was frequently the case that one part of a trust would identify no such teams, and then another source within the trust would identify a service that clearly met our inclusion criteria. For this reason, and despite the multiple avenues we used to identify teams, it is possible that there are more ADUs in the country than were identified by this survey.

A second limitation was that, because we found that teams close and open relatively quickly, accurately identifying the number of these services in the country at any one time is challenging.

The third limitation regards the quality of the data obtained in the survey. Many teams did not answer all of the survey questions. In the case of joint/voluntary services this was often because the questions were not relevant, or, as with questions about ethnicity and sexual orientation, because they did not keep records of these variables, but even among the NHS-ADUs some data were missing. The aim was for the survey to be as comprehensive as possible while remaining feasible for busy clinicians to complete, but perhaps a shorter survey would have encouraged a higher completion rate. There is the possibility of social desirability bias from this self-report survey, and that respondents interpreted questions in different ways.

Research implications

The results of this survey demonstrate the need for further research into these services. Although some previous research has compared outcomes for people using ADUs with those for people using inpatient wards,12 there is little evidence comparing ADUs with other non-residential services. The finding12 that ADUs are as effective as inpatient wards is promising, but it would be helpful to investigate the place and effectiveness of ADUs in the wider acute care context. Understanding how the ADU complements other crisis and community provision by increasing the support options available is vital. There is a lack of research considering the acute crisis care system as a whole, and how the range of services available can work together to meet the needs of different people. Investigation of the service user and carer experience of ADUs is also, as far as we can find, entirely lacking, and this is particularly important to rectify. Although this survey was focused on ADUs in England, this is an issue of international relevance, and so comparison with services in other countries would be helpful.

Given that CRTs are widely available as the standard service for non-residential crisis care, it is important to know whether or not ADU provision enhances outcomes for people using acute services. However, the lack of model specification for ADUs and the resulting heterogeneity of services means that in any such research, similar types of service should be considered. Research into the different models of ADU care available, and their relative merits in terms of service user outcomes and experiences, would be beneficial, as would a thorough economic evaluation of the costs and benefits of ADUs in comparison with other acute services. The current Acute Day Units as Crisis Alternatives to Residential Care (AD-CARE) study aims to address these issues.

Implications for policy and practice

A detailed health economic analysis of ADUs would be highly useful for policy-makers and service planners, particularly given the current economic and political climate in the UK. Such an analysis would provide vital information about the best ways to configure services, given the economic pressures that NHS trusts and wider communities find themselves under.

This survey suggests that, on average, around 1215 people use NHS-ADUs or voluntary/joint services per day in England. Putting this in context, as of 2017 there were 18,730 mental health inpatient beds in England. Taking the conservative Marshall12 estimate of the proportion of inpatients suitable for ADUs [23.2%, 95% confidence interval (CI) 21.2% to 25.2%], this suggests that, potentially, approximately 3130 additional service users per day could benefit from ADU care. Given the known pressures on beds, frequent out-of-area placements, and the inherent desirability of offering choice regarding acute care, commissioners and policy-makers should consider the place of ADUs in the acute care pathway. The development of a national policy and the implementation of a standard ADU model would mean that these services would be less vulnerable to closure during economically challenging periods.

For existing NHS-ADUs, it may be worth considering further how former and current service users and carers can contribute to services, and the ways in which voluntary sector ADUs manage this could be of interest to NHS-ADUs. Greater sharing of best practice between services would certainly be desirable, as the heterogeneity of services suggests that this is currently not a regular occurrence.

Design

The design was a cohort study of ADU users and CRT users, comparing re-admission rates with those of the acute care pathway during a 6-month period, and satisfaction with services. Five sites with ADUs and CRTs were identified and recruited from the national survey carried out in WP 1, although one site dropped out of the study shortly after starting recruitment, leaving us with four study sites. Recruitment began in March 2017 and was completed by the end of March 2019, with follow-up completed by the end of September 2019. We invited people who were consecutively admitted to each ADU to participate in baseline interviews and then in telephone or online follow-up 8–12 weeks after baseline, with electronic health records (EHRs) outcome data collected at 6 months.

We also invited people from the trusts who used CRTs to participate in the parallel non-ADU cohort.

Objectives

1. To describe the clinical and sociodemographic characteristics of people who use each of the ADUs selected for the in-depth case studies.

2. To determine their pathways into the ADU, length of stay, treatments received, experience, empowerment, loneliness and re-admissions.

3. To compare these characteristics with those of a cohort of people who receive acute CRT care in the same locality but without ADU input.

Setting

Data were collected from ADUs and CRTs in four NHS mental health trusts in England. We investigated the possibility of including a voluntary/joint voluntary–NHS ADU service in the study, but no such units were able to provide the data required to enable a comparison with a local CRT service. Therefore, the study focused on NHS ADUs, as identified by the mapping and survey work of WP 1.

Sample size

We calculated the sample size to detect a 12–13% absolute reduction in the main outcome of re-admissions to the acute pathway at 6 months after baseline (using admission figures from previous research). Our CORE programme grant in crisis teams was powered at 80% to detect a 15% difference between trial arms (50% vs. 35%). Data from London crisis services suggest that baseline re-admission rates could be lower, at 40%. We explored various sample size calculations, including different assumptions regarding this baseline re-admission rate. These showed that 310 people in each arm would afford 90% power to detect differences such as 50% compared with 36.8%, 45% compared with 32.0% or 40% compared with 27.4%. Inflating for a design effect by 30% to accommodate the clustered study design required 400 per arm. These numbers also afforded > 90% power to detect an effect size or difference of 0.3 standard deviations in the client experience measure the CSQ-8 [crisis team mean CSQ 25, standard deviation (SD) 6]. Therefore, we required 400 ADU participants and 400 CRT participants (total n = 800).

Participants

Measures

Data were collected from participants using a questionnaire (provided in hard copy or online) at baseline, and another questionnaire 8–12 weeks later, and baseline and 6-month data were collected from EHRs by researchers. The validated measures used in the questionnaire are listed in the next few sections. A full list of the variables included in data collection at each time point is available in Table 12.

The primary outcome was re-admission for acute treatment during the 6-month study period, which was collected from service use data in the EHRs. After discharge from the service used at baseline, any subsequent use of acute mental health services during the 6-month study period was recorded, along with the duration of each admission to services and any detention under the Mental Health Act (MHA).

The full baseline questionnaire is available in Report Supplementary Material 2, the full follow-up questionnaire is available in Report Supplementary Material 3 and the full EHR data collection schedule is available in Report Supplementary Material 4.

Procedure

At baseline, ADU/CRT staff screened all service users consecutively admitted to their service from the start date. All service users who met the inclusion criteria were approached by ADU/CRT staff/trust-employed clinical studies officers/clinical researchers and asked if they were willing to discuss participation further (except at sites where service users had already given consent to be contacted directly about research projects, in which case researchers contacted service users directly once their eligibility and any risk-related safety requirements had been established). The researchers kept a record of potential participants to be contacted and the date and the name of the clinician with whom this had been agreed. Researchers noted the service user’s agreement to being contacted by a researcher in their EHRs. Those who agreed were contacted by a researcher with an information sheet and offer to answer any questions. Potential participants were given at least 24 hours to consider whether or not they wanted to take part, and if they were still interested their consent was taken by a researcher, who also collected the baseline data. Consent and data collection could occur up to 14 days post discharge.

Participants were offered reimbursement of £20 (in vouchers) for taking part (£10 for the baseline interview and £10 for the follow-up interview 8–12 weeks after baseline). Participants were telephoned by a researcher 8–12 weeks after baseline for the collection of follow-up data. At 6 months after baseline, participant data were collected from EHRs and clinical notes.

Although many people find talking about their experiences to be helpful, some found that completing the questionnaire brought up issues that caused emotional distress. Where this happened, the researcher provided immediate emotional support to the participant, offered to pause or postpone the interview, and, if the participant asked, contacted a person of the participant’s choice (e.g. current care co-ordinator, carer, friend, family member, colleague) on their behalf.

If service users or carers reported any untoward feedback about the service used, the researcher conducting the interview asked the participant whether or not they wanted the researcher to pass this on to the service or other relevant person. If the untoward feedback was of a nature that led the researcher to be concerned for the safety of others, the participant was informed that the feedback would be passed on to the relevant service or person to be addressed, but that, if desired and if possible, they would remain anonymous.

