Work package 4: BETTER PATHWAYS

Introduction

There has been a steady reduction in the number of acute psychiatric admission beds in many countries around the world. Since the 1960s it has become clear that with more effective pharmacological treatments, community-based alternatives to hospitalisation were possible and less costly, as well as often being preferred by service users and their families. Despite this shift towards community-based care, acute inpatient psychiatric services remain an indispensable element within an integrated mental health system. Indeed, acute wards consume the greatest proportion of the mental health budget and employ the greatest number of staff in the UK.6 However, this high investment does not always translate into better services. Average length of stay is higher than the Organisation for Economic Cooperation and Development average (21 days for mood disorder to 38 for psychosis171) and many inpatient units have significant problems with the provision of treatment and unacceptably high bed occupancy levels, which are often well over 85% and sometimes over 100%.172 Furthermore, a significant association has been shown between ward crowding and incidents of physical and verbal aggression.173

The traditional model of inpatient care in the UK involves service users being assigned to wards based on their geographical residence and bed availability, with no specific proposed limit on the duration of stay. The resulting case mix of acutely ill service users with others further along the pathway to recovery contributes to inefficiencies and delayed discharges. Pressures from managing newly admitted acutely ill service users also contributes to delayed discharge of those who need sustained therapeutic intervention, as the ward staff are spending time on bed management tasks as opposed to engaging with service users in potentially therapeutic communication or activities.117 Previous studies have also shown that 27.2% of inpatients were judged by ward staff as not requiring hospitalisation, with the figure increasing to 61% for individuals with stays of > 6 months.31 The main reason discharge was delayed in this study was the lack of suitable community accommodation or specialised rehabilitation services. Paradoxically, this system probably also contributes to the ‘revolving door’, as some service users (possibly those who have a home to go to and fewer social needs) experience repeated cycles of brief admissions with no resolution to their underlying problems.174

In this context, there have been many attempts to improve the efficiency of acute inpatient care and the care pathways that lead to it. Almost 40 years ago, Hirsch et al.175 showed that service users who were randomly allocated to a specially designed short admission programme improved as much as the comparison group, despite a 33% reduction in the length of their hospital stay. Subsequently, crisis resolution and HTTs have been developed in an attempt to prevent hospitalisation and to facilitate earlier discharge for cases in which admission was unavoidable. The results have been mixed with little reduction in inpatient bed use.176,177

In an attempt to relieve the overoccupancy that leads to poor service user (and nurse) experience, a novel ‘triage’ ward system has been developed. In a triage system, all acute admissions are directed to one inpatient facility that provides a brief period of assessment and rapid stabilisation of the acute episode. The triage ward has a capped length of stay (typically 7–10 days). Service users who are assessed as needing a longer period of inpatient treatment are subsequently transferred to other inpatient settings that offer more rehabilitative care. Early experience of this model has claimed considerable benefit both clinically and in terms of greater efficiency. In a London example, 406 service users were admitted to the ward over a 6-month period with 170 (42%) being discharged home within a week. The mean occupancy rate on the triage ward during that period was 70% of capacity, which even allowed for the temporary admission of service users from neighbouring catchment areas where no inpatient beds were available.178

Although the primary aim of the triage system is to improve efficiency, the separation of the immediate management of the acute episode from longer-term inpatient care may also offer some advantages in terms of the experience of the service users and nurses. Rapid crisis resolution and discharge home are likely to be preferred by most and, for those with longer stays, a quieter environment that can offer a ‘milieu’ approach might also be conducive to better service user and staff experience. On the other hand, the triage team will be under pressure to make fast discharge decisions and to deal with high levels of acute symptoms, all of which might lead to higher levels of staff burnout. Furthermore, staff in the longer-stay locality wards may be faced with the challenges of a service user group with longer-term, more challenging behaviours and consequent delays in discharge, all of which can be demoralising. Service users may dislike rapid moves between different parts of the service and having to form relationships with yet another clinical team. User and staff views of the triage system are therefore essential to assess benefits and potential disadvantages of this service model in addition to the hard data on duration of care and costs. Despite their appeal and gaining popularity and proliferation, ‘triage’ services have not been subject to robust investigation. It is also unclear whether or not, when compared with a system without triage, (1) the claimed advantage of the model is sustained in the face of wider pressures on efficiency, (2) it achieves the hoped for cost benefit or (3) the savings from those service users discharged within the 7- to 10-day target are offset by even longer stays in those service users who are transferred to less acute settings.

This WP investigated whether or not the triage system contributes to reduced lengths of hospital admission overall and investigated the impact that the triage system has on the experience of service users and staff. We were also interested in whether or not one system is associated with a frequently reported benefit to service users – greater service user/staff interaction – and what this may cost.

