Economic analysis

Overall approach and aims

The objective of the health economic analysis is to bring together findings from earlier chapters to identify potential impacts on the local health economy in each study area following the introduction of PDUs. The analysis makes use of the findings of all earlier chapters in this report. Using these different types of information, the chapter describes the economic impact and the RoI, from an NHS perspective, in each of the different study site areas. This compares the costs associated with PDU implementation with subsequent cost offsets as well as additional costs incurred. In addition, we also look at some impacts at an individual service user level; individual service scenarios provide illustrative estimates of changes in quality of life that would be needed for PDUs to be considered cost-effective from an NHS perspective.

All costs are reported in 2019–20 prices, with costs for acute hospital-based contacts and specialists community mental health teams taken from national reference costs.83 Costs for some psychiatric inpatient stays, as well as hourly costs for some additional community and hospital-based staff costs, are taken from the annual Unit Costs of Health and Social Care.84 These staff costs, estimate hourly costs, using mean full-time equivalent basic salary for AfC bands 4–9 of the April 2019/March 2020 NHS staff earnings estimates and include salary overheads and other oncosts. Further information on unit costs used in this chapter and their sources are reported in Appendix 8, Table 55.

Estimating the costs of psychiatric decision units

Our service mapping review (see Chapter 3), where the resources required to implement PDUs in the different areas have been collected,82 has been used to estimate the costs of providing PDU services in each site. An online survey completed by senior staff within PDUs included basic features on the structure of each unit, capacity, average and maximum LOS. The survey also collected information on staffing levels for both the day and night shifts in each PDU. This included type and number of staff members as well as their NHS AfC pay bands. We used these data to estimate the annual costs of running each PDU service. We assumed that there were two shifts, each of 12 hours in length and that the service operated 365 days per year.

The cost per PDU visit varies considerably from £741 in BSMHFT to £4800 in SHSCFT, with cost per PDU service user also ranging between £996 and £7442 (see Table 21). This is partly due to the considerable variation in staff ratios across the four MHTs, as well as differences in the level of throughput. BSMHFT, which has very short average stays of 12 hours, has an average of 1253 PDU visits for 932 service users per annum, compared with 438 visits by 283 service users in STHFT.

Area-level economic impact

The ITS reported in Chapter 4 and the synthetic control study in Chapter 5 have examined the impact of PDUs on area-level acute hospital ED attendances, as well as area-level psychiatric hospital admissions (both informal and informal) at a (clinical) population-level. In this section, using the results of these analyses we estimate economic impacts from an NHS perspective that may be associated with the implementation of PDUs at an area level. Chapters 4 and 5 set out in detail the methods used and results of the two approaches. As noted in Chapter 4, the ITS provides data on service use over a 4-year period (spanning 2 years prior to and 2 years post PDU implementation) in MHTs and acute hospital EDs in three of the four sites. The synthetic control study (described in Chapter 5) sought to match and compare mental health admissions and LOS, as well as ED attendances, for the mental health and acute trusts in our four study areas with trusts with similar characteristics in areas of England where PDUs have not been implemented.

While there are some similarities in the conclusions reached from each method there are also differences in findings. In this section, we look at potential impacts of using both of these approaches independently.

Interrupted time series analysis

For the ITS, we only considered changes in the pre- to post-PDU long-term trend rather than short-term step change impacts or just post-PDU impacts. Conservatively, in our analyses we also only assumed that there may be an economic impact where statistically significant changes in service use were seen; we did not look at the economic impacts of consistent but not significant trends. As noted in Chapter 4, insufficient data were available from SHSCFT to conduct an ITS, so ITS impacts on MHTs was restricted to three areas: BSMHFT, LPFT and SWLSTG.

Impacts on mental health trusts: interrupted time series analysis

We start by looking at the impacts on overall inpatient admissions, both informal and formal, into study MHTs. All these impacts were calculated from a segmented Poisson regression (GLM with log link) and therefore can be interpreted as a percentage change in admission/attendance frequency on PDU implementation per week (trend change). The ITS revealed that there was no significant impact on admissions at BSMHFT, but there was a highly significant change in the pre- to post-PDU (long-term) trend for all inpatient admissions at SWLSTG, with a decrease of 0.28%/week over the 2 years post PDU. A highly significant change in the pre-to post-PDU (long-term) trend of 0.45%/week for all inpatient admissions was also seen at LPFT. These modest impacts on inpatient admissions would equate to annual costs averted of £32,405 for LPFT and £47,919 at SWLSTG. We assumed that the LOS for each averted inpatient admission would be the median LOS, given that there were no significant changes in lengths of stay in any of the three sites.