Analysis

We derived descriptive statistics comparing the baseline characteristics of the CRT cohort and ADU cohort for the sample as a whole, and within sites. We explored baseline differences in demographics, clinical data and content of care using parametric and non-parametric tests, as appropriate.

For the primary outcome (time to acute admission during the 6-month study period) we compared the time to re-admission in the CRT cohort compared with the ADU cohort using Cox regression and adjusting for trust, age, sex, diagnosis [severe mental illness (SMI) or not], employment, Health of the Nation Outcome Scales (HoNOS) score, SWEMWBS score and whether or not the person had previously been an inpatient. Covariates were added using a stepwise procedure, starting with a univariate model with trust as a fixed effect, and then adding age, sex and employment, followed by diagnosis, followed by HoNOS score, followed by SWEMWBS score, followed by previous inpatient. We tested for any interaction between team (ADU/CRT) and trust as a sensitivity analysis.

In addition, we analysed a number of secondary outcomes:

1. We compared satisfaction mean CSQ-8 scores at the 8- to 12-week time point in the two cohorts using linear regression, using trust as a fixed effect, and adjusting for important covariates (as above). We tested for effect modification by individual ADU. We did not make head-to-head comparisons of outcomes between the different ADUs.

2. We compared mean SWEMWBS scores at the 8- to 12-week time point in the two cohorts using linear regression, as above.

3. We compared mean CES-D-10 depression scores at the 8- to 12-week time point in the two cohorts using linear regression, as above.

4. We compared the duration of admissions in the CRT cohort compared with the ADU cohort using Poisson regression, adjusting for the same covariates as above.

Cohort descriptives

Primary outcome: re-admission

The unadjusted model with trust as a fixed effect produced a hazard ratio (HR) for re-admission in the ADU group of 0.94 (95% CI 0.67 to 1.32). Adjusting for age, sex and employment in model 2 reduced the HR to 0.90 (95% CI 0.64 to 1.28). In model 3 the remaining covariates were added (SMI diagnosis, whether or not participant had previously been an inpatient, baseline HoNOS score and baseline SWEMWBS score). The addition of these covariates reduced the fully adjusted HR in model 3 to 0.78 (95% CI 0.54 to 1.14). None of the models showed statistically significant differences between the ADU and CRT groups.

As only three participants identified as transgender, they were included in the analysis under the sex category indicated by the name they provided. A sensitivity analysis was run excluding these three participants, which showed no significant changes in the model (model 4).

We then tested for any interaction between team (ADU or CRT) and trust in model 5, and found evidence of an interaction between ADU/CRT and trust and the primary outcome of re-admission. Model 5 showed that in trust 1 the risk of re-admission was statistically significantly lower for ADU participants than for CRT participants (0.46, 95% CI 0.25 to 0.83; p < 0.01). In trust 2 there was no statistically significant difference (0.74, 95% CI 0.39 to 1.42). In trust 3 the risk of admission was statistically significantly higher for ADU participants than for CRT participants (9.11, 95% CI 1.22 to 68.09; p = 0.03). In trust 4 there was no statistically significant difference between ADU and CRT participants (0.83, 95% CI 0.27 to 2.54).

Full results are shown in Table 15.

Secondary outcomes

We compared service user satisfaction between CRT and ADU participants, using the CSQ-8 measured during the 8- to 12-week follow-up period, and found statistically significantly higher scores (indicating higher satisfaction) in ADU participants than in CRT participants (Table 16). The unadjusted standardised mean difference in satisfaction on the CSQ was 0.39. On average, and taking account of trust, age, sex, employment, SMI diagnosis and previous inpatient use, the mean CSQ score among ADU participants was 2.27 (95% CI 1.24 to 3.30) points higher than that of CRT participants (p < 0.001).

Acute day unit participants also showed statistically significantly higher levels of well-being at the 8- to 12-week follow-up point than CRT participants. ADU participants had, on average, and taking into account the same covariates as above, a SWEMWBS score 1.38 (95% CI 0.58 to 2.17) points higher than that of CRT participants (p = 0.001).

Acute day unit participants also had statistically significantly lower levels of depression at the 8- to 12-week follow-up point than CRT participants (see Table 16). ADU participants had, on average, and taking into account the same covariates as above, CES-D-10 depression scores –1.7 (95% CI –2.6 to –0.7) points lower than those of CRT participants (p < 0.001).

Summary of results

This WP aimed to compare the outcomes, in terms of re-admission to acute mental health care, satisfaction and well-being, in two cohorts: those using ADUs and those using CRTs. Having recruited 431 ADU participants and 312 CRT participants (n = 744), we found important sociodemographic and clinical differences between the groups. ADU participants were more likely to be unemployed, have an SMI diagnosis and have previously been admitted to an inpatient ward and have higher HoNOS scores (indicating more severe symptoms and poorer social functioning). These findings suggest that those served by ADUs are more unwell than those using CRTs.

Even taking into account these differences between the groups, there was no significant difference between them in terms of re-admission (HR 0.94, 95% CI 0.67 to 1.32). At 8–12 weeks, after accounting for baseline differences in participant characteristics, ADU participants had significantly higher satisfaction [CSQ score: 26.66 (SD 5.04) vs. 24.37 (SD 6.57)], with better well-being [SWEMWBS score: 20.51 (SD 4.96) vs. 19.02 (SD 5.03)] and lower depression [CES-D-10 14.4 (SD 6.0) vs. 16.6 (SD 5.7)] than CRT participants. These results indicate that, despite serving a more unwell client group, ADUs produce comparable outcomes in terms of re-admission, and better satisfaction, depression and well-being results than CRTs. The effect sizes for the difference in satisfaction appeared clinically meaningful, as did the differences in depression and well-being scores, even adjusting for baseline difference between ADU and CRT participants (see Tables 14 and 16).

Previous research

Although no previous research has directly compared the outcomes for ADU and CRT participants, there is evidence comparing ADUs with inpatient wards, and evidence about satisfaction and re-admissions in CRTs. In a randomised controlled trial comparing day hospitals with inpatient care, Priebe et al. 13 measured satisfaction using the Client’s Assessment of Treatment Scale (CAT)27 instead of the CSQ, but, similar to this study, they found significantly higher satisfaction among the day-hospital group than among inpatients at discharge and at 3 months (although no difference was found at 12 months). The CSQ has been used in several other studies of acute mental health services; for example, in a cohort study of people using standard acute inpatient wards compared with those using alternative residential crisis services (e.g. crisis houses), those using alternative services had significantly higher satisfaction scores [26.4 (SD 4.9) vs. 23.19 (SD 6.6)]. 28 These scores are comparable to those found in the current study, with greater satisfaction among those using ADUs and crisis houses than among those using CRTs and inpatient wards.

In terms of re-admission, the most recent meta-analysis of research on ADUs12 found no difference in re-admission rates between ADU and inpatient services. However, this meta-analysis did find some evidence (albeit of low quality) that those using ADUs had better outcomes than inpatients in terms of subsequent employment. This is particularly interesting given our finding that ADU participants in this cohort had lower rates of employment than CRT participants. Werbeloff et al. 29 investigated the re-admission of CRT service users over 1 year and found that having a psychotic disorder increased the risk of re-admission. This study found HoNOS scores in two CRT cohorts that were similar to the scores of our CRT participants [13.3 (SD 6.4) vs. 11.8 (SD 5.5)].

Strengths and limitations

This cohort study has four key strengths: (1) the direct comparison of those using ADUs and CRTs, which to our knowledge has not been undertaken previously; (2) the relatively large sample size; (3) the range of geographic locations of the participating services; and (4) the high follow-up rates. As discussed above, there is a lack of evidence about ADUs, the majority of work having been carried out some time ago and on services that were substantially different from those available currently. This study provides a comprehensive overview of the sociodemographic and clinical characteristics of those using ADUs and CRTs, and offers insights into the comparative benefits of each type of service in terms of outcomes. To our knowledge, this is the largest cohort study of ADU participants to date, which lends weight to these results. The participating services were in areas serving urban, suburban and rural communities, with high and low deprivation levels. We were able to follow up 99.99% of participants via their EHRs at the 6-month time point, which included our primary outcome of service use during the 6 months, and 76% of participants at the 8- to 12-week time point, which included the secondary outcomes of satisfaction, well-being and depression.