Method

The setting

The study was carried out in one large mental health organisation responsible for all mental health services. The defined geography covered by this health-care trust comprises four catchment areas, each of which has its own inpatient hospital provision prioritised for catchment area residents but service users from any area, including other areas in the UK, may be admitted. This WP focused on two of these areas, one of which operated a triage service.

Triage service

Triage services are located in an area of London with an adult population of approximately 275,000 and an average index of deprivation score of 30.97 (ranking 31st most deprived out of the 326 boroughs in England).163 The inpatient resource comprises 79 beds (four mixed-gender wards). There is also a PICU ward with 10 beds. The triage ward accepts all inpatient admissions and service users remain for a maximum of 7–10 days. All service users are expected to have senior medical input within 24 hours of admission and community mental health services are engaged in this process. The intensive assessment process allows for immediate discharge planning on the day of admission. Stabilisation takes place over 3 days and the HTT that is responsible for care in the community is consulted at days 3 and 4 and a decision to admit to longer-term care or home treatment is made with a resulting discharge or transfer by day 7. These assessments are carried out by health-care professionals with the ability to discharge or define the inpatient service need. There are three other longer-term treatment wards in this service system and which ward service users are admitted to depends on geography. This admission model is based on a holistic approach aiming for more efficient and continuous care, with the service user’s Community Mental Health Teams (CMHTs) and HTTs significantly involved in both the admission and discharge processes. In the whole period of acute care, the HTT acts as an extension of the inpatient services providing high levels of support immediately after discharge. All four wards are mixed gender and had an average of 18 beds. All the wards are housed in a single building.

Traditional service

The comparison routine service is an area well matched for population size and deprivation (300,000 residents) with an average index of deprivation score of 31.24 (ranking 29th out of 326 boroughs163). Beds are provided in five wards (one of which closed during the year), with 18 beds each, which accept acute admissions. There is also a PICU with 10 beds. Service users are admitted to a ward with an available bed for their whole stay. As in the triage system, there is involvement of the service user’s CMHT and a separate HTT that liaises with each of the wards to facilitate discharge. Two wards were mixed gender, one ward was female and two were male. One of the wards is a specialist early intervention unit for those having a first episode of psychosis and had a younger population than the other wards. Wards had an average of 18 beds. Initially wards were on three sites, but during the course of the study consolidated to a single site.

Two studies were carried out. One to explore the impact of a triage system on overall lengths of stay and the second partially nested within this: a survey of the views of service users and staff. These studies are reported separately.

Study 1: is a triage system more efficient than traditional care?179

Design

Data on all admissions and discharges were extracted from the comprehensive electronic patient record using the Clinical Records Interactive Search system. This has regulated access to anonymised information extracted from the trust electronic clinical records system which is updated daily. Data were extracted for dates of admission and discharge from all the wards in the two study areas and pathways on discharge, including any input from HTTs. Demographic and diagnostic characteristics of each admitted service user were also extracted. Two data sets were created to contain:

1. All admissions to the two systems between 1 January and 31 December 2009 for analyses of length of stay and costs. There were no admissions with lengths of stay that exceeded the date of data collection and so no censoring was necessary.
2. All re-admissions in the year following discharge from their first admission (i.e. index admission) in 2009. This allowed for a fair comparison by limiting the same period of time for re-admission to occur, regardless of the index admission date or the length of stay. There were no observation periods that exceeded the date of data collection.

Analyses

First, in order to determine whether or not there were differences between the inpatient systems we needed to identify whether or not the systems admitted different service user groups. This was investigated with t-test and chi-squared tests. Then we investigated whether or not the triage system discharged a greater proportion of patients back into the community within 10 days compared with routine care, that is, whether or not the triage did act as a fast throughput ward and, therefore, is different from the routine care system.

Second, we carried out an analysis of the length of stay for each admission as (1) length of inpatient care, the period from the night of admission to the day of discharge from the hospital, and (2) length of acute care, which includes any involvement of the HTT as a step down from hospital. As these analyses are by admission and some patients had multiple admissions during the period, all models accounted for non-independence of observations using random-effects linear regression including a random intercept at the patient level. Owing to the skewed distribution of length of stay, we used bootstrapping to estimate SEs and CIs. In addition, to adjust for any confounders as a result of differences in service user characteristics between the two areas, we constructed multivariable models using the same techniques and adjusted for service user characteristics that have been shown to be important in predicting length of stay in other studies.180,181

Third, we carried out an analysis of re-admissions to hospital. To estimate difference in odds of a re-admission (readmitted or not within 1 year) between the care systems we used a logistic regression model and added potential confounding variables using a stepwise selection technique with a threshold of p < 0.20. Categorical variables were tested using likelihood ratio tests.

