The aim of this WP was to describe supported accommodation in England and factors associated with positive outcomes and costs.
The main objective of the national cross-sectional survey (WP2i) was to describe the provision and quality of supported accommodation for people with mental health problems in England and to investigate service and service user factors associated with service users’ quality of life, autonomy and satisfaction with care. Specifically, we investigated the following research questions:
- Is there a difference in service users’ quality of life, autonomy, ratings of the therapeutic milieu of services and satisfaction with care between the three main types of supported accommodation (i.e. residential care, supported housing and floating outreach services)?
- Can any variation in these be explained by service and service user characteristics?
Work package 2i: national survey of supported accommodation services
Methods
Our sample size was estimated to assess the difference in proportion of people moving on from each of the three types of supported accommodation 30 months after recruitment (assessed in the second component of WP2 – the cohort study). Our original estimate was that we needed to recruit 90 services (30 of each type) and 450 service users (five from each service) from 14 nationally representative areas based on an intraservice cluster correlation coefficient of 0.07 and a mean cluster size of five. We selected the 14 areas using an index developed by Priebe et al.11 for their postal survey of supported accommodation, which ranks local authority areas on the basis of mental health morbidity, social deprivation, urbanicity, provision of community mental health care, supported accommodation, local authority mental health-care spend and housing demand. Recruitment was carried out between 1 October 2013 and 31 October 2014.
The researchers contacted key personnel working in housing departments in each of the 14 areas to first scope the local provision of mental health supported accommodation services and the number of places available in each. Residential care homes for adults with mental health problems in each area were also identified from the Care Quality Commission (the registration authority for care homes in England and Wales) website to ensure that none was missed (www.cqc.org). Services that were unusually large (> 80 service users) or small (< 6 service users) were excluded to increase generalisability of the sample. Five areas were replaced with another area closest on the sampling index: three because the area had no residential care services and two because the researchers were unable to clarify local supported accommodation provision (see Appendix 2, Table 5).
After 15 services of each of the three types were recruited, we reviewed our sampling strategy. We recalculated our intraservice correlation coefficient using responses to one item of the managers’ research interview [their estimate of the proportion of service users who had moved to more independent accommodation (residential care and supported housing) or required less support (floating outreach) in the last 12 months] and found it to be 0.18. A larger service user sample (624 rather than 450) was thus required. Owing to the relatively small number of residential care services identified, the number of services needed was adjusted to 21 residential care, 35 supported housing and 35 floating outreach, and the number of service user participants per service was increased to a maximum of 10.
The researchers randomly ordered the services in each area and within each of the three service types and contacted service managers sequentially to invite their participation. Service managers were given up to 4 weeks to reply. When the target number of services of each type had been recruited or the list exhausted for each area, the researchers moved on to the next area (see flow chart in Killaspy et al.22).
Where service managers responded to the initial contact, the researchers arranged a time to discuss the study further by telephone. If the manager was willing for their service to participate, the researchers arranged a face-to-face meeting to explain the study to staff. A list of potentially eligible service users was then obtained from the service manager (excluding those they considered to be unable to give informed consent to participate and those absent from the accommodation for any reason). Each potentially eligible service user was allocated a unique identifier code. The service user list for each service was randomly ordered by the researchers in blocks of six. Service users in the first block were approached in any order by the researchers to seek their participation and subsequent blocks were generated until five service users had been recruited or the list was exhausted for each service. All potential participants received a participant information sheet about the study and had at least 2 days to read it and address queries to the researchers before giving their informed consent to participate.
Data collection
The researchers completed face-to-face interviews with the service manager, keyworkers and service users at which the following data were collected.
Service manager: description of the service
- Service quality was assessed using the QuIRC-SA.20
- A proforma was used to gather details of the annual budget, referral processes and input from local mental health services.
Keyworker staff: service user assessments
- A pro forma was used to collect service user participants’ clinical and risk history.
- A Likert-type scale assessed the staff member’s expectation of the service user moving on (residential care/supported housing clients) or managing with less support (floating outreach clients) in the next 12 months.
- Challenging behaviours were assessed with the Special Problems Rating Scale (SPRS),31 which records the presence and severity of 14 challenging behaviours, giving a total mean score of 0 to 2.
- Needs were assessed with the Camberwell Assessment of Needs Short Assessment Scale (CANSAS),32 which rates 22 domains of care on a 3-point scale (0 = no need, 1 = met need, 2 = unmet need).
- Use of substances was rated with the Clinician Alcohol and Drugs Scale (CADS),33 from which a dichotomous problematic/non-problematic score can be derived.