The study has three main limitations: (1) recruitment bias, (2) lack of randomisation of participants and (3) EHR data quality. As is typical in cohort studies of this nature, we were reliant on people who were attending participating services being willing and able to take part. The nature and severity of some service users’ illness excluded them from the study, and as a result our findings may not be applicable to the most unwell users of these services. However, the sample who were recruited from both ADUs and CRTs included those with a variety of diagnoses and people who had previously been admitted, and in both groups substantial use of mental health services was reported at follow-up. Furthermore, many participants were single and unemployed and had diagnoses that included psychosis and personality disorder. This indicates that our sample were indeed people who had complex mental health needs.

We recruited just under 750 people to the study, having originally aimed for a target of 800, which would have given 90% statistical power. It seems unlikely that this slight under-recruitment would have affected our main findings. The adjusted HR for re-admissions was 0.78 (95% CI 0.54 to 1.14), which is not suggestive of any difference between ADU users and CRT users, even if a larger sample had been recruited. The differences in satisfaction, well-being and depression, each of which was in favour of ADUs, were all statistically significant, with convincing effect sizes.

The fact that participants were not randomised could be seen as a limitation; however, we were able to recruit a larger sample size than would have been possible in a randomised controlled trial, and in the analysis we adjusted for baseline differences between the two groups, including history of admission. We collected a large number of data from routinely entered EHRs, and so we were reliant on these records being accurate and up to date. Missing data were evident for most variables, and a number of errors were identified in the data provided to us.

These cohort results are discussed further in the context of findings from other WPs in Chapter 10.

General principles of the economics analysis

This analysis adopts a health-care perspective, focusing on acute mental health care. To examine the costs associated with acute mental health care for those accessing ADUs (the ADU group) compared with the costs for those not accessing ADUs (the CRT group), it was necessary to collect information on acute mental health care resource use and mental health outcomes. The time horizon was 6 months and, consequently, the discounting of costs and outcomes was not necessary. This analysis uses the price year of 2017 and adopts an intention-to-treat principle, whereby the groups at baseline were analysed regardless of whether or not they transferred between groups during the 6 months. In this analysis, statistical significance is determined by a p-value of < 0.05.

Identifying resource use and costs

The first step in the analysis of acute mental health costs was to identify the different types of mental health care that were of interest. As in the primary statistical analysis, these were refined to:

• acute mental health ward

• ADU

• CRT

• crisis house.

To examine the costs to the two groups, it was necessary to capture the number of days in contact with each of the aforementioned services. To do this, the details of acute mental health contacts were collected from EHRs for the 6 months from baseline. To calculate the costs of acute mental health care, resource use data were combined with unit cost data. The unit costs for acute mental health wards and CRTs were collected from Personal Social Services Research Unit data,30,31 and the unit costs for crisis houses were estimated from published case studies. 32,33 Given that no existing unit costs were found in the literature for ADUs, it was necessary to estimate a unit cost. To do this, participating ADUs were contacted and information on the annual budget for running the ADUs was requested and combined with information on average caseloads from the WP 1 mapping study. This information was used to calculate a cost per day while in contact with ADU services. Information on acute service resource use was then combined with the unit cost data to calculate the total cost of acute mental health care over the 6-month period. Given the lack of established unit cost data for ADUs and crisis houses, sensitivity analyses were required to examine the impact of changes in unit costs.

Analysis of cost

The analysis first examines the resource use data descriptively. Differences in resource use were examined using standard statistical tests (t-tests for continuous variables and chi-squared test for categorical variables). The total cost end point was then analysed using a series of increasingly complex models, as discussed below, to estimate the differences in costs. In all models, CIs were calculated using non-parametric bootstrapping methods:

• Model 1 – adjustment for clustering. This regression model examined the difference in costs between the two groups including provider trust as a fixed effect.

• Model 2 – partially adjusted model. This model builds on model 1 by including variables that were statistically significantly different at baseline as covariates in addition to provider trust.

• Model 3 – fully adjusted model. This model builds on model 2 by including all the prespecified covariates, which included age, sex, SMI, employment, baseline HoNOS score, trust and previous inpatient admission.

Analysis of outcomes

The HoNOS was used as the primary outcome measure in the cost-effectiveness analysis. The HoNOS is collected routinely within mental health services and is reverse scored, so that the lower the score, the better the mental health state of the individual. The most recently measured HoNOS score in EHRs was recorded at both baseline and 6 months. The differences in HoNOS scores recorded at baseline and follow-up are first presented descriptively, with t-tests used to examine the difference between the two cohorts.

To examine the mental health outcomes for the two groups and to account for baseline differences, changes in HoNOS scores were calculated (HoNOS score at 6 months minus HoNOS score at baseline). Three models were used to examine the impact of the cohort on the change in HoNOS score during the follow-up period. As above, CIs were calculated using non-parametric bootstrapping. The models included were:

• Model 1 – adjustment for clustering. This regression model examined the difference in HoNOS change between the two groups including provider trust as a fixed effect.

• Model 2 – partially adjusted model. This model built on model 1 by including variables that were statistically significantly different at baseline as covariates in addition to provider trust.

• Model 3 – fully adjusted model. This model built on model 2 by including all of the prespecified covariates, which included age, sex, SMI, employment, baseline HoNOS, trust and previous inpatient admission.

Exploratory cost-effectiveness analysis

To assess cost-effectiveness, methods that account for clustering, baseline differences and correlation between costs and outcomes were required. To do this, seemingly unrelated regressions were used to jointly model costs and HoNOS score change while accounting for baseline differences between the two groups by including the covariates from model 3 in the joint regression. Had costs and outcomes fallen into the north-east cost-effectiveness quadrant (i.e. more costly, more effective), the cost per HoNOS point associated with the ADU group would have been calculated and presented. To examine the probability of cost-effectiveness and to characterise uncertainty, a net-benefit regression framework was used to generate cost-effectiveness acceptability curves (CEACs). These were used to examine the probability that ADUs were more cost-effective than CRTs.

Given the lack of preference-based measures in this study, it was possible to conduct only a cost-effectiveness analysis and not a cost–utility analysis. Had ADUs been more effective but more costly than CRTs, we had planned to examine the number of quality-adjusted life-years that the ADU group would need to gain to achieve cost-effectiveness using the upper National Institute for Health and Care Excellence threshold of £30,000 per QALY. 34 The changes in HoNOS score were used to inform the likely qualitative direction of the quality-adjusted life-years gained and the plausibility of this cost-effectiveness interpretation.

Sensitivity analyses

In addition to the unadjusted and partially adjusted models of cost and outcomes, several further sensitivity analyses were performed to examine the sensitivity of results to changes in assumptions:

• Sensitivity analysis 1 – given the lack of established unit costs for ADUs and for crisis houses, it was necessary to examine the impact of significant changes to the unit costs for these two services. Therefore, in the first sensitivity analysis, the cost of both services is increased by 25% so that the sensitivity of the results to these increases in cost can be examined.

• Sensitivity analysis 2 – the second sensitivity analysis replicates the first analysis; however, in this instance the unit costs for ADUs and crisis houses are reduced by 25%.

• Sensitivity analysis 3 – a number of participants in the CRT group went on to access ADU services. The primary analysis adopts an intention-to-treat principle. This sensitivity analysis excludes those who crossed over from CRT to ADU service use.

• Sensitivity analysis 4 – multiple imputation to partially address concerns with follow-up HoNOS data. When it was not clear that a participant’s follow-up HoNOS score data had been updated in their medical records (i.e. the HoNOS data in the medical records were recorded as the same at baseline and at follow-up), these follow-up data points were treated as missing in this sensitivity analysis. Multiple imputation using chained equations with predictive mean matching for continuous variables and logit for binary variables was used to impute missing data. Primary analysis covariates were included as covariates in the imputation, and 35 imputation sets were created. The main cost-effectiveness analysis was then repeated, using multiple imputation estimation procedures in Stata^® 16 (StataCorp, College Station, TX, USA).

Service use

The unadjusted acute mental health care service use data for both groups during the 6-month follow-up period are shown in Table 17. During the follow-up period, the ADU group used slightly more acute mental health ward services (mean 3.78 vs. 2.53 ward days) than the CRT group; this difference was not statistically significant. Unsurprisingly, the CRT group had statistically significantly more days in contact with CRT services than the ADU group (7.15 vs. 3.84 days in contact). There was no significant difference in days in contact with ADU services or crisis house services. The mean ADU use was marginally higher in the CRT group, driven in part by a few heavy users of ADUs. The fact that the heaviest user of ADU services was a participant from the CRT group demonstrates that there was significant contamination between the two groups, highlighting the need to carry out a sensitivity analysis excluding these individuals.