Fourth, in an analysis of the costs of the two systems, we used the same statistical models and adjusted for the same confounders as for the length of stay analyses. We extracted data on the number of HTT visits that each service user received and multiplied the number of inpatient days and the number of HTT visits by unit cost figures from NHS reference costs182 (£304), face-to-face contact with a crisis resolution HTT (£179)182 (see Appendix 6). As it is a fairly common practice that some service users are granted periods of trial leave prior to final discharge, sensitivity analyses were conducted that repeated the multivariable analyses of length of stay and costs, but defined length of stay as the number of days spent physically present in hospital during an admission (‘bed-days’), that is, excluding any leave periods before official discharge. The cost sensitivity analysis was performed under the assumption that inpatient costs while the service user was on leave were negligible and, thus, the analysis applied the same cost estimates for inpatient care (£304). Percentile-based CIs are reported for all length of stay and cost analyses.

All analyses were carried out using Stata v12.1.

Study 2: service user and staff perceptions of care

These data were collected four times at 6-month intervals over an 18-month period (largely coterminous with the comprehensive admission census described above).

This series of cross-sectional samples and their timing was deliberately chosen because of concern that a single cross sectional assessment would be open to the effects of seasonal changes and rare events, such as unexpected organisational changes that could not be controlled for in the analysis.

We carried out two sets of comparisons. In the first we compared staff and service user perceptions in the triage system (triage ward plus remaining wards) with those in routine care. In the second set of comparisons, we investigated specific differences between types of ward.

Inclusion criteria

Service users were included if they could communicate in English, if they provided informed consent and if they had been on the ward for a minimum of 7 days. For triage participants, this minimum was 3 days so as to minimise the loss of those who were rapidly discharged to the community. There were no diagnostic exclusion criteria. We interviewed service users only once in the study. This excludes 10% of individuals who agreed to take part but were in hospital at more than one time period and tended to be those who had long admissions rather than repeat admissions.

Permanent nursing staff of any grade who were working on the wards at the time of recruitment were eligible to be included in the study. Temporary staff were also eligible if they had completed seven shifts in the previous month (to ensure that they were sufficiently familiar with the ward environment).

In order to have representative views, the aim was to recruit at least 50% of staff and service users who were on the ward at the times of data collection.

Ethics approval for the study was granted by Bexley and Greenwich Research Ethics Committee (reference 07/H0809/49).

Measures

Service user measures

1. VOICE:84 this was developed and validated as part of WP1. VOICE is easy to understand and complete and, therefore, is suitable for use by service users while in hospital. The 19-item measure has good validity and internal and test–retest reliability. A high score indicates a more negative perception.
2. SSS-RES:78 this is a derivative measure adapted from the SSS-3079 designed to evaluate residential services for people with serious mental illness. The original SSS-30 has been used in a variety of settings and demonstrates sound psychometric properties.164 Higher scores indicate less satisfaction with services.
3. Demographic and clinical information: for each service user participant we recorded age, gender, marital status, ethnicity, education, socioeconomic status, accommodation and employment status, diagnoses, admission date, legal status, length of stay, time on ward, diagnosis history and medication history. For nurse participants we recorded age, gender, ethnicity, level of employment band, length of employment and education, and employment in previous wards/trusts.
4. GAF:147 this is a 100-point rating scale of a service user’s overall level functioning that is based on observation, interview and medical records. Scores can range from 0 (lowest functioning) to 100 (highest functioning).
5. NOSIE:166 a 12-item scale designed for nurses largely focused on the assessment of socially unacceptable/unusual behaviour. Nurses were asked to rate participants on items such as ‘sits, unless directed into activity’.

Staff measures

1. VOTE:85 this is a 20-item measure that captures staff perceptions of the daily pressures of working in acute inpatient mental health wards. It was developed and validated in WP1. Each item is answered by choosing from a six-point Likert scale. All item scores are added together to form a total score. This score represents each staff participants’ view and high scores are negative. Test–retest reliability and the internal consistency of the measure and subscales were good.
2. MBI:49 this can be totalled to produce a score that measures work-related ‘burnout’ over 22 items, answerable by a six-point Likert scale. High scores indicate higher levels of burnout. The MBI comprises three individual subscales: emotional exhaustion (high scores mean high burnout), depersonalisation (high scores indicate high burnout) and personal accomplishment (low scores indicate high burnout). It has good psychometric properties and is widely used.
3. IWS:81 this is a 44-item scale that can be totalled to produce a score that measures health professionals’ levels of job satisfaction. It is answerable by a seven-point Likert scale.