- Social functioning was assessed with the Life Skills Profile (LSP).34 The 39 items rate various aspects of the service user’s social and everyday functioning from 1 to 4. Higher scores denote better functioning. A total score (from 39 to 156) and five subdomain scores can be derived.
- The adapted version of the Client Service Receipt Inventory (CSRI)35 was used to collect contacts with professionals/services within and outside the supported accommodation service, details of any medical and/or psychiatric admissions and contacts with family members over the past 3 months. These data were used in the health economic assessment.
Service users
- A pro forma was used to collect sociodemographic details and details of any incidents of abuse (verbal, physical or sexual abuse), self-harm or exploitation by others from within or outside the service over the last 12 months.
- Quality of life was assessed with the MANSA;29 the service user rates 12 life domains on a scale from 1 (could not be worse) to 7 (could not be better). A total mean score between 1 and 7 is generated.
- Social inclusion was assessed with the Social Outcome Index (SIX),36 which gives a rating from 0 to 6 from four social domains [i.e. employment, housing, living alone or with family/partner, and contact with friend(s)].
- Service users rated their autonomy using the Resident Choice Scale (RCS).37 The degree to which they have choice over 22 aspects of daily activities is rated on a four-point scale, giving a maximum possible score of 88.
- Users of residential care and supported housing services rated the therapeutic milieu of the service with the Good Milieu Index (GMI).38 General satisfaction with the service and the degree to which it facilitates confidence and abilities are rated on a scale of 1 to 5, with higher scores denoting greater satisfaction.
- Satisfaction with care was assessed with the CAT-SA.21
Data analysis
Data were entered into a purpose designed database by the researchers. After data cleaning, data were transferred to Stata® (StataCorp LP, College Station, TX, USA) statistical software for analysis. Differences between services, including the adapted QuIRC domain scores, service user characteristics and ratings of standardised assessments were investigated using simple descriptive statistics and regression analyses. Multilevel regression was used to investigate the association between service factors (adapted QuIRC domain ratings and contextual factors) and service user factors (sociodemographic characteristics, clinical history, social functioning, needs, substance misuse and challenging behaviours) with service user ratings of quality of life, autonomy, good milieu of the service and satisfaction with care (see www.ucl.ac.uk/psychiatry/research/epidemiology-and-applied-clinical-research-depa/projects/quest-project for full details of the analysis plan). We assumed the convention that in any regression analysis at least 10–20 participants for each predictor should be entered into the model.
Results
A total of 22 residential care, 35 supported housing and 30 floating outreach services were recruited. From these 87 services, 619 users were recruited (residential care, n = 159; supported housing, n = 251; floating outreach, n = 209). A total of 193 keyworkers completed ratings on service users.
The full characteristics of the three types of service and their users are shown in tables 1 and 2 of Killaspy et al.22
In summary, floating outreach provided more places than the other two types of service. Both floating outreach and supported housing services expected to work with their users for a median of 2 years, whereas for residential care this was 5 years. All three types of service used similar processes to assess new referrals. Almost all services had clinical input from a community mental health team or community rehabilitation team, despite the fact that only one-third of floating outreach clients were subject to the Care Programme Approach (vs. almost all residential care and supported housing clients). Supported housing services scored highest on six of the seven domains of the QuIRC (floating outreach scored slightly higher for human rights).
Most service users were male (66%), single (66%) and unemployed (82%). Users of residential care services were older and had been known to mental health services longer than users of the other two types of service. Most (68%) had a primary diagnosis of psychosis but one-third of floating outreach service users had depression/anxiety. The proportion of service users with substance misuse problems was relatively small (16% alcohol, 12% drugs), with the lowest prevalence among residential care service users. Users of residential care and supported housing had slightly more previous admissions than users of floating outreach and more were subject to a community treatment/restriction order. Overall, 40% of service users had committed an act of violence at some time, but there were few serious incidents of risk to others in the last 2 years. More users of supported housing (26%) and floating outreach (21%) had self-harmed within the last 2 years than users of residential care (4%). Risk of serious self-neglect was reported for 57% overall (72% for users of residential care services and at least 50% for users of the other two types of services). Vulnerability to exploitation was reported for around one-third of users in supported housing and floating outreach services and for 41% of those in residential care. Overall, 67% to 78% of service users across the three types of accommodation were considered a risk to themself or others.
There were few differences in social function and challenging behaviours between service users in the three types of accommodation, but those in residential care, unsurprisingly, had more needs than those in supported housing or floating outreach. However, there were few unmet needs across the three types of accommodation. Users of supported housing and floating outreach were more likely to report having been a victim of crime in the last 12 months than those in residential care (residential care 8%, supported housing 25%, floating outreach 22%) and around half of these incidents involved physical assault. Residential care service users had higher ratings of satisfaction with their safety than users of the other two types of service.