Data were also available on liaison psychiatry contacts. For all service users no significant changes in pre-to post-PDU (long-term) trends were observed, but when focused solely on service users who had a previous episode in the last 24 months the ITS indicated a highly significant 2.79% monthly reduction in contacts in LPFT and a smaller but still highly significant reduction of 0.28% contacts in BSMHFT. This would be equivalent to small annual costs averted of £4739 and £2630 in the two areas, respectively. Overall, at site level the ITS suggests that all MHTs (BSMHFT, LPFT and SWLSTG) averted some costs following PDU implementation of £2630, £37,144 and £47,919, respectively.

However, these estimates of averted costs would almost entirely be no longer be seen if data across all three sites were pooled. As reported in Chapter 4, after adjustment for random effects, this meta-analysis indicated no significant impacts on inpatient admissions or LOS and hence no impact on costs. The meta-analysis did, though, find a highly significant but modest reduction of 0.24% weekly contacts with psychiatric liaison services for service users who had experienced a prior episode for poor mental health in the previous 24 months. This would be equivalent to a small annual saving across all three sites of £3765.

Impacts on acute trusts: interrupted time series analysis

For pre-to post-PDU (long-term) trends, the ITS analysis reported significant increases in weekly accident and ED attendance at ULHFT and STHFT of 0.26% and 0.46%, respectively. This is equivalent to an additional 4.67 and 9.32 attendances at accident and EDs per annum at a total cost of £880 and £1755, respectively. There was no significant change at SGUHFT, while at SWBHFT there was a significant fall in ED attendances for service users who had a previous admission in the previous 24 months. This was equivalent to an annual reduction of 4.75 in accident and ED presentations, with total costs averted of £895. Only one of the trusts, STHFT, saw a significant change in arrivals by ambulance or police; this increased by 0.23% points per week. This is equivalent to 4.66 additional ambulance visits per annum at a cost of £1362.

At ULHFT, SGUHFT and STHFT, there were highly significant increases from pre-to post-PDU in the proportion of attendances where service users were admitted from the accident and ED to an inpatient ward at the acute trust of 0.12% points, 0.13% points and 0.21% points per week, respectively. This leads to an additional 2.15, 2.64 and 4.26 inpatient admissions per annum at a total cost of £1598, £1962 and £3161, respectively, using the assumption that a typical inpatient stay is related to the weighted average of all NHS tariffs for short-stay non-elective treatment for the most common form of deliberate self-harm, poisoning.

Meta-analysis, pooling data across all four site areas, after adjustment for random effects, did not reveal any significant long-term change in accident and emergency mental health related attendances. It did, however, report a highly significant but small increase of 0.14% points per week increase in discharge to acute hospital trust wards. This would imply an additional 9.03 acute hospital admissions per annum at a cost of £6700.

Table 22 summarises changes in NHS impacts following the introduction of the PDUs. Overall, at an area level, the ITS analyses, because of the limited number of significant effects observed and the small magnitude of these effects, suggests only very modest impacts on NHS costs at an area level. For the three areas, BSMHFT, LPFT and SWLSTG, where data on both MHT and acute trust activity were available, there would be an annual net reduction in total NHS costs of £3525, £34,666 and £45,957, respectively. For STHFT, only acute hospital trust impacts could be measured, with annual increased costs of just £6278. When looking at the results of meta-analysis alone, there are even fewer significant effects, with a net overall increase in costs of just £2935.

Synthetic control analysis

The synthetic control study (see Chapter 5) matched and compared mental health admissions and lengths of stay, as well as ED attendances, for the mental health and acute trusts in our study areas with trusts with similar characteristics in areas of England where PDUs have not been implemented. The analysis did not however include data from SHSCFT as this site did not contribute any data to HES during the study period.

Impacts on mental health trusts: synthetic control analysis

For the remaining three MHTs, the synthetic control study did not identify any significant impact of the opening of the PDU on admissions to inpatient wards at any of the MHTs. However, a highly significant lower average LOS of 1.5 days for service users admitted to mental health inpatient wards at BSMHFT compared with control trusts was observed. There was no significant difference in LOS in the other two MHTs. Meta-analysis pooling data from all three trusts suggested a significantly lower average LOS of 6.4 days compared with the control areas. These data should, however, be treated cautiously, as the diagnostic codes in HES data for acute mental health admissions may include some specialist admissions or non-hospital acute admissions, such as diversions to home treatment teams.