Activities and costs

CITRINE:142 this was developed and validated as part of WP2. This instrument records activity data on psychiatric inpatient wards including ward activities and time spent with health professionals that is perceived to be meaningful by the service user. In combination with data on the unit cost of staff time, this can be used to calculate the cost of such perceived contacts.136 For more detailed explanation of the costing approach, see Chapter 3.

Analyses

There were four objectives:

1. To investigate whether or not service users’ perceptions of the therapeutic environment (VOICE) and service users’ satisfaction with services (SSS-RES) differed between the two systems.
2. To investigate whether or not staff perceptions of the therapeutic environment (VOTE), staff burnout (MBI) and work satisfaction (IWS) differed between the two systems.
3. To establish whether or not the perceived amount of care received and its cost (CITRINE) differed between the routine and triage systems and to establish the cost-effectiveness of the triage system according to the perception of service users.
4. To establish whether or not our sample was representative of all the service users admitted to inpatient services in the two catchment areas.

For all of these objectives we carried out two sets of analyses. The first identified the key question: does the triage system as a whole produce any advantages or disadvantages? The second set explored three groups of wards: the triage, the remaining wards in the triage system and the routine system wards.

The service user outcomes and the cost of service use were modelled using linear regression in models that included phase of sampling as a covariate. Phase included as a trend was preferred if a likelihood ratio test between the two models provided no evidence that the categorical variable provided a better fit for the data, and phase by system interactions were tested in the same way. When interactions were significant, they are presented as the effect on phase compared with the baseline (reference) phase for each system. Sensitivity analyses were performed using mixed-effects regression models to take into account the clustering of the service users by sampling ward. In the cost regression, Eicker–Huber–White SEs were calculated to allow for non-normality of residuals. In addition, regression estimates were weighted by the ward-specific inverse probability weight (i.e. the inverse of the proportion of service users admitted to each ward that were contained in the sample). In the adjusted cost analyses, a number of potential confounders were controlled for. The choice of these variables was based on a previous study by Sabes-Figuera et al.142

For staff measures, when some staff members were interviewed more than once, a random-effects regression was fitted. Phase, and phase by system, covariates were tested using the same methods as in the service user models. Ward was included as an additional random effect in sensitivity analysis. The different exclusion criteria per system (length of stay < 7 days, length of stay < 3 days in the triage system), as defined in our protocol, resulted in the inclusion of seven participants (3%) in the triage system who would have been excluded from the routine system. We performed each model excluding these seven participants to test the influence on the findings and found that none of the results changed.

Exploratory analyses included a comparison between three categories of ward: triage ward, non-triage wards in the triage system and wards in routine care. This new three-level categorisation of system provided us with the specific contrasts of interest: (1) the triage ward versus the remaining non-triage wards in the triage system and (2) the contrast between routine system wards and the non-triage wards in the triage system. In the service user models, these contrasts were estimated with and without indicators of service user functioning on the ward to adjust for any differences between the systems.

To explore the relationship of the differences in costs with the differences in benefits between the two systems, we constructed two cost-effectiveness planes using bootstrap techniques. The first plot showed the joint distribution of mean costs and VOICE differences using the patient-level data. To obtain the second plot, we simulated the distribution of mean cost and VOTE differences using ward-level averages. In addition, we calculated cost-effective acceptability curves to show the probability that the net benefit of the triage system was higher than the routine care system at different levels of willingness to pay to improve VOICE and VOTE scores, respectively.

Results

Study 1: is the triage system more efficient than traditional care?

There were a total of 935 admissions to the triage system and 899 to routine care over our census year beginning on 1 January 2009. There were few differences and none of the differences between the characteristics of people admitted in the two systems reached statistical significance (Table 30). Psychosis accounted for the largest diagnostic group in both systems and around 40% were involuntary admissions. For a fair comparison of the two systems, all admissions were analysed regardless of whether or not service users were resident in the catchment area. Admissions from individuals recorded to be homeless were assumed to be inside the catchment area. In both areas, a similar proportion of all admissions were for ‘out of area’ residents (routine care 21% and triage 24%).

Are there differences in length of stay between the two systems?

Although more admissions to the triage system had rapid discharge, both systems have service users with lengthy admissions. A univariable analysis showed that although the length of inpatient stay was, on average, lower in the triage system, the difference was not statistically significant (mean difference –3.87 days, 95% CI –12.20 to 4.47 days; p = 0.364). A multivariable analysis (Table 31) taking account of service user characteristics also failed to provide evidence of a difference between the systems (mean difference 0.53 days, 95% CI –9.77 to 7.21 days).