Regression analyses that took account of clustering by area and service showed lower quality-of-life scores for those in floating outreach and supported housing than for those in residential care services, but higher levels of autonomy in both supported housing and floating outreach than in residential care (see table 3 in Killaspy et al.22). In these analyses, the therapeutic milieu (which could not be assessed in floating outreach services) was rated higher by users of supported housing than by users of residential care [mean difference 0.973, 95% confidence interval (CI) 0.126 to 1.821; p = 0.024], yet there were no statistically significant differences between the three service types in terms of satisfaction with the care received.
Appendix 2, Tables 6 and 7, show the results of the regression analyses that investigated service and service user characteristics associated with service users’ quality of life, autonomy, ratings of the therapeutic milieu (residential care and supported housing only) and satisfaction with care. The QuIRC domains therapeutic environment and recovery-based practice were found to be highly correlated. It was decided to keep recovery-based practice in the models as this domain had been found to be a predictor of successful discharge from inpatient mental health rehabilitation units in a previous study.39 As data could not be collected for living environment domain scores for floating outreach services, this domain was also dropped from the regression models. The 19 variables included (see Appendix 2, Tables 6 and 7) were agreed by the PMG on the basis of those where differences were found in the summary statistics and where there was clinical justification.
In these analyses, which accounted for differences in service user characteristics, service users’ quality of life (MANSA) did not differ between supported housing and residential care but remained lower for those in floating outreach than for those in residential care. Supported housing was still predictive of higher autonomy (RCS) than residential care but floating outreach was not. In addition, positive associations were found between service users’ quality of life and the mental health morbidity/housing index of the local area, service user age, primary diagnosis (psychosis vs. non-psychosis) and problematic drug use. Negative associations were found with the number of places occupied per service, the adapted QuIRC recovery-based practice domain score and the number of unmet needs. However, the small size of the coefficients suggests that these associations had very little clinical impact. Service users’ autonomy was negatively associated with the QuIRC-SA domain treatments and interventions score and unmet needs, but, again, small coefficients suggest that these associations were of limited clinical significance. For each additional occupied place per service, there was a small reduction in service users’ ratings of satisfaction with care (CAT-SA) and the therapeutic milieu (GMI). The GMI ratings were positively associated with the QuIRC-SA human rights domain score, service users’ age, staff ratings of their social function (LSP) and risk history. Once again, small coefficients limit the clinical relevance of these findings.
Health economic component
Methods
The use of services was estimated from staff and service user interviews using an adapted version of the CSRI.35 Both health-care and social care costs were included in the analyses. Services included those provided externally as well as contacts with staff of the supported accommodation service. Participants provided information on whether or not specific professionals had been seen in the previous 3 months, how often they had been seen and whether or not this was in a group setting. Information relating to the previous 3 months was obtained on contacts with care co-ordinators, psychiatrists, other doctors, psychologists, community mental health nurses, occupational therapists, social workers, counsellors and art therapists. Contacts with staff of the supported accommodation service were broken down into face-to-face contacts, group sessions and personal care. It was assumed that group sessions consisted of four participants. In addition, details of admissions to hospital in the previous 12 months for mental health or physical health reasons were provided. Service costs were calculated by combining the service use data with appropriate unit cost information.40 Total costs of services used in the previous 3 months were calculated, as were total inpatient costs for the previous 12 months. An overall total for the previous year was calculated by multiplying the 3-month costs by 4 and adding to the inpatient costs. Comparisons in terms of service use and costs were made between the three groups (i.e. residential care, supported housing, floating outreach). Total cost differences were assessed using a mixed-effects multilevel regression model, controlling for clinical and demographic factors. To generate cost-effectiveness, planes we used a bootstrapped linear regression model for both the cost and the MANSA score, and controlled for area. One thousand cost-outcome differences between (1) residential care and supported housing and (2) residential care and floating outreach were generated and plotted.
Results
There were clear differences in the proportion of each group using specific services (see tables 4 and 5 and the appendix of Killaspy et al.22). Those in residential care were most likely to receive input from the supported accommodation staff through group sessions and to be in receipt of personal care. Those in supported housing and floating outreach had similar levels of face-to-face and group contacts with supported accommodation staff. Users of supported housing had the highest input from community team care co-ordinators (79% vs. 64% of residential care and 45% of floating outreach service users) and the highest rate of psychiatric admission (24% vs. 11% of residential care and 9% of floating outreach service users). Floating outreach service users generally had lower levels of service use than those in the other two groups. Of those who used specific services, the intensity of use did not differ markedly between the groups for most services. However, those in residential care had more nurse contacts, face-to-face sessions and personal care contacts than users of the other two services. They also had longer stays in hospital for mental health reasons, although this was influenced by some outliers.