Despite these uncertainties the synthetic cohort study implies potential economic benefit for all three MHTs following implementation of PDUs relative to control trusts. If average lengths of stay were lower by 15.5 days per 10,000 trust catchment population per month in BSMHDT compared with control MHTs, this could imply economic costs avoided in the area of £1.27 million per annum. This does not necessarily mean that costs have reduced since the PDU was introduced, it could also mean that there has been a lower rate of increase in LOS in comparison with control areas.

Similarly, if lengths of stay across all three MHTs were on average lower than the control areas by 6.4 days per 10,000 trust catchment population per month, as shown in the meta-analysis, the total in costs averted for all three MHTs compared with control trusts could be up to £1.16 million.

Impacts on acute trusts: synthetic control analysis

The synthetic control study revealed that there were no impacts of PDU opening on acute hospital trusts in BSMHDT. There were no significant differences in ED attendances between acute trusts and their comparators, with one exception. For STHFT, the study reported a significant reduction of 1.5 mental health-related ED attendances per 10,000 trust population per month compared with the control areas. This would equate to 87 ED visits averted per month, with total costs averted of £196,620 per annum. There were no significant differences in ambulance and policy conveyance to hospital between acute trusts and their comparators, with one exception: ULHFT reported a significant 9.1% reduction per 10,000 trust catchment population compared with control areas. This would mean 23 fewer ambulance/police conveyances each month compared with the counterfactual areas, potentially avoiding costs of £80,617 per annum. No significant differences in acute inpatient admissions to the same provider were reported for any of the acute trusts. There were also no significant differences in any category of resource use when data were pooled across all acute trusts in meta-analysis.

Return on investment

An objective of this chapter was to draw on findings of earlier chapters to identify potential impacts on the local health economy in each study area following the introduction of PDUs. One element of this is a RoI, from an NHS perspective, in each of the different study site areas. Essentially this compares the costs associated with PDU implementation with observed changes in costs of health services.

Table 23 summarises the RoI for each of the four sites. Here we have assumed that all the impacts on costs from the ITS and synthetic cohort studies are independent and cumulative. Even with this assumption, the limited impacts of PDU implementation on resource use reported by the analyses in Chapters 4 and 5, means that the level of direct annual RoI for most areas is low, being just 2.41% of PDU annual costs in SWLSTG, 8.89% in LPFT and 9.05% of PDU annual costs in STHFT. The exception is BSMHFT, which because of the highly significant reduction in average length of inpatient stay of 15.5 days, has an annual return of 1868% compared with the costs of PDU investment. Our survey in Chapter 3 indicates that the four PDU services were operating between 63% and 83% of full capacity; if services were operating at full capacity and assuming a proportionate change in cost impacts would have a negligible impact on the RoI in three areas but would increase to a maximum of 2990% in BSMHFT (see Appendix 8, Table 56 for full results).

Our original intention was to use the results of the RoI analysis to look at the potential costs and benefits of scaling up access to services across England, but this is not appropriate given the uncertainty around the levels of return, which are low with the clear exception of BSMHFT. While the synthetic control method indicates that the difference in mental health inpatient LOS following the introduction of the PDU in BSMHFTis highly significant, this method does not allow us to isolate the effects of the PDU from other factors that could contribute to difference in LOS. The introduction of the PDU would not be expected to have such a major effort on LOS, given its focus on reducing the number of short (< 5 days) rather than long hospital stays; the relative contribution to average LOS of other innovative services introduced in the same time frame in BSMHFT, including the more intensive use of liaison psychiatry, using the rapid assessment, interface and discharge service needs to be explored. The limited change on resource use seen in both the ITS and synthetic cohort studies in the other areas should also not be interpreted as indicating that there is no economic case for investing in PDUs, but it does indicate that they may be unlikely to offset a substantial proportion of their cost to the NHS.

That said, there are additional resources saved on which we have not placed an economic value; these include any reductions in waiting time in ED. The synthetic cohort study in Chapter 5 pointed to significantly lower waiting times in the ED at KHFT in comparison with control sites. While we have valued contacts with EDs using tariffs that do not vary by time spent waiting, reducing waiting times potentially frees up time of ED staff to see and treat other patients. Furthermore, evidence from previous analysis suggests that individuals detained under section 136 of the Mental Health Act may wait in EDs on average up to 12 hours compared with a general population average of 2.5 hours, owing to the complexity of their needs. This, it has been suggested, is the equivalent of being able to treat 10 other patients in the same time period.100 Our study also does not look at whether PDUs services have displaced existing local mental health services (e.g. have replaced triage wards or other community-based assessment services) or are instead an addition to existing services. There may also be impacts on the use of specialist CMHTs that we do not know about. The costs of providing PDUs will be lower and the RoI greater if they have substituted for some existing services.