The use of HTT also differed between systems with a greater proportion of service users from triage followed up by the HTT (32%) at discharge than from the routine care system (26%, OR 1.50, 95% CI 1.12 to 2.01; p = 0.006). Repeating the multivariable analysis of length of stay but including the additional days spent with HTT showed that the mean length of acute care is actually somewhat longer in the triage system (mean difference 7.83 days, 95% CI –1.62 to 15.63 days; p = 0.076).

Figure 11 shows probability density distributions of length of inpatient stay with and without involvement of the HTT in both systems. It is apparent that more admissions to the Triage system resulted in a rapid discharge but both systems had a long ‘tail’ of admissions that were in excess of 60 days. The use of HTT differed. In the routine care system, HTT duration was largely focused around one to two weeks following discharge from hospital whereas in the Triage system service users generally had longer contact with the HTT. When the two components were considered together, as a period of acute care, there were no differences between the systems. Looking in detail, we noticed that for the acute care episodes of short duration, both systems used a similar percentage of days on home treatment to days in inpatient care. However, for the longer care episodes we noticed that the Triage system discharges individuals to HTT care for longer.

FIGURE 11 Length of care components (WP4). (a) Inpatient stay; (b) home treatment; (c) acute care; and (d) inpatient care/acute care. a, Probability density functions estimated using a kernel-smoothing function calculated on full sample, distribution censored at 100 days; and b, locally weighted scatterplot smoothing curves, censored at 100 days. Acute care = length of stay including any use of Home Treatment. Acute care admissions lasting ≤ 100 days accounted for 85% of the full sample. Reproduced from Efficacy of a triage system to reduce length of hospital stay. Williams P, Csipke E, Rose D, Koeser L, McCrone P, Tulloch AD, et al. Brit J Psychiat 2014;204:480–5.179 Reproduced with permission of Royal College of Psychiatrists via PLSClear. (Figure not reproduced here; see the original article.)

A sensitivity analysis that excludes the use of leave periods (where a service user is allowed home temporarily and his/her bed re-used) from the measure of length of stay, shows the average number of days spent on the ward was lower in the Triage system (mean difference: –8.36 days, 95% CI –16.54 to –1.51; p = 0.030).

About one in three of the admissions in each system resulted in a re-admission within 1 year of discharge. A multivariable model showed no significant difference in the probability of re-admission between the two systems (odds of re-admission: triage vs. routine = 1.05, 95% CI 0.77 to 1.44; p = 0.758) (Table 32). Results from the univariable analysis were (odds of re-admission: triage vs. routine = 0.90, 95% CI 0.80 to 1.22; p = 0.925).

Do costs differ between the two systems?

Similar to the analysis of length of stay, our exploratory cost analysis suggested that there were no statistically significant differences between the two systems. The mean cost of an acute care episode (inpatient length of stay and number of home treatment visits) was £15,233 in the Triage system compared with £15,476 in the routine system. Adjusting for potential confounders, the mean difference in the cost of acute care for service users treated within their catchment area was estimated to be higher in the triage system (cost difference £391, 95% CI –£2535 to £2748; p = 0.77).

Study 2: the experience of service users and staff

Four hundred and fifty-four service users were recruited, which is 61% of all those eligible to take part. They were mostly single men of non-white ethnicity, aged, on average, 40 years and who were unemployed. Most had experienced a previous admission with a median of 4 years since their first admission, and the most common diagnosis was psychosis (56%). Mean NOSIE and GAF scores were similar between systems. The mean NOSIE score of the sample was 16.0 and the mean GAF (level and symptom scores) were 43.3 and 42.2, respectively (Table 33).

Are service user participants representative of those using inpatient care?

We also utilised available anonymous data of the inpatient population on the same wards and over the same period. A comparison of the clinical and demographic profiles demonstrated that they were very similar (see Table 33). We excluded individuals with very short lengths of stay in the current sample (< 3 days) and so the length of stay sample will not mimic that reported for the total population but with that exclusion there were no major differences between the triage and routine care systems in admission duration.

Staff

A total of 484 observations were collected from 284 different staff members during the four phases of the study. This represents, on average, 57% of nursing staff who were eligible to take part at each time point. Most were women, of non-white ethnicity, aged 36 years on average. They had been employed for approximately 20 months on wards in the routine care system and 18 months in the triage system. Most staff were qualified nurses in both systems. It was not possible to collect data on those nurses employed on the wards who did not agree to participate.