Table 18 presents the same service use information in terms of cost. Across both groups the highest costs were associated with acute mental health ward use (over £1000 per service user in both groups). The ADU group experienced higher acute mental health ward costs over the 6-month period than the CRT group (£1526 vs. £1021). The next highest costs were related to CRT days, with the CRT group accruing statistically significantly higher costs than the ADU group (£851 vs. £457). Crisis houses were the next most costly component of acute mental health care, with the ADU group costing slightly more than the CRT group (£400 vs. £265). Finally, ADU stays were the least costly, with both groups having a relatively low ADU cost per service user.

Analysis of acute mental health costs

The results of the analysis of acute mental health costs are shown in Table 19.

The final fully adjusted model 3 further reduces the cost difference between the two groups to a negligible amount (the mean cost is £37 lower in the ADU group than in the CRT group). Again, whether or not the individual has had a previous inpatient admission is the largest predictor of cost and is highly significant. Those who have SMI have higher costs than those who do not, but this is not statistically significant at the 5% level (p = 0.051). Trust 4 is again associated with lower costs than trust 1 at a statistically significant level. Following adjustment for baseline imbalances, there is no significant difference in the costs accrued by the two groups.

In model 1, NHS trust is included in the linear regression as a fixed effect. In this model, the ADU group is associated with increased costs (£332) compared with the CRT group. As demonstrated by the wide CIs, considerable uncertainty surrounds this mean cost difference. Of note, trust 4 had significantly lower costs than the comparator, trust 1.

Given the extensive baseline imbalance, model 2 adjusts for statistically significant baseline differences. When this is done, the cost difference between the ADU group and CRT group largely disappears, with the ADU group associated with marginally lower costs than the CRT group. The significant imbalance at baseline between the groups in terms of ‘previous inpatient admission’ appears to be particularly important. In model 2 this is the largest and most significant predictor of mental health costs, with those who have previously had an inpatient admission accruing significantly more costs than those who have not. The difference between trusts 1 and 4 remains significant; however, the cost difference is reduced following adjustment.

The final fully adjusted model 3 further reduces the cost difference between the two groups to a negligible amount (the ADU group is a mean of £37 cheaper than the CRT group). Again, whether or not the individual has had a previous inpatient admission is the largest predictor of cost and is highly significant. Those who have SMI have higher costs than those who do not, but the difference is not statistically significant at the 5% level (p = 0.051). Trust 4 is again associated with lower costs than trust 1 at a statistically significant level. Following adjustment for baseline imbalances, there is no significant difference between the costs accrued by the two groups. The full results can be seen in Table 19.

Analysis of Health of the Nation Outcome Scales scores

The raw unadjusted HoNOS scores are given in Table 20 for the groups at both baseline and 6-month follow-up. Both groups saw improvements between baseline and follow-up. The HoNOS scores in the CRT group improved from 12.43 to 11.91 and those in the ADU group improved from 14.11 to 13.26.

The analysis of the change from baseline in HoNOS scores is shown in Table 21. When adjusted solely for clustering, model 1 suggests that the ADU group improved slightly (not statistically significantly) more than the CRT group, with a larger fall in HoNOS score between baseline and follow-up.

When model 2 is partially adjusted by adding the statistically significant baseline variables, this difference disappears, with the CRT group experiencing a larger fall in HoNOS score than the ADU group. In this model, both baseline HoNOS score and previous inpatient admission are significant predictors of larger falls in HoNOS score. This suggests that the baseline differences in groups may be responsible for the majority of the difference between the two groups in model 1. Employment is also a significant predictor of a decrease in HoNOS score, with the suggestion that those in employment improved more than those who were unemployed over the 6-month period. Additionally, both trust 3 and trust 4 saw statistically significant improvements in HoNOS score compared with trust 1 (the comparator).

The implications of the final fully adjusted model 3 are similar to those of model 2. Again, the CRT group improves more than the ADU group, with baseline HoNOS score the most significant predictor of improvement. Employment status remains a statistically significant predictor of improvement from baseline. Trust 3 sees greater improvements than the comparator trust 1. Although it maintains the same direction, previous inpatient attendance is no longer statistically significant in this model. Given the strong and significant associations between baseline HoNOS score and HoNOS score change in models 2 and 3, this suggests that the larger change from baseline for the ADU group seen in model 1 is likely to be regression to the mean, with those in a worse initial health state having more potential to improve. This is reflected in the reversal of direction for the treatment group variable once covariate imbalances are included.

Cost-effectiveness analysis

The results of the fully adjusted joint analysis of costs and HoNOS score change are shown in Table 22.

When costs and outcomes are analysed jointly, a marginal cost saving is associated with the ADU group compared with the CRT group. However, in terms of HoNOS score change, the ADU group performed less well than the CRT group. As demonstrated by the large CIs surrounding the estimates, there is a great deal of uncertainty around these results. To examine this uncertainty and the corresponding probability of cost-effectiveness, a CEAC was created, which is shown in Figure 4.

FIGURE 4.

Cost-effectiveness acceptability curve: probability of cost-effectiveness.

If we do not value benefits (i.e. no willingness to pay for HoNOS improvements), then there is a 52% probability that the ADU group is the more cost-effective option. As the willingness to pay for HoNOS improvements increases, the CRT group becomes the superior choice, with a 65% probability of being the more cost-effective option at a willingness to pay of £2000 per HoNOS point decrease before levelling off at 70% at a willingness to pay of £6000 per HoNOS point. Given the lack of difference in costs, this change is driven largely by the fact that, after baseline adjustment, the CRT group was associated with greater improvements than the ADU group.

The results of the four sensitivity analyses are shown in Table 23.

In the first two sensitivity analyses, the costs associated with care are altered. In terms of the joint analysis of costs, this has relatively little impact on the overall findings. In this first instance, the costs associated with the ADU are marginally higher than those associated with CRT, but this difference is negligible. Likewise, in sensitivity analysis 2, the cost savings associated with the ADU group increase compared with those associated with the CRT group. However, in the context of the uncertainty around the cost estimates, this has very little impact on the overall picture, as demonstrated in Figures 5 and 6, which show that, once any willingness to pay for benefits is considered, the CRT becomes more likely to be the cost-effective option. Sensitivity analysis 3 explores the impact of excluding those who crossed from the CRT group to the ADU group. In this analysis the ADU group is associated with higher costs than the CRT group. This cost difference is, again, relatively small when the uncertainty around the estimates is considered. The impact of this change is shown in Figure 7, which shows that CRT is the more cost-effective option at all levels of willingness to pay for benefits. In the final sensitivity analysis, which imputes for missing data, the ADU group is associated with lower costs than the CRT group; however, again there is substantial uncertainty around this estimate. The CRT group performs better than the ADU group in this analysis, with a slightly larger coefficient than in the primary analysis. This has relatively little impact on the overall interpretation, as shown in Figure 8. If outcomes are ignored, there is a 60% chance that ADU is the more cost-effective option; however, once outcomes are valued, CRT becomes the more cost-effective option.

FIGURE 5.

Sensitivity analysis 1: CEAC for inflated ADU/crisis house costs.

FIGURE 6.

Sensitivity analysis 2: CEAC for deflated ADU/crisis house costs.

FIGURE 7.

Sensitivity analysis 3: excluding those who crossed over from CRT to ADU.

FIGURE 8.

Sensitivity analysis 4: multiple imputation.

Design

This used a qualitative study design, involving a series of interviews conducted by a peer researcher with staff and service users of ADUs.

Objectives

1. Explore the views of service users, carers and practitioners regarding the strengths and weaknesses of ADUs and their component interventions.

2. Explore service user, carer and practitioner views concerning the role of ADUs in the acute care pathway.

3. Integrate these findings with those from WP 2.1 to generate theory around how outcomes are achieved.

4. Develop a set of recommendations outlining best practice in this field.

5. Explore the reflections of the study team on ADUs.

Setting

Data were collected from people working in and using ADUs in four NHS mental health trusts in England.