The services with the highest costs were inpatient care and face-to-face contact with supported accommodation staff. Excluding inpatient care, the costs of service use during the previous 3 months were highest for users of residential care, followed by those receiving floating outreach. Inpatient costs were lowest in the floating outreach group and similar in the other two groups. This was also reflected in the total costs pertaining to a 1-year period such that total costs were highest for residential care service users and lowest for floating outreach service users.
The multilevel models showed that after making adjustment for demographic and clinical characteristics, the residential care group had costs that were, on average, £1483 more than for supported housing and £5381 more than for floating outreach. The average costs for supported housing were £3898 more than for floating outreach. We found that residential care service users had a quality-of-life (MANSA) score that was 0.274 points higher than users of supported housing and 0.722 higher than users of floating outreach, and those in supported housing had a score that was 0.449 points higher than for those in floating outreach (see table 3 in Killaspy et al.22). These figures suggest that it costs £5412 to achieve an extra 1-point improvement on the MANSA (representing a clinical improvement of > 14%) if residential care is chosen rather supported housing, £7453 if residential care is chosen rather than floating outreach and £8682 if supported housing is chosen rather than floating outreach. Appendix 2, Figure 4, shows the uncertainty around these point estimates and indicates that although residential care is most likely to generate higher costs and better outcomes (quality of life) than supported housing, there is still a reasonable probability of lower costs and better outcomes for supported housing. Appendix 2, Figures 5 and 6, show that residential care and supported housing are both highly likely to produce better outcomes but higher costs than floating outreach.
With regard to service users’ autonomy, both supported housing and floating outreach produced better outcomes than residential care and were less expensive. That is to say that they were ‘dominant’. Supported housing was more expensive than floating outreach but it was associated with greater autonomy, with a cost of £46,405 for every extra unit improvement in autonomy achieved.
Links to other work packages
The QuIRC-SA and CAT-SA, developed in WP1, were used to assess service quality and service user experiences in WP2i. Service users in supported housing and floating outreach had similar levels of risk, social functioning, challenging behaviours and autonomy, and received similar levels of input from their supported accommodation support workers, suggesting equipoise between the two service types, which supported the rationale for a comparison of these two service models in the feasibility trial (WP4).
Limitations
A number of limitations must be acknowledged. In spite of using a sampling strategy explicitly designed to minimise bias and produce a nationally representative sample, service users who declined to participate or who lacked capacity to consent may have introduced a sampling bias. In addition, our findings cannot be generalised to supported accommodation services and systems outside England. Finally, the cross-sectional nature of WP2i means that we cannot infer causality from our findings.
Key findings
- Compared with floating outreach, service users in residential care and supported housing had more severe mental health problems.
- Over half of all service users were at risk of serious self-neglect and over one-third had been vulnerable to exploitation over the previous 2 years.
- One-quarter of those in supported housing and one-fifth of those receiving floating outreach services had been the victim of crime in the last year, compared with 8% of those in residential care.
- Residential care was the most expensive service and floating outreach was the cheapest. Residential care and supported housing were both highly likely to produce better outcomes but had higher costs than floating outreach.
- As assessed by the QuIRC-SA, supported housing demonstrated the best quality of care compared with the other two service types.
- After adjusting for clinical differences between service users, quality of life was similar for those in residential care and supported housing, but lower for those in floating outreach. Autonomy was greater for those in supported housing than in residential care and floating outreach. Satisfaction with care was similar across service types.
Work package 2ii: cohort study
The aim of WP2ii was to assess the proportion of people in supported accommodation who successfully moved on to more independent accommodation over 30 months and to identify service and service user factors (including costs) associated with move-on.
The primary outcome, ‘successful move-on’, was defined as the proportion of participants in each service type who moved to more independent accommodation successfully without placement breakdown over the 30-month follow-up period. Because floating outreach is provided to people living in a permanent tenancy, for this service type, the primary outcome was defined as managing with fewer hours of support per week rather than moving home.
Our specific research questions were:
- What proportion moved on to more independent accommodation overall and by service type?
- What proportion moved on to more independent accommodation and sustained it for the 30-month follow-up (i.e. did not move back to more supported accommodation)?
- How much of the variation in outcome was due to service type and service quality (measured by the QuIRC-SA domains), before and after accounting for service user characteristics (age, sex, diagnosis, length of stay, morbidity)?
We also investigated a secondary outcome, defined as the proportion of participants who moved on to more independent accommodation and sustained it for 30 months (i.e. without placement breakdown, any move back to more supported accommodation or any hospital admission) overall and by service type.