Do experiences differ between the triage and routine care systems?

Service users

The detailed results for these analyses are shown in Table 34. No significant difference was found in perceptions of the ward between the two systems (VOICE scores adjusted mean difference –0.77, 95% CI –4.44 to 2.90; p = 0.68). However, this analysis demonstrated that perceptions of wards deteriorated over time. For every increase in phase (6-month interval), the mean VOICE score for the whole sample increased (coefficient 2.72, 95% CI 1.00 to 4.43; p < 0.005). This result remained significant after adjusting the model for clustering. The results remained stable in the two sensitivity analyses adjusting for the number of days spent on the ward until interview. The exploratory model that separated the triage ward from the remaining triage system wards also found no evidence of differences between the three types of ward.

No significant differences or associations were found in satisfaction between triage and routine care or between routine care and the remaining wards in the triage system (Figure 12). However, there was a trend towards lower satisfaction in the triage ward compared with the remaining locality wards (mean difference –6.91, 95% CI –14.66 to 0.84; p = 0.08). This remained stable in the two sensitivity analyses adjusting for the number of days spent on the ward until interview. A trend of deteriorating satisfaction with phase of the study was also observed (p = 0.07).

FIGURE 12 Service user VOICE for each system by phase (WP4) by (a) system; and (b) system and ward type. (Figure not reproduced here; see the original article.)

One potential explanation for the deteriorating service user perceptions is that we also found a trend of decreasing length of stay with phase. As the phase increased by one unit, the average total length of stay for participants decreased by 16.16 days (coefficient –16.16, 95% CI –26.84 to –5.49; p = 003). This finding remained in the sensitivity analysis which included clustering by ward (Figure 13 and Table 35).

FIGURE 13 Service user SSS-RES for each system by phase (WP4) by (a) system; and (b) system and ward type. (Figure not reproduced here; see the original article.)

Table 36 provides results of a number of exploratory analyses of the potential predictors presumed to be associated with service user experience (VOICE) from prior research. Worse service user experience (i.e. higher VOICE scores) was associated with higher skilled worker groups, increasing severity of symptoms and those who were admitted for compulsory treatment. A strong association with the number of contacts service users make with staff and better service user experience is associated with more staff contacts. The trend with phase and with severity of symptoms remained significant in this adjusted model.

Nursing staff

The same models were examined for staff but in these models we used time as a categorical variable. Interactions between system and phase were significant for VOTE [χ²(3) = 11.03; p = 0.01] and for MBI [χ²(3) = 8.89; p = 0.03] but not for the IWS model [χ²(3) = 5.80; p = 0.12].

There was no significant difference in staff experience at baseline between the two systems (adjusted mean difference: 1.68, 95% CI –2.05 to 5.41; p = 0.38) but staff experience in both systems changed over the period of the study and these changes differed between systems. In the routine care system, there was a deterioration in the VOTE measure at 6 months (in phase 1) (mean difference: 6.58, 95% CI 3.83 to 9.33; p < 0.001), but this was not seen in the triage system (mean difference 0.94; p = 0.503) and the difference between the two systems was significant (mean difference –5.64, 95% CI –9.53 to –1.75; p = 0.005). By the end of the study (phase 3), there was evidence of deterioration compared with baseline in both the routine care system (mean difference 4.80, 95% CI 1.30 to 8.30; p = 0.007) and the triage system (mean difference 4.78, 95% CI 1.93 to 7.63; p = 0.001) and no evidence that this was a differential change (p = 0.994). These results remained in a sensitivity analysis adjusting for clustering by ward (Figure 14 and Table 37).

FIGURE 14 Staff outcomes for each system by phase (WP4). (a) VOTE; (b) MBI; and (c) IWS. a, Phases in which difference between systems was statistically significant. (Figure not reproduced here; see the original article.)

Similar results were seen for the other staff measures. Staff ‘burnout’ as measured by the MBI in the triage system appeared to be relatively stable throughout the study timeframe but an exploratory analysis separating the triage ward from the locality wards in that system showed that burnout was highest on the locality wards, even when compared with staff in the routine care system (mean difference 3.76, 95% CI –7.28 to –0.25; p = 0.04). Staff were also more dissatisfied in the triage system than in the routine care system (IWS mean difference 8.06, 95% CI 0.95 to 15.16; p < 0.05) and this increased over the study for both systems, with the most dissatisfaction in the final phase compared with baseline (mean difference 11.90, 95% CI 6.36 to 17.43; p < 0.001). The largest difference appeared to be between the routine care system and the triage locality wards (mean difference 9.03, 95% CI 1.19 to 16.87; p = 0.03) that had the lowest level of satisfaction. No evidence was found for a difference between the routine care system and the triage ward itself (mean difference 5.77, 95% CI –4.82 to 16.37; p = 0.29).