Participants

We aimed to use convenience sampling to collect data from five service users, three carers and three staff members at each study site ADU. We anticipated that the total sample numbers of 20 service users, 12 carers and 12 staff members would enable us to capture a diversity of views within the funding limitations of the study. 37 To ensure that we included people who broadly reflected the range of ADU service users, we aimed to recruit people who varied in terms of age, sex, ethnicity and diagnosis. For the staff sample, we aimed to include a range of professional disciplines to reflect the make-up of ADU staff teams identified in our earlier mapping of ADUs in England. 1

Measures

Semistructured interview schedules were developed for each stakeholder group (staff, carers and service users). These were reviewed and revised by the study’s PPI group (see Report Supplementary Material 23). Some interview questions were broad to allow discussions to be guided by participants’ unique experiences (inviting descriptions of what the interviewee’s experience of the service was like, and how the service helped people); other questions had a more specific focus (particularly concerning comparisons with other acute services, ADU staff and service users, ADU environments, and safety). Scope was allowed for exploring other topics as these arose.

Procedure

Service users were recruited by local researchers as part of our larger study. The eligibility requirements remained the same for WP 2.2 as for WP 2.1 (aged ≥ 18 years; used the ADU for at least 1 week; understood English sufficiently to participate in an interview; had capacity to provide informed consent; and did not pose a high risk to others or to themselves). Interested participants were given a study information sheet and provided with further information during a telephone call with a study researcher. For staff interviews, researchers approached ADU managers to ask about their own participation, and, using a snowballing sampling approach, asked them to identify two more members of staff from the ADU team. For carer recruitment, study researchers asked staff to identify carers from their caseload, and local researchers asked all participants in our larger cohort study if they had family members or friends who might be interested in participating. Written informed consent was provided by all participants.

Interviews took place on site at the relevant ADU. All interviews with service users and carers took place during the period of service users’ treatment, between 11 and 90 days after service users’ first attendance at the unit. All interviews were conducted by two peer researchers (MD and JW) who have lived experience of using acute mental health services as well as experience of qualitative research. The researchers disclosed their peer status to service user participants during the call to explain the study.

Interviews took place between May 2017 and May 2018 in private rooms that the researchers booked at each of the ADU sites. The interviews were audio-recorded and transcribed.

Analysis

Interview data were analysed using thematic analysis38 in NVivo (QSR International, Warrington, UK) software. A primarily inductive approach was adopted, but researchers also sought answers to the initial research questions about the strengths and weaknesses of ADUs and their component interventions, and the role of ADUs in the acute care pathway. Data were analysed primarily by the two peer researchers who collected data, with collaborative input from all other authors. A small subsample of data was initially explored to develop a preliminary coding framework. This was subsequently revised and refined through a cyclical process of reading, coding, reflecting on the data, and collaborative discussion. As part of this, the study’s PPI group met twice during the course of the analysis and provided guidance and input from the perspective of those with lived experience.

Sample characteristics

Individual interviews were carried out with 21 service users (four at ADU 1, seven at ADU 2, and five at ADU 3 and ADU 4), and 12 staff members (three at each site). We were able to recruit only three carers from sites: one each from ADU 1, ADU 3 and ADU 4. Sample characteristics of service users and carers are shown in Table 24; sample characteristics of staff are shown in Table 25.

Findings

Eight broad areas were identified as being represented strongly in the data across all stakeholder groups. These were organised into two larger domains related to the two research questions with which this work is concerned: (1) day-to-day functioning of ADUs (covering group therapy, one-to-one support, environment and safety, structure and purpose, and suggested improvements) and (2) the wider picture (covering ADUs in relation to other treatment options, the role of ADUs in the acute care pathway and the wider picture).

Strengths and limitations

To our knowledge, this is the first study to explore, in depth, the strengths and weaknesses of ADUs, and their role in the acute care pathway, across a range of contexts and from the multiple perspectives of relevant stakeholders, particularly service users and staff members. The size of the data corpus – drawing on 36 individual interviews – and the range of respondent characteristics gives some cause for confidence that the findings are representative of the views of these two stakeholder groups.

The target number of carer interviews, however, was not reached; only 3 of 12 intended interviews took place, and they produced data that were far less rich than those for the other stakeholder groups, with interviews with carers averaging 11 minutes, compared with an average of 30 minutes for the other groups. Two of the three interviewees did not report being particularly engaged with the ADU, and the third was also a mental health professional and so had an unusually high level of knowledge and engagement with the staff team. Carer interviewees were sought both through study clinical studies officers asking participants in the cohort study and by asking staff to identify carers. Possible reasons for this include the following: all of the ADUs in the study seemed to take a passive approach to carer engagement, or, in the case of ADU 4, attempts at carer outreach were met with limited success; service users who attend ADUs may be more likely to be lacking in social support (and indeed this may be one reason for their referral to ADU treatment rather than to home treatment from a CRT); and carers have historically proved a difficult group to access for qualitative research.

Although the sample was well balanced in terms of age and sex, 81% of service user respondents were white British. This is broadly in line with the national picture; the area where ADU 2 is located is very ethnically diverse, and four of the seven respondents from that area were white British. Previous qualitative research has shown that ethnic minority groups face significant barriers to help-seeking for mental illness,42 and to have only one ethnic minority respondent in the sample may be cause for concern that ethnic minority voices are not adequately represented in this work.

A limitation of the study was that interviews were carried out in a small number of trusts, and those trusts were self-selecting in agreeing to participate in the study. It may be that only those ADUs that were highly functioning, and therefore had generally satisfied service users, were willing to take part in the study, thus biasing our sample. This self-selection bias is, of course, inherent in all research where participants are required to opt in, although we attempted to minimise this issue by inviting all ADUs in the country to participate.

This study placed strong emphasis on service user involvement: both of the peer researchers who were responsible for gathering data and analysing results have lived experience of mental health problems and using acute mental health services. In addition, the UCL team was partnered with the McPin Foundation, which provided a Lived Experience Advisory Panel who were consulted repeatedly throughout the process. This has ensured authentic service user involvement at the three key levels of consultation, collaboration and control,43 which helped ensure that service user voices were represented appropriately throughout the study.

The results presented here are primarily descriptive, with more in-depth analysis being carried out for future publications.

Research implications

This paper forms part of the wider AD-CARE study, part of the purpose of which is to determine the efficacy of ADUs as a method of acute crisis care. The findings of this study should be considered again in context with the results of the other parts of the study, particularly the comparative cohort study which has as its focus the same four ADUs, and seeks to assess their effectiveness at preventing re-admission and satisfaction with services provided. This is discussed in Chapter 10.

One question that has not been adequately addressed is the cost of ADU treatment compared with other forms of acute crisis care. A 2006 study suggested that the cost per day of ADU treatment was 70% that of inpatient care,13 but the ADU model has been developed significantly since then. Anecdotal evidence from this study suggests that the cost per day is closer to 50% that of inpatient care, which is a significant difference worth exploring further; cost-effectiveness is likely to be a major factor influencing policy regarding ADUs.

Clinical implications

Only around 30% of NHS trusts in England currently have ADUs, and the closure of several ADUs during the study period, including in two of the trusts involved in this study, implies that the trend is towards this proportion decreasing. If the results of this study are representative of the national opinion about ADUs, there is a case to be made that this trend should be reversed, given the value that service users and staff place on ADUs and their model of care.

Given the well-documented pressure on mental health acute wards and reduction in available hospital beds,4 acute care options need to be reviewed urgently. ADUs provide a model that this study shows is highly popular, and may also be more effective and less expensive than crisis care alternatives.

Design

This study was a multilevel secondary analysis of routinely collected annual administrative data from providers of secondary mental health care services in England during 2013–15.

Aims

1. To assess whether or not acute re-admission rates are reduced in areas/trusts with a more enhanced crisis care pathway, defined as having an ADU in the pathway, determined from WP 1.

2. To assess whether or not in trusts with ADUs, individuals who access NHS-funded ADUs have different outcomes from similar (propensity score matched) people who have had an acute episode but do not access ADUs.

After data availability in the MHMDS was examined, a change was made to the protocol and we excluded aim 2 from our study [see Why acute day units were (not) identifiable in the Mental Health Minimum Data Set].

Objectives

• To determine the re-admission rate to the acute care pathway in a national sample.

• To determine the characteristics (predictors) associated with re-admission.

• To determine whether or not trusts with an ADU differed in terms of re-admission rates to the acute pathway.

• To explore the use of MHA in re-admissions.