To minimise the influence of time in service and clinical differences between service users in different types of supported accommodation, we planned to undertake further investigation of service user level outcomes for the subgroup of individuals in the main cohort who had been living in supported accommodation for < 9 months at recruitment. Nine months was chosen as a pragmatic balance between being relatively new to the service and being there long enough for staff and service users to be able to rate the various outcome measures. For this subgroup, we aimed to investigate the following research questions:
- Is there a difference in service users’ quality of life, autonomy and staff-rated social function at the30-month follow-up between users of different types of service, after adjusting for the baseline score of each outcome?
- Is there a difference in service users’ quality of life, autonomy and staff-rated social function at the30-month follow-up between users of different types of service after adjusting for baseline score, service quality and service user characteristics?
- How much of the variation in these outcomes is due to service type and quality, after accounting for service user characteristics?
- Are there differences in service use and costs at the 30-month follow-up?
Methods
Data collection
After participants were recruited in WP2i, the researchers maintained contact with the relevant service manager every 3 months to monitor whether or not the service user had moved on to less supported accommodation, moved to more supported accommodation or had any admission(s) to hospital. For those who had moved, details of the new supported accommodation service were obtained and contact was made with the new service staff. If the service user moved on to a fully independent accommodation, with no supported accommodation staff involvement, their care co-ordinator (when applicable) was contacted to provide ongoing monitoring of their accommodation status. This process was continued over the 30-month follow-up period.
At the 30-month follow-up point, the researchers completed telephone interviews with supported accommodation staff or care co-ordinators and confirmed details of any moves to alternative supported or independent accommodation during the 30-month period and the length of time in each accommodation. From these data, an overall assessment of whether or not the person had progressed successfully was made [i.e. moved to less supported accommodation (for those in floating outreach services, this was operationalised as having fewer hours of support per week) without placement breakdown].
If a relevant staff member could not be identified (e.g. if the service user had moved to a fully independent tenancy without floating outreach support and been discharged from mental health services), NHS case records were accessed to collect primary outcome data on move-on. For all participants who were reported to have had a hospital admission, case notes were also checked to clarify this. The researchers also completed the adapted version of the CSRI35 with the staff member, corroborating the length of any admissions from the case notes for the health economic analysis where possible.
For the cohort of participants who had been in their accommodation for < 9 months at recruitment, researchers conducted additional face-to-face interviews with service users and their support workers or care co-ordinators. If the service user had moved on to fully independent accommodation and had no contact with supported accommodation or mental health staff, only the service user interview could be completed. These interviews comprised the same measures completed at recruitment, assessing, (1) from staff interviews, social functioning (LSP),34 substance use (CADS),33 challenging behaviours (SPRS)31 and needs (CANSAS),32 and, (2) from service user interviews, quality of life (MANSA),29 social outcomes (SIX),36 autonomy (RCS),37 therapeutic milieu (GMI)38 and satisfaction with care (CAT-SA).21
Data analysis
Data were entered into a purpose designed database by the researchers. Data checks were completed on all records, comparing the recorded data with the data entered into the database. After cleaning, data were transferred to Stata statistical software for analysis. Descriptive analyses were conducted for all variables.
Primary outcome
For the primary outcome, a logistic mixed-effects model was fitted using xtmelogit, with a random intercept for service and a fixed effect for area as this was used in the sampling frame as a design variable. Univariate analysis was used to identify service and service user variables with a significant association (p < 10%) with the primary outcome for inclusion in multilevel models investigating predictors of the primary outcome. As the analysis of baseline data showed that the QuIRC-SA domains therapeutic environment and recovery-based practice were very highly correlated (Spearman’s rho = 0.87) and the variance inflation factor exceeded 10, it was decided to dispense with the therapeutic environment domain score as the recovery-based practice domain score had previously been shown to predict successful discharge from inpatient rehabilitation services.39 The QuIRC-SA domains included in the univariable analysis were therefore restricted to treatments and interventions, self-management and autonomy, social interface, human rights, and recovery-based practice. Living environment was excluded as it does not apply to floating outreach services.
Service user variables included in the univariate analysis included sociodemographic and clinical characteristics [age, sex, clinical history, diagnosis (non-psychotic vs. psychotic disorder), length of stay with service], social functioning (LSP), total unmet needs (CANSAS), substance misuse (CADS), challenging behaviours (SPRS), risk of self-neglect and/or vulnerability to exploitation, risk to others and risk of self-harm. Variables were examined for collinearity. Those that did not allow discrimination between service users because the counts were too high or too low were excluded.
Selected variables were included separately in the univariable model to ascertain whether or not there was a difference in successful move-on between service types after adjusting for these variables. Those that showed a significant association (p = 10%) were included in the multivariable model. In our multivariable models, we assumed the convention that at least 10–20 participants for each predictor should be entered.