What services do people receive in the two systems, what are their costs and what is the cost-effectiveness?

Despite similar levels of nurse staffing, service users in the triage systems reported about half the number of contacts with nurses than those in the routine care system (Table 38). The cost of meaningful contacts with OT and activities was also found to be statistically significantly lower in the triage system. The total costs of staff contacts were estimated to be £41 lower in the triage system than the routine care system in the unadjusted analysis (95% CI 7 to 75), but this estimate was highly unstable over time and between wards. Figure 15 shows the development of the costs of perceived service use across the study. There was a marked reduction in staff contacts in the routine care system; therefore, after adjusting for potential confounders, the triage system was estimated to have lower costs in the first two phases of the study only. On the other hand, at phase 2, costs were found to be statistically significantly higher in the triage system. Two units (an early intervention ward and one of the acute wards) in the routine care system had notably higher estimated mean costs than the other wards (£294 and £305 vs. £159). Therefore, the use of cluster robust SEs increased the width of CIs considerably, such that differences were no longer statistically significant in the unadjusted comparison. In the adjusted comparison the costs of service use in the triage system remained statistically significantly lower at phase 0 and higher at phase 2 than the routine care system.

FIGURE 15 Cost of perceived service use (WP4). a, Phases in which difference between systems is statistically significant after adjusting for confounders. (Figure not reproduced here; see the original article.)

Within the triage system, service users on the triage ward were significantly less likely to take part in activities and have contact with other care professionals. The overall cost of perceived service use was marginally lower in the triage ward than other wards in the triage system (mean difference –17, 95% CI –62 to 27), but this difference was not statistically significant and equally unstable across the different phases of the study as in the comparison of the two systems. The cost of meaningful contacts was statistically lower in the triage system at phase 3, but the mean estimate was not consistently lower across the study. In summary, although there was some indication that service use patterns in the two systems might be different, the results of the service use analysis were not robust and these differences may be attributable to other factors. Similarly, our data suggested that there were no substantial differences between the overall cost of perceived care received in the triage ward and long-stay wards in the triage system.

The cost-effectiveness analysis also did not provide clear support for either of the two systems. Using VOICE, there was a high probability that the triage system was more cost-effective (Figure 16) at any willingness to pay and that it would lead to both a decrease in costs and improvement in service user satisfaction (Figure 17). However, the aforementioned lack of robustness in mean cost estimates over the time horizon of the study should be kept in mind. In addition, in the case of VOTE, the analysis suggested that the triage system would probably lead to a deterioration in staff perception (67%) (Figure 18). Therefore, only if little value was placed on this outcome measure and it was the only outcome measure would the triage system be cost-effective (Figure 19).

FIGURE 16 Cost-effectiveness plane using VOICE with 95% confidence ellipse (WP4). (Figure not reproduced here; see the original article.)

FIGURE 17 Cost-effectiveness acceptability curve using VOICE (WP4). (Figure not reproduced here; see the original article.)

FIGURE 18 Cost-effectiveness acceptability curve using VOTE (WP4). (Figure not reproduced here; see the original article.)

FIGURE 19 Cost-effectiveness plane using VOTE with 95% confidence ellipse (WP4). (Figure not reproduced here; see the original article.)

Discussion

Strengths and limitations

Triage systems are one attempt to improve the efficiency and efficacy of acute inpatient psychiatric services and the current study is, to our knowledge, the first attempt at a rigorous evaluation of the approach. Our data included a comprehensive account of all admissions and discharges across two similar geographical catchment areas and linked this with a contemporaneous assessment of service user and staff perceptions of the quality of inpatient care in both systems. A limitation of our study is that it describes just one triage system in one inner city location provided by one NHS trust at one period in history. With these limitations in mind, the results of our investigation suggest that although a triage ward does achieve a rapid turnover of acute cases, this has little effect on the overall length of stay across the system and, in our data at least, a triage system cannot be said to provide a clear advantage over traditional care. There are three possible explanations for our observations. First, and most plausible, is the continuing pressure to improve efficiency in routine care that has made traditional catchment area services more like ‘triage’ than they might have been had we conducted the study 10 years ago when the triage system was first promoted. Second, it may be that the major contribution to length of stay has less to do with the vigour or success of crisis management and more to do with delayed discharge at the ‘other end’ of the system, with substantial numbers of service users remaining in hospital because of failures in the delivery of community and social care including access to adequately supported accommodation. Finally, there is the possibility that triage wards encourage admissions of some service users who would not be hospitalised in a routine care service and that our service user data were insufficiently sensitive to pick this up.