• To explore access to emergency services and use of the MHA after discharge from index admission (supplementary analyses).

Data source

• The MHMDS: a national data set containing anonymised individual-level data about adults using secondary adult mental health services for the financial years 2013/14, 2014/15 and 2014/15 (incomplete, up to November 2015).

• For supplementary analyses we used Hospital Episode Statistics (HES) data sets, which include administrative data for inpatient admissions to NHS acute hospitals (the Admitted Patient Care data set) and attendance at NHS A&E departments (A&E data set) in England. The linkage service was provided by the NHS Digital and was at the (pseudo/anonymised) individual level.

Study period

• Cohort entry: index admission starting between 1 April 2013 and 31 May 2015.

• Cohort exit: 6 months after discharge from index admission discharge or re-admission to acute care pathway if earlier than 6 months.

Study population

The population was individuals who were discharged from the acute care pathway in England during our study period and reported to the MHMDS who met the following inclusion criteria:

• aged ≥ 16 years

• not living with dementia or learning disabilities

• receiving mental health care provided by a single NHS trust (thus excluding non-NHS trusts and acute trusts and care provided by multiple NHS trusts).

Main study outcome

The main outcome was re-admission to the acute care pathway within 6 months of index admission discharge.

Supplementary study outcomes

• Re-admission to inpatient ward.

• Re-admission under the MHA.

• Attendance at A&E.

• Attendance at A&E for self-harm.

• Admission to acute hospital.

Predictors of re-admission selected a priori

• Age.

• Sex.

• Marital status.

• Ethnicity.

• Payment by Results (PbR) care cluster.

• IMD.

• Type of index admission (CRT or inpatient ward).

• Length of stay at index admission.

• ADU trust (vs. non-ADU trust).

Definitions

• ADUs: non-residential services offering daytime treatment and care to adults experiencing a mental health crisis who would otherwise be considered for acute psychiatric hospital admission, or other alternatives to admission including services provided by the CRT. Trusts with ADUs were identified in the initial mapping exercise conducted in 2016 (see Chapter 4) and further revised for this study’s period (2013–15).

• Index admission to the acute care pathway: the first admission to an inpatient ward and/or CRT.

• Re-admission to the acute care pathway: the first re-admission occurring in the first 6 months following discharge from index admission. Note that throughout this study we refer to admission and re-admission as admission and re-admission for treatment in the acute care pathway respectively (unless otherwise stated).

• Main outcome: re-admission to the acute care pathway within 6 mon.hs after discharge from index admission, meaning a new episode of inpatient or CRT treatment.

• Re-admission to inpatient care (ward): re-admission to the acute care pathway as an inpatient; if an individual was also admitted to CRT at some point during the same episode, the admission was still considered as re-admission to inpatient care.

• Attendance at A&E: the first attendance to A&E during the 6-month period following discharge from index admission or before/at re-admission to the acute care pathway.

• Admission at an acute general hospital: the first admission to an acute general hospital (mental health hospitals excluded) during the 6-month period following discharge from index admission or before/at re-admission to the acute care pathway.

• Index admission under the MHA: an admission during which an individual was formally detained under any section of the MHA 1983 or other acts at any time during the index admission period.

• Re-admission under the MHA: an admission during which an individual was formally detained under any section of the MHA 1983 or other acts within the first 7 days after the re-admission start date.

Information governance

As outlined above, this study was granted ethics approval (references 16/LO/2160 and 17/CAG/0101).

The data in this study contained confidential service user-level information, and the statistician (GF) and the chief investigator of the study (DO) were granted access to anonymised data by the NHS Digital Data Access and Advisory Group. All data analyses took place through the UCL Data Safe Heaven.

Data Safe Heaven is built using a walled-garden approach, whereby the data are stored, processed and managed within the security of the system, avoiding the complexity of assured end-point encryption. Data Safe Haven has been certified to the ISO27001 information security standard44 and conforms to NHS Digital’s Data Security and Protection Toolkit.

The structure of the Mental Health Minimum Data Set

For each financial year there are the following data files:

• The Records data file includes an individual’s demographic data, LSOA and the provider and commissioner of the mental health services. From this file we extracted data on age, sex, ethnicity, marital status, LSOA and mental health service provider.

• The Episodes data file includes data on care delivered over extended periods of time, and these periods can overlap. From this file we extracted data on admission to ward, admission to mental health team and PbR care cluster assignment.

• The Events data file includes data on care delivered at specific moments in time, e.g. diagnosis and outpatient contact with a health-care professional. From this file we extracted data on primary and secondary diagnosis (and HoNOS).

Each individual in contact with mental health services reported to the MHMDS has a personal identifier, which does not change across the years and is linked with records, episodes and events. We used the personal identifier to link the annual data sets and the data fields of interest. Each episode and each event also have a personal identifier. We were warned that the same episode (or event) may have more than one identifier across the years or within the same year, creating a considerable number of identical episodes (or event) when merging the yearly data sets. Exploratory analysis confirmed this and, as part of the data management, a large amount of time was spent identifying and removing duplicate entries.

For details on data fields, we referred to the document ‘MHMDS specification.xlsm’ version 4, available on the NHS Digital website. 45

How the acute care pathway was identified in the Mental Health Minimum Data Set

Based on our definition of acute care pathway, we identified the following admissions:

• admissions to CRT

• admissions to inpatient ward.

Admissions to the CRT were extracted from the data field ‘TEAMEP’ in the data file ‘Episode’. The ‘TEAMEP’ data field is defined as ‘a continuous period of care for a patient under the responsibility of one or more mental health teams’ and contains a record for every time an individual was under the care of a mental health team and the type of team. We chose code = A02 for the CRT. In the TEAMEP table we also found a further code for CRT ( = A03) and home treatment ( = A04), but when this was compared with the NHS Dictionary we found some discrepancies in the coding. We contacted NHS Digital for clarification and also investigated the length of admissions by team coding (A02 vs. A03 vs. A04), as we expected CRT admissions to be the shortest (see Report Supplementary Material 5). NHS Digital reassured us that crisis resolution and home treatment team was coded A02 in the MHMDS.

Admissions to inpatient ward were extracted from the data field ‘PROVSPELL’ in the data file ‘Episode’. The ‘PROVSPELL’ is defined as ‘a continuous period of inpatient care under a single Hospital Provider, starting with hospital admission and ending with discharge from hospital’ and contains a record of every time an individual was a ward inpatient.

For details on how we defined admissions and extracted variables from MHMDS, see Report Supplementary Material 5.

How the study population was identified in the Mental Health Minimum Data Set

First, we identified individuals who did not meet the inclusion criteria:

• Those aged < 16 years, from the data file Records.

• Those whose care was provided by a non-NHS trust or an acute trust or those whose care was provided by more than one trust, from the data file Records.

• Those with dementia or learning disabilities, from the data files Events and Episodes, using the following exclusion criteria:

  • Having a primary or secondary diagnosis of F70–F79, F00–F09 or G30–G32 (from the data fields PRIMDIAG and SECDIAG of the data file Events).

  • Having been assigned an organic PbR care cluster (18–21 or organic superclass care cluster) (from the data field CLUSTER of the data file Episodes).

  • Ever being admitted to the following mental health teams: organic, young-onset dementia, memory services/clinic, forensic learning disability service, learning disability services or community team for learning disabilities.

Second, we identified people who had been admitted to the acute care pathway (i.e. admitted to hospital and/or crisis team; see Chapter 6, Methods: cohort study) at least once during the study period, and excluded those who had not been admitted. For details, see How admissions to the acute care pathway were identified.

Last, we excluded people whose discharge date was unknown and those who were discharged after 31 May 2015 to ensure that 6 months of follow-up data were available for the analysis.

The remaining data set comprised 231,998 persons in 56 NHS trusts. For a visual summary, see the flow chart in Figure 9.

FIGURE 9.

Flow chart of how persons admitted to the acute care pathway between April 2013 and May 2015 reported to the MHMDS were identified for analysis. Adapted with permission from Osborn et al. 46 This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

Adapted with permission from Osborn et al. 46 This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

How admissions to the acute care pathway were identified

Exploratory analysis identified a large proportion of persons who had duplicate admissions (more than one admission starting and ending on the same date) but had different identifiers. The proportion of duplicate/identical admissions varied largely by NHS trust.