Sensitivity analyses
To address factors that may have influenced our primary outcome (clinical profile of service users, geographical location of service, definition of primary outcome for floating outreach, service user time in service), the following sensitivity analyses were conducted on the primary outcome by service type:
- a propensity score analysis that collated an average treatment effect from the following variables: social function – LSP score,34 age, diagnosis of psychosis/no psychosis and a composite risk variable (vulnerability to risk of exploitation with or without risk to others, with or without self-harm in the last 2 years)
- excluding participants who did not have a diagnosis of psychosis
- replacing the geographical area variable with the geographic area sampling index score
- only categorising floating outreach service users as having a positive outcome if the number of hours per week of support had reduced by at least 50% since recruitment
- comparing service users who had been in the supported accommodation for < 9 months at recruitment with those who had been there for ≥ 9 months.
Secondary outcome
A logistic mixed-effects model was fitted using xtmelogit, with a random intercept for service and a fixed effect for area to assess the secondary outcome by service type.
Subgroup analysis
We planned to use multilevel models to investigate factors associated with service user ratings of quality of life, autonomy and satisfaction with care in the subcohort of participants who had been living in their accommodation for < 9 months at recruitment. However, the sample for whom data were available was too small and thus only descriptive data are presented.
Results
Participant flows in the cohort are shown in Figure 2. After accounting for withdrawals (n = 7) and deaths (n = 26), we followed up 586 out of 619 (95%) participants over 30 months (residential care n = 146; supported housing n = 244; floating outreach n = 196).
FIGURE 2 Participant flow during the prospective study. FO, floating outreach; RC, residential care; SH, supported housing; SU, service user. (Figure not reproduced here; see the original article.)
Descriptive data
Descriptive data for participants at follow-up by service type are shown in Appendix 3, Table 8. Overall, 110 out of 586 participants (18.8%) had a hospital admission during the 30-month period. Incidents of risk to others were highest among users of residential care (14% vs. 11.5% for users of supported housing and 4.1% for those in receipt of floating outreach). Episodes of self-harm were highest among users of supported housing (17.3% vs. 14.8% receiving floating outreach and 4.2% in residential care). Around one-third (30.5%) of supported housing service users who had not moved on were considered by staff as ready for move-on, compared with 8.5% of those in residential care and 6.9% of those receiving floating outreach.
Missing data
Overall, there were very little missing primary or secondary outcome data (see Appendix 3, Table 8).
Primary outcome
In total, 243 out of 586 participants (41.5%) achieved the primary outcome of successful move-on to less supported accommodation. The proportions achieving the primary outcome in residential care, supported housing and floating outreach were 15 out of 146 (10.3%), 96 out of 244 (39.3%) and 132 out of 196 (67.3%) participants, respectively. The unadjusted odds ratio (OR) of achieving the primary outcome for users of floating outreach compared with residential care was 28.81 (95% CI 11.53 to 72.02; p < 0.001). For floating outreach compared with supported housing service users, the OR was 5.11 (95% CI 2.47 to 10.57; p < 0.001). The unadjusted OR of achieving the primary outcome for users of supported housing versus residential care was 5.64 (95% CI 2.30 to 13.84; p < 0.001) (see Appendix 3, Table 9).
The univariable analysis identified positive associations with the primary outcome for the QuIRC-SA service quality domain scores for human rights (OR 1.09, 95% CI 1.02 to 1.16; p = 0.007) and recovery-based practice (OR 1.04, 95% CI 1.00 to 1.08; p = 0.054), as assessed at recruitment. The QuIRC-SA social interface domain score was negatively associated with the primary outcome (OR 0.95, 95% CI 0.91 to 0.98; p = 0.001). Service user total unmet needs, length of time in the supported accommodation service and a composite risk variable (risk of vulnerability to exploitation with or without risk of self-harm) at recruitment were also negatively associated with the primary outcome (see Appendix 3, Table 9).
After adjusting for these variables in the multilevel model, users of floating outreach services were almost eight times more likely to be managing with less support at follow-up than users of residential care (OR 7.96, 95% CI 2.92 to 21.69; p < 0.001) and almost three times more likely than supported housing service users (OR 2.74, 95% CI 1.01 to 7.41; p < 0.001). Users of supported housing were almost three times more likely to have moved on to less supported accommodation than those in residential care (OR 2.90, 95% CI 1.05 to 8.04; p = 0.04).
Sensitivity analyses
The results of the sensitivity analyses are shown in Appendix 3, Tables 10 and 11. All analyses showed a similar pattern of results to the main adjusted and unadjusted models, with higher odds of floating outreach service users achieving the primary outcome than users of supported housing and residential care, and higher odds of supported housing service users achieving the primary outcome than those in residential care.