It should also be stressed that the time period for this analysis is still relatively short; if individuals benefit from contact with PDUs then their long-term mental health service use costs may be considerably lower, but our analysis at most only covers a 2-year post-PDU implementation period. Untreated poor mental health can have consequences that last for many years or even decades, so a longer time frame may also be needed to fully capture these consequences. Our return on analysis has been conducted from an NHS perspective, but there may be other impacts on other agencies, such as the police and local authorities arising from any improvement in mental health outcomes for service users.

Critically, the most fundamental impacts of all are the short-, mid- and long-term outcomes on PDU service users. Longitudinal analyses are still needed at an individual service user level that compare the long-term costs and outcomes of PDU service users directly with those of individuals that have not used these services. If PDUs help improve levels of recovery from poor mental health as well as social inclusion, this will also have economic benefits for these individuals such as economic independence and wider society (higher levels of productivity from participation in education, paid employment, family caring and volunteering). These could not be included in our RoI analysis. PDUs will also have impacts on quality of life. Another fundamental question is whether contact with PDUs leads to incremental gains in quality of life compared with care as usual.

The remaining sections of this chapter look at some of these missing potential individual-level impacts of PDUs, by drawing on analysis of service use and pathways of care observed in cohort study in Chapter 6, as well as the qualitative analysis in Chapter 7. We also examine the quality of life of individuals attending PDUs and therefore there is scope for substantial improvements in quality of life following PDU contact.

Changes in cost of mental health services following psychiatric decision unit assessment

In addition to ED attendance and psychiatric inpatient admissions, contact with PDUs might be expected to have an impact on the use of community mental health services. Longitudinal cohort data collected as part of decision have been used to explore these impacts. Chapter 6 describes how in each of the four study areas data have been collected on patterns of service contact and utilisation in the 9 months prior to and following an initial visit to a PDU. We have then examined differences in resource use and costs to the NHS in the 9 months after a PDU visit for each of the four sites. Unit costs are drawn from NHS national reference costs and the Personal Social Services Research Unit’s Unit Costs of Health and Social Care used in estimating these costs. Specific sources are reported in Appendix 8, Table 55.

Table 24 shows summary results of pairwise comparisons on per-PDU and post-PDU costs, using 2000 bootstraps, bias corrected and accelerated to account for the non-normal distribution of costs (full results for each site are given in Appendix 8, Tables 57–60). Overall increased contacts with services and inpatient hospitalisations meant that post-PDU costs are significantly greater than pre-PDU costs in all four areas (p = 0.000). Mean total costs post PDU were £6945 greater in BSMHFT £8446 in LPFT, £5803 in SWLSTG and £7908 in SHSCFT. Similarly, the mean costs of inpatient stays were significantly greater in all four sites: £4696 greater in BSMHFT £5059 in LPFT, £3671 in SWLSTG and £5179 in SHSCFT. The cost of liaison psychiatry costs was significantly lower in the post-PDU in LPFT (p = 0.001) and SWLSTG (p = 0.000) although the absolute level of costs was low. Use of both CMHTs, as well as crisis/home treatment teams, were also significantly greater in all four sites. Potentially PDU visits may have led to different patterns of onward referral and signposting to community-based services.

Regression analyses

Multiple statistical analyses were performed to explore the extent to which individual characteristics and pre-PDU costs were associated with costs of service use for the 9-month period following initial visit to the PDU at each of the four PDU study areas. Having determined that there was no statistical difference in costs between the pre COVID-19 cohorts and the primary cohorts where service access may have been affected by COVID-19, we used data from both time periods in three cohorts (BSMHFT, LPFT and SWLSTG) in regression analysis, to increase available cases. There were no pre-pandemic cohort data available for SHSCFT.

In these analyses, the dependent variables were total costs after visiting PDU over the 9-month period in each of the four sites. GLM was used for these analyses to allow flexibility to deal with skewness in the dependent variables and we tested for the best-fitting distributional forms in the regression equations. All pre-PDU baseline variables for the individual characteristics were tested for statistical significance using a significance level of 0.05 and were explored for potential inclusion in the cost models. They were added one at a time to the cost model, given statistical significance and theoretical importance. Variables considered for potential inclusion were: age, sex, ethnicity (White British vs. all other), marital status, employment status (working vs. all other), accommodation (mainstream housing vs. insecure housing, supported housing, health facility, unknown), and primary diagnosis using ICD-10 subcategories (psychoactive substance use, schizophrenia, mood disorders, neurotic disorders, eating disorders, personality disorders, disorders of psych development, behavioural and emotional disorders, other). Stata 15 and SPSS version 24 were used for these analyses.