We noticed that even for the same person there were also admissions with overlapping dates, for example more than one admission starting on the same day but ending at different days or admissions starting and ending within the date range of another admission. We assumed that these admissions were all part of one single admission or episode as long as there was not a gap of more than 1 day between the end date of one and the start date of the next.

We developed an algorithm to identify and merge these identical entries (see Flow chart for decision-making in Report Supplementary Material 6–9). We proceeded as follows:

1. First, we applied this algorithm to clean the hospital admissions. We removed duplicate and merged admissions into one single admission if dates were overlapping. If an admission started on the same day that or the day after the previous admission ended, we considered these two admissions as one single admission episode (see Report Supplementary Material 7).

2. We used the same approach to identify, remove and/or merge admissions to CRTs. We noticed that some admissions to CRTs lasted longer than 180 days. We assumed that this was a data entry error as the duration of an admission to a CRT is much shorter than this (see Report Supplementary Material 8).

3. Finally, we created a variable for ‘admission to acute care pathway’ by merging admissions to hospital and admissions to CRTs into one single admission or acute care episode (see Report Supplementary Material 9).

4. To prepare the data set for analysis, we also excluded people whose date of discharge from their index admission was later than 31 May 2015 (see Report Supplementary Material 10).

Table 26 summarises the number of people identified at each stage and the final number of people of our study population for analysis.

How predictors of acute re-admission were identified

The following predictors were included:

• age at the time of index admission, extracted from the Records file

• sex, extracted from the Records file

• marital status at the time of index admission, extracted from the Records file

• ethnicity, extracted from the Records file

• PbR care cluster, extracted from the Episodes file

• IMD (see Index of Multiple Deprivation)

• being admitted in a trust with an ADU compared with being admitted in a trust with no ADU, from AD-CARE WP 1.

How individuals detained under the Mental Health Act were identified

First, we identified all episodes during which an individual was detained under the MHA between April 2013 and November 2015. We excluded MHA sections starting before 1 January 1990, those ending before the index admission start and those starting after 6 months from discharge. To explore whether or not a re-admission occurred under the MHA, we excluded sections that started later than 7 days from the date of re-admission. Sections of the MHA under which an individual was detained using the same code and starting on same date but ending on different dates were considered as one single episode under the MHA and the section with the most recent date was retained. For details of data cleaning for MHA definitions, see Report Supplementary Material 21.

How individuals attending accident and emergency and admitted to acute hospitals were identified

To obtain information on those in our study population who attended an A&E department or were admitted to an acute hospital within 6 months from discharge or before re-admission to acute care, we used the HES data sets, namely the Accident & Emergency (A&E) data set and the Admitted Patient Care data set. NHS Digital provided a bridge file that linked the two main data sets (MHMDS for the years 2013–16 and HES) at the individual level. After excluding people who were not in the acute care pathway during our study period and those who were admitted to a mental health hospital (and recorded in the Admitted Patient Care data set), we identified 29,298 people who attended A&E and 16,877 people who were admitted to an acute hospital and met our criteria (see flow chart in Report Supplementary Material 22). For details on how we extracted variables from HES, see Report Supplementary Material 5.

Why acute day units were (not) identifiable in the Mental Health Minimum Data Set

We were unable to find a reliable code for using an ADU in the MHMDS to address our second aim in Methods: Mental Health Minimum Data Set study, Aims, regarding individual ADU use.

This section provides some detail regarding the steps we took to attempt to overcome this problem. An initial search in the MHMDS dictionary showed that there was no field or category within a field indicative of an episode of ADU care. As data are entered in the MHMDS for the purpose of reimbursing the services provided by the NHS trusts, we expected that ADU services could be identified. We explored several fields and categories that could be used as proxies for ADU.

We hypothesised that if one of these fields was indeed a proxy for ADU then we would expect virtually zero individuals accessing this ‘proxy’ service in trusts that had been identified as not having an ADU in the WP 1 mapping survey.

Second, we explored the following fields that might be suggestive of ADU:

1. HPCONT – an event that describes the location where a day contact with a health-care professional (HPCONT) took place. It includes outpatient clinic, day hospital or day centre, which potentially could be used as alternative wording for ‘ADU’.

2. DAYATT – an event that describes an attendance at an NHS day-care facility, a term that potentially could be used as alternative wording for ‘ADU’.

3. DAYEP – an episode that describes a continuous period of care for a service user receiving NHS day-care services, which potentially could be used as alternative wording for continuous access to ‘ADU’.

Finally, we compared the number and percentage of individuals accessing the services above in trusts with and trusts without ADUs from the mapping exercise (WP 1). The percentage of individuals with these codes is based on the total number of individuals in the acute care pathway as identified in WP 3 in ADU and non-ADU trusts.

All fields that we explored had been accessed by considerable numbers of individuals in both ADU and non-ADU trusts (Table 27). The proportion of individuals recorded in NHS day-care services (DAYEP), in outpatient clinics (HPCONT) and in NHS day-care facilities (DAYATT) was higher in trusts that had an ADU, suggesting that some trusts may use these fields to record ADU services within the scope of reimbursement. However, because some individuals were also using these services in trusts with no ADUs, we could not be confident in using any of these fields as a binary (yes/no) variable to identify individuals who definitely accessed an ADU service. We also concluded that these codes could be used for many other types of service, including longer-term day services, that are not part of the acute care pathway.

Aim (restricted to aim 1)

• To assess whether or not acute re-admission rates are reduced in areas/trusts with a more enhanced crisis care pathway, defined as having an ADU in the pathway, determined from WP 1.

Objectives (restricted to aim 1)

• To determine the re-admission rate to the acute care pathway in a national sample.

• To determine the characteristics (predictors) associated with re-admission.

• To determine the characteristics of people admitted in ADU trusts compared with those admitted in non-ADU trusts.

• To determine whether or not trusts with an ADU differed from the acute pathway in terms of re-admission rates.

Supplementary objectives

• To determine the individual characteristics associated with re-admissions to inpatient care.

• To determine the individual characteristics associated with re-admissions under the MHA.

• To determine the individual characteristics associated with attendance at A&E, attendance at A&E for self-harm and admission to acute hospital before or at cohort exit.

Method

Characteristics of the study population

We identified 231,998 individuals admitted to the acute care pathway between April 2014 and May 2015. The clinical and sociodemographic characteristics of the participants are described in Table 28. Briefly, they were mainly young/middle aged, of white ethnicity, single and living in more deprived areas. The majority had been admitted in a trust that had no ADU. In terms of clinical presentation, the most common care cluster group was non-psychosis. On average, the index admission lasted 10 days (median 10 days, IQR 1–30 days), and the most common index admission was admission to a CRT (156,968/231,998; 67.7%).

Over one-fifth (49,547/231,998; 21.4%) of individuals were re-admitted to the acute care pathway within 6 months of being discharged from their index admission, on average after 34 (median) days (IQR 10–88 days), and nearly half were admitted to inpatient care (23,290/49,547; 47.0%). Individuals who were re-admitted were more likely to be women, single, of mixed or black ethnicity, living in a more deprived area and to receive care for severe psychosis (see Table 28).

Characteristics of people admitted and re-admitted to the acute care pathway in trusts with and trusts without acute day units

We identified 61,450 (26.5%) individuals whose care was provided by an ADU trust and 170,548 (73.5%) individuals whose care was provided by a non-ADU trust. There were differences between these two groups. Individuals admitted in an ADU trust were more likely to be single, younger and of non-white ethnicity; they were more likely to be admitted to hospital (rather than a CRT) (Figure 10) and to be in a care cluster for psychosis (moderate or severe) (Table 29). There were no differences in the proportions of people being re-admitted and in the characteristics of individuals being re-admitted to the acute care pathway within 6 months between ADU and non-ADU trusts (see Report Supplementary Material 16 and 17). The probabilities of not being re-admitted during the 6 months following discharge from index admission to acute care are shown in Figure 11 (survival curve).

FIGURE 10.

Venn diagrams showing the number of individuals in the acute care pathway included in the study by type of index admission: (a) ADU trusts (n = 61,450); and (b) non-ADU trusts (n = 170,548).

FIGURE 11.

Survival curve showing the probabilities of not being re-admitted within 6 months following discharge from index admission to acute care (analysis time in number of days). Re-admission to acute care pathway by ADU and non-ADU trusts.