Secondary outcome
Few (17/243, 7%) individuals who moved on had an admission after they had moved. This was most likely among supported housing service users, among whom 12 out of the 96 (12.5%) who moved on had a subsequent hospital admission [vs. none of the 15 who moved on from residential care and 5/132 (3.8%) of those who moved on from floating outreach]. The results of the analysis of our secondary outcome multivariable model are shown in Appendix 3, Table 12.
Subgroup analyses
The subgroup analyses results are shown in Appendix 3, Tables 13 and 14. The mean satisfaction with care (CAT-SA) scores were highest in residential care [mean 8.2, standard deviation (SD) 2.0]. There was little difference in mean quality-of-life (MANSA) and autonomy (RCS) scores between users of the three service types. Social function (LSP) was lowest in residential care (mean 118.3, SD 19.4) and highest in floating outreach (mean 127.2, SD 12.1).
Health economic component
Methods
Data on services used by residents were collected from staff at the 30-month follow-up with a short version of the CSRI. For the subcohort of participants who had been in the service for < 9 months at baseline, data were also collected from service users directly. Costs were calculated as described in Work package 2, Health economic component, and a similar multilevel model was used for analysing cost differences for the full sample. Information on inpatient use during the whole 30-month follow-up was available. We did not extrapolate the 3-month non-inpatient costs across the 30-month period.
The association between the primary outcome measure and costs was investigated in two ways. First, costs were compared for each group among those who achieved the outcome and those who did not. Second, the primary outcome variable was entered into the multilevel models to investigate the overall relationship with cost. We do need to be cautious in interpreting the results though because it is to be expected that those who do move to a lower level of care will have correspondingly lower costs. Adjusting for participant characteristics does allow us to quantify the impact more precisely.
Results
From the staff-reported service use information in Appendix 3, Table 15, it can be seen that supported housing service users were more likely to have had care co-ordinator contacts in the 3-month period up to the 30-month follow-up than users of residential care. Only one-quarter of floating outreach service users had these contacts. Contacts with psychiatrists and other doctors were relatively common, although less so for the floating outreach service users. Face-to-face and group contacts with supported accommodation staff were both most likely for the residential care participants, followed by those from supported housing and then those in the floating outreach. During the whole 30-month follow-up period, the supported housing service users were twice as likely as floating outreach service users to have had psychiatric inpatient episodes. There was little difference in the proportions having inpatient stays because of physical health problems between the three service types. Appendix 3, Table 16, reveals little difference in the intensity of service use among those actually in contact with services. The average number of face-to-face staff contacts with supported accommodation staff was higher for users of floating outreach. For those with admissions to hospital for psychiatric reasons, the number of days over the 30-month period was highest for the residential care service users. Excluding inpatient days, care costs were highest for personal care in residential care service users, followed by face-to-face contacts with supported accommodation staff and other doctor contacts (see Appendix 3, Table 17). Among supported housing service users, face-to-face contacts with supported accommodation staff had the highest costs followed by care co-ordinator costs. Face-to-face contacts with supported accommodation staff also had the highest costs for floating outreach service users. Total non-inpatient costs were, on average, around twice as high for residential care service users (£1434) compared with supported housing (£718) and floating outreach (£640) service users. After controlling for demographic and clinical variables in the multilevel regression model, users of residential care had costs that were, on average, £440 (95% CI –£245 to £1124) more than those for supported housing service users and £601 (95% CI –£54 to £1257) more than those for floating outreach service users. These CIs indicate that the cost differences were not statistically significant. Psychiatric inpatient costs were similar for users of residential care and supported housing, both about twice the cost of floating outreach service users. After controlling for demographic and clinical variables, residential care users had inpatient costs that were, on average, £5214 (95% CI –£2844 to £13,272) more than for supported housing and £7481 (95% CI –£210 to £15,172) more than for floating outreach service users.
Appendix 3, Table 18, shows the self-reported use of services at follow-up for those who had been living in their accommodation for < 9 months at baseline. Around two-thirds of residential care and supported housing service users had contacts with their care co-ordinators at some point in the previous 3 months, whereas only around one-quarter of floating outreach service users had such contacts. Most participants had contacts with psychiatrists, with a slightly higher proportion among floating outreach service users. About two-thirds of residential care and supported accommodation service users had face-to-face contacts with supported accommodation staff; far fewer in the floating outreach group had these. Around half and one-third of the residential care and supported housing service users, respectively, had group contacts, whereas none in the floating outreach group did. The average numbers of contacts for those with them show relatively similar numbers for the groups for most services (see Appendix 3, Table 19). There were noticeable differences for community mental health nurses and for personal care but the numbers receiving such care were very low. The highest service costs (excluding inpatient care) were for face-to-face contacts with supported accommodation staff, contacts with care co-ordinators and contacts with doctors other than psychiatrists (see Appendix 3, Table 20). Total mean non-inpatient costs were highest for supported housing service users (£649), followed by users of residential care (£512) and floating outreach (£398). We did not test these differences for statistical significance because of the small numbers in each group in this subcohort. Psychiatric inpatient costs were far higher among residential care service users than among the supported housing and floating outreach groups. Costs for inpatient care related to physical health problems were similar between the groups.