Tables 25–28 provide summary results for all four areas. For all models a gamma log function model was best fitting. In summary, female sex is an explanatory factor for higher costs in three of the four sites and increased age is associated with higher costs in all four sites. Not being White British was associated with higher post-PDU costs in one area but did not feature in best-fitting models for the other three areas. Insecure or unknown housing status relative to being in mainstream housing was associated with lower costs in both BSMHFT and SHSCFT. Associations between different primary mental disorder diagnoses and post-PDU costs varied across sites.

In the case of BSMHFT, total costs for the 9 months prior to PDU visit were highly significantly associated with post-PDU costs, although their impact on costs was minimal, with a one-unit increase associated with a 0.004% increase in post-PDU costs (p = 0.000). Sex and ethnicity were also significant and had substantial associations with post-PDU costs. Being a woman was associated with 43.15% higher total post-PDU costs than being a man (p = 0.033), while not being White British was associated with 55.28 % higher total post-PDU costs (p = 0.012). Age was also a potential explanatory factor for higher costs; for an increase of age by 1 year, there was a significant increase (2.07%) in the post-PDU total costs (p = 0.004). Relative to a primary diagnosis of a mood disorder (ICD-10 codes F30–F39), diagnosis was not associated with post-PDU costs; this lack of association may be due to the high level of missing data for diagnosis (82%).

In the case of LPFT, total costs for the 9 months prior to PDU visit were highly significantly associated with post PDU costs, although again their impact on costs was minimal, with a one-unit increase associated with a 0.003% increase in post PDU costs (p = 0.004). Sex and ethnicity were also significant and had substantial associations with post-PDU costs. Being a woman was associated with 37.15% higher total post-PDU costs than being a man (p = 0.027). Age was also a potential explanatory factor for higher costs; for an increase of age by 1 year, there was a significant increase (1.55%) in the post-PDU total costs (p = 0.003). Living in insecure housing was associated with 48.8% lower post-PDU costs than those in mainstream housing (p = 0.010); this might be because they are more difficult to reach that people with a fixed address; similarly having an unknown housing status was associated with 56.6% lower post-PDU costs than those in mainstream housing (p = 0.010).

Relative to individuals with a primary diagnosis of mood disorder, having a psychoactive substance use (F10–F19) was associated with 148% higher costs (p = 0.013). Similarly, a diagnosis of schizophrenia, schizotypal, delusional, and other non-mood psychotic disorders (F20–F29) was associated with 406% higher post-PDU costs than for those with mood disorders (p = 0.000). Anxiety, dissociative, stress-related, somatoform and other non-psychotic mental disorders (F40–F48) were associated with 209% higher post-PDU costs (p = 0.000) and personality disorders (F60–F69) with 504% higher post-PDU costs (p = 0.000).

For SWLSTG, being a woman was associated with 45.1% higher total post-PDU costs than being a man (p = 0.013). Age was also a potential explanatory factor for higher costs; for an increase in age by 1 year, there was a significant increase (1.4%) in the post-PDU total costs (p = 0.012). Relative to individuals with a primary diagnosis of mood disorder, many other diagnostic conditions were associated with lower post-PDU costs in contrast to the other study areas. Having a psychoactive substance use (F10–F19) was associated with 79.1% lower post-PDU costs (p = 0.000). Anxiety, dissociative, stress-related, somatoform and other non-psychotic mental disorders (F40–F48) were also associated with 70.4% lower post-PDU costs (p = 0.000). Although primary diagnoses of pervasive and specific development disorders (F80–F89) and behavioural and emotional disorders with onset usually occurring in childhood and adolescence (F90–F98) were also significantly associated with lower PDU costs, only 5 of 494 individuals in the SWL data set had these diagnoses. Ethnicity was not selected for the best-fitting model.

For SHSCFT, which only had 204 individuals in the primary cohort, few significant associations were found between individual characteristics and post-PDU costs. Total costs for the 9 months prior to PDU visit were highly significantly associated with post-PDU costs, although their impact on costs was minimal, with a one-unit increase associated with a 0.004% increase in post-PDU costs (p = 0.001). Age was also a potential explanatory factor for higher costs; for an increase of age by 1 year, there was a significant increase (1.6%) in the post-PDU total costs (p = 0.03). Different diagnostic categories were not associated with post-PDU costs, with the exception of behavioural