Variance across trusts and the null model

The null model suggested that the odds of being re-admitted in an ‘average’ trust were 0.25 (corresponding to a probability of 0.20) and that the 95% predicted odds of being re-admitted across trust ranged from 0.15 to 0.42 (from 0.13 to 0.30 in probabilities).

The caterpillar plot (Figure 12) shows that for a substantial number of trusts (irrespective of whether or not the trust had an ADU) the 95% CI did not overlap the horizontal line at zero, indicating that the odds of being re-admitted in these trusts were significantly above average (above zero line) or below average (below zero line) and that there was no substantial difference between ADU and non-ADU trusts.

FIGURE 12.

Caterpillar plot of the residuals by trust (null model).

Missing data

Among the 231,998 people included in this study, 90,783 (39.1%) had at least one missing observation for sex, marital status, ethnicity, care cluster or IMD (Table 30). The variables with the largest proportion of missing observations were care cluster (with 25.7% missing observations) and marital status (with 21.7% missing observations).

An investigation into the frequency of missingness by re-admission and age suggested that data were not missing completely at random and that there was strong evidence that re-admission and age were independently predictive of missingness in ethnicity, care cluster and marital status (see Report Supplementary Material 19–20). Therefore, a multivariable analysis using only complete cases (i.e. people with no missing data) would not be appropriate.

Predictors to re-admission to the acute care pathway

Findings from the multivariable multilevel analysis (Table 31) suggest that severe and less severe psychosis (compared with non-psychosis), ageing (compared with aged ≤ 24 years), being female (compared with male), being single (compared with non-single), being of mixed ethnicity (compared with white ethnicity), living in most deprived areas (compared with living in least deprived areas) and being previously admitted to inpatient care are individual characteristics associated with an increased risk in re-admission to acute care. In contrast, being of Asian ethnicity (compared with white ethnicity) reduces the risk. Most (98%) variation was attributable to individual patient- rather than trust-level effects (2.0%). We did not observe any risk increase in trusts without ADUs compared with in those without ADUs (OR 0.96, 95% CI 0.80 to 1.15).

Re-admissions of inpatient care and re-admissions under the Mental Health Act

Nearly half (23,290/49,547, 47.0%) of re-admissions were to an inpatient ward. The risk of being re-admitted was 11% in an ‘average’ trust, and it was between 5% and 26% in 95% of trusts. Multivariable analysis (Figure 13) suggested similar associations between re-admission and individual characteristics to the main analysis, although some associations increased in effect size (psychosis, ageing, index admission at an inpatient ward, mixed ethnicity) or became statistically significant (being of black ethnicity increased the risk of re-admission to inpatient ward only) or became non-significant (e.g. IMD).

FIGURE 13.

Forest plots showing mutually adjusted ORs for re-admission inpatient care by predictors. Reproduced with permission from Osborn et al. 46 This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated. The figure includes minor additions and formatting changes to the original figure.

Reproduced with permission from Osborn et al. 46 This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated. The figure includes minor additions and formatting changes to the original figure.

Fourteen per cent (6814/49,547) of re-admissions were under the MHA. This percentage is quite similar to the proportion of index admissions that occurred under the MHA (16%; 36,454/231,998). The characteristics of individuals re-admitted under MHA are presented in Table 32. Multivariable analysis suggests that previous admissions under the MHA, severe and non-severe psychosis are strongly associated with re-admission under MHA, as, to a lesser extent, is being of black, mixed or Indian ethnicity and ageing (Figure 14).

FIGURE 14.

Forest plot showing the adjusted ORs of being re-admitted under the MHA within 6 months of discharge from the index admission.

Attendances at accident and emergency and admissions to acute general hospital

Following discharge from index admission, 29,298 (12.6% of all cohort) people attended A&E and 16,877 (7.3%) were admitted to acute hospital. The first attendance at A&E occurred, on average, after 10 (IQR 1–52) days, and self-harm was the cause in 10% (2947/29,298) of these cases. The first admission to acute general hospital occurred, on average, after 44 (IQR 10–100) days, and the most frequent primary diagnosis (3679/16,877; 21.8%) was poisoning by drugs or medications (ICD-10 T36–T50), the most common (1206/16,887; 7.2%) being poisoning by non-opioid analgesics, antipyretics and antirheumatics (ICD-10 T39). The characteristics of individuals attending A&E and those admitted to acute general hospital are presented in Table 32. The results of the multilevel multivariable analyses are shown in Figures 15 and 16. For both outcomes (attendance at A&E and admission to general acute hospital), non-psychosis, being of white ethnicity, being female and staying in an inpatient ward at the index admission increased the risk. However, A&E was more likely to be attended by young people, whereas older people were more likely to be admitted to acute hospital. Living in a more deprived area increased the risk of attending A&E but not of being admitted to hospital.

FIGURE 15.

Forest plot showing the adjusted ORs of attending A&E within 6 months from index admission discharge.

FIGURE 16.

Forest plot showing the adjusted ORs of being admitted to general acute hospital within 6 months from index admission discharge.

Summary of main findings

This study identified more than 230,000 individuals who were admitted and discharged from the acute care pathway in 56 NHS mental health trusts in England between 2013 and 2015. Over 20% of people were re-admitted to mental health acute care within 6 months, half of whom were admitted to an inpatient ward. The median time to re-admission was only 34 days after discharge. Analyses suggested no association between re-admission and the provision of an ADU within a trust’s catchment area. There was large variation in the proportion of acute re-admissions across trusts in England, but multilevel analysis suggests that most of these variations are explained at the level of the individuals who are admitted rather than by trust-level characteristics. The predictors of any acute re-admission identified by the analyses were psychosis, older age, being female, being single, being of mixed ethnicity, living in deprived areas, having shorter index admissions and being admitted to an inpatient ward at the index admission. The predictors of inpatient admission and admission under the MHA were broadly similar.

Almost 13% of the sample attended A&E after discharge from acute care, with a median time to attendance of 10 days. A&E attendees were more likely to be female, of white ethnicity and younger, and are less likely to have a psychosis care cluster. As with other outcomes, people discharged from inpatient stays were more likely to attend.

Strengths and limitations of study

The strength of our study is the size and comprehensive national coverage of the MHMDS, which allowed us to investigate the proportions and predictors of re-admissions to mental health acute care and inpatient wards. To our knowledge, this is the first study that has been able to address this question nationally. However, we were not able to identify access to ADUs at the individual level in the MHMDS data set and were able to answer only the first of our study aims by using a binary indicator at the trust level for ADUs.

The lack of differences between trusts with and trusts without ADUs may reflect the relatively small numbers of people who use ADUs compared with CRTs and psychiatric inpatient services.

The nature of administrative data collected for reimbursement purposes rather than research limited our analyses as we did not have information about (and thus we could not adjust for) important factors that may affect re-admission, such as social support, housing status previous mental health and medical history and employment status. Using the MHMDS is extremely time-consuming, requiring long periods of data cleaning to identify variables, to exclude erroneous and duplicate entries, and to create cohorts that span different years in the data set. Because of the time limitation of this project (18 months), we were unable to conduct further stratified analyses by factors such as psychosis (vs. non-psychosis), age and sex, although the factors associated with re-admission may affect these groups in a different manner.

The MHMDS comprises data submitted from trusts’ records, and the quality and precision of data entry may vary across trusts. For example, there were large variations in the coding of CRT admission, suggesting that coding as well as provision of CRT may vary across trusts. We identified a substantial number of repeated entries, which were dealt with a series of algorithms that allowed us to approach data cleaning in a standardised manner.

Recommendations for future research

Working with MHMDS demands a systematic approach to data cleaning but can still provide answers to research questions regarding mental health care at a national level. Future analysis on re-admission to mental health care should also consider stratifying analysis by mental health disorders groups (i.e. psychosis vs. non psychosis), age and sex. The MHMDS does not contain granular detail to allow a more accurate prediction of which individuals are at greater risk of re-admission.

The disparity between re-admission rates of white and ethnic minority patients is concerning and warrants further investigation. Our findings align with existing evidence about ethnic variations in involuntary admissions. 48

For future research it would be helpful if more accurate coding frameworks were developed to allow different service models in mental health services and the MHMDS to be identified, so that their effectiveness can be explored.

The re-admission rates following discharge are clearly of concern, and further research is needed to determine models of community care or support after discharge that might reduce these re-admission rates.

The overall findings regarding national re-admission rates after a period of acute care in England are discussed in the context of the AD-CARE programme in the following chapter.