Across the full sample, those who achieved the primary outcome had mean (SD) costs at 30-month follow-up of £388 (£700), whereas those who did not achieve this had costs of £1214 (£2594). After adjustment, those who moved on to lower levels of care had costs that were, on average, £427 lower than those who did not – a difference that was statistically significant (95% CI £43 to £811). The mean (SD) inpatient costs for those who achieved the primary outcome were £2713 (£10,062) and for those who did not they were £15,142 (£40,463). The multilevel model revealed that inpatient costs for those who moved on were £14,608 (95% CI £8593 to £20,624) less than for those who did not. For the subcohort, the mean (SD) non-inpatient costs were £456 (£751) for those achieving the primary outcome and £626 (£573) for those who did not. The respective figures for inpatient costs were £2517 (£4705) and £10,732 (£41,776).
Appendix 3, Table 21, shows the costs for each of the three participant groups. In each case, there were cost savings for those who had moved to lower levels of support with the exception of inpatient care for the subcohort in the supported housing group.
Links to other work packages
WP2ii followed the cohort of participants recruited in WP2i and our analyses to identify predictors of our primary outcome used the data gathered at WP2i recruitment.
Limitations
Findings from WP2ii must be viewed in the light of a number of limitations. First, successful move-on (the primary outcome) for floating outreach service users was operationalised as having fewer hours of support per week than at baseline. Arguably, this is a lower threshold for ‘success’ than that applied to service users of residential care and supported housing and thus the higher ‘move-on’ we found for floating outreach service users may have been overestimated. Nevertheless, our sensitivity analysis that reclassified floating outreach service users as having a successful outcome only if the number of hours of support they were receiving had reduced by more than half found similar results. Second, as we were unable to monitor service users directly, it is possible that participants who moved on to full independence (discharged from floating outreach services) may have returned to some form of supported accommodation without the research team knowing. However, we attempted to address this through case note reviews of all participants at the 30-month follow-up. Third, follow-up data for some service users was collected through case notes only; although this allowed us to assess both the primary and secondary outcomes (a strength of our design), some other data, such as contact with family for our health economic analysis, could not be collected. Fourth, service use data provided by staff and participants may have be prone to recall error. However, the period of interest was 3 months and it is assumed that this is short enough to mitigate against this possibility. Fifth, it is possible that in some cases there may have been confusion over service definitions and double counting between care co-ordinator and support staff contacts. However, if this occurred it is unlikely to have been a major issue because care co-ordinator contacts were far less frequent than support staff contacts. Finally, owing to the small sample size for the < 9-month subcohort, we were unable to conduct inferential analyses for this group.
Key findings
- We achieved a high follow-up rate, collecting primary outcome data on 95% of participants at 30-month follow-up.
- Successful move-on was achieved by 15 out of 146 service users (10.3%) in residential care, 96 out of 244 service users (39.3%) in supported housing and 132 out of 196 service users (67.3%) from floating outreach.
- Successful move-on was associated with the service’s quality scores at recruitment; QuIRC-SA domain scores for human rights and recovery-based practice were positively associated with the primary outcome and QuIRC-SA social interface domain score was negatively associated with it.
- More unmet needs, greater risks and longer length of time in the service were negatively associated with participants moving on successfully.
- After adjusting for these variables, floating outreach service users were more likely than those in residential care (OR 7.96, 95% CI 2.92 to 21.69; p < 0.001) and supported housing (OR 2.74, 95% CI 1.01 to 7.41, p < 0.001) to achieve successful move-on. The adjusted OR of achieving successful move-on for supported housing service users compared with those in residential care was 2.90 (95% CI 1.05 to 8.04; p < 0.001).
- Very few service users had an admission after moving on.
- For service users who had been in their accommodation for < 9 months at recruitment, satisfaction with care (CAT-SA) was highest among those in residential care. This group also had the lowest social functioning. There was little difference in quality-of-life (MANSA) and autonomy (RCS) scores between users of the three service types in this subgroup.
- Service costs for those who had moved on were significantly lower than for those who did not.