Although it could be expected that service users might express a preference for one system over the other for a variety of reasons, we did not find this to be the case. Even when accounting for other variables (e.g. socioeconomic status) that are associated with differences in perceptions, there is no significant difference in perceptions of care by service users being treated in the two systems. Even when looking at the triage ward on its own, service users did not view it differently from the other wards. The only change in service users’ perceptions and satisfaction was a deterioration in these measures over time that occurred in both systems and coincided with a move to shorter lengths of stay and a considerable reduction in the budgets of both systems, including a ward closure in the routine care system.

The perceptions and burnout of nursing staff did not differ between the two systems, although both got worse over time, which mirrored service user perceptions. Although we suggested that working on a high-pressure triage ward would increase burnout and dissatisfaction, we did not support this hypothesis. In fact, when looking at the wards themselves, a difference did emerge. Burnout scores were highest on the triage system’s locality wards compared with the triage ward itself and the routine care wards, and the triage system had less satisfied nurses than routine care. When teased apart into types of ward, nurses on the locality wards in the triage system had the lowest satisfaction. This finding can be explained as the consequence of the triage ward system, as the service users with the more persistent and challenging behaviours tend to concentrate over time on the locality wards, particularly if community residential alternatives are slow coming forward or not available. As noted in our introduction, staff in acute wards often report that as many as one-third of the service users have delayed discharges because of failures in ‘social’ care, and the protracted stay contributes to both staff and service user dissatisfaction. This may be compounded in a triage system as the triage ward is perceived not just as the innovation attracting interest and attention but also as a setting in which there is a steady turnover of service users with none of the problems of long-stay service users.

In terms of the costs, the analyses suggest that the cost of meaningful staff contacts might be lower in the triage system, however, this was not significant. Similarly, there was no clear evidence for differences between the triage ward and long-stay wards in the triage system in terms of activity costs.

A striking and unexpected finding in the study was that both service users and staff reported a worsening in satisfaction with inpatient services across the period of our research regardless of whether or not they were managed in triage or routine care. Sadly, these data chime with other reports from elsewhere in the NHS that suggest a more pervasive influence on staff morale and patient experience working across the wider organisation in recent years.18,170 They are also surely not the only data to demonstrate this decline in qualitative experience coinciding with a period of budget tightening and system pressure to increase efficiency that was reflected in our data by a universal reduction in length of hospitalisation and continuing very high levels of occupancy on the wards.

This issue was in the news as the then Deputy Prime Minister, Nick Clegg, made a speech on 20 January 2014,185 alongside a publication by the Department of Health,3 promising action to direct attention at improving the quality of care in our hospitals that should certainly improve the service user and carer experience. At the same time, the Guardian4 also ran a feature describing the harmful effects of bed closures on inpatient services from the point of view of a former psychiatric nurse revisiting his previous place of employment and finding it much deteriorated from a decade ago.

Clinical implications

Although the results of the study do not suggest any benefit of triage system over routine care, it would be wrong to conclude that the approach does not have some merit or may not be worthy of further study. Our data show that, when considered in isolation, the triage ward is able to discharge a greater proportion of hospitalised service users in the 7- to 10-day period than is achieved by standard care and there is no apparent increase in re-admission of these patients as was thought likely. Furthermore, contrary to expectations, staff in the triage ward were no more likely to report burnout than staff in routine care. What is clear is that there may be consequences for staff morale and satisfaction on the remaining wards in such a care pathway and this may require special attention.

Implications for future research

As noted in the section Strengths and limitations, this is a study of one triage system in one area of the inner city in London. This was the only service available at the time. Since then, triage systems have been adopted in several areas and are also contained in our own data in WP3. Future research would benefit from the inclusion of more of these new sites and perhaps a more detailed appraisal of the clinical characteristics of service users as there are anecdotal concerns that triage might encourage hospitalisation of service users who would otherwise be managed as effectively by crisis HTTs.

In conclusion, our study suggests that introducing triage wards is not enough to address the wider challenge of shortening the length of hospitalisation, as this is also determined by the availability of social care in the community, including access to appropriate housing and community support, much of the provision of which lies outside the control of the NHS.

Conclusion

The study demonstrated that, for an inner city acute service, triage wards confer no specific benefit to length of stay and they are not preferred by either staff or service users. Contrary to our hypothesis, triage ward staff did not exhibit more detrimental effects than staff on other wards. There is also some limited evidence that nurses on wards that are part of the triage pathway experience lower morale and less satisfaction with their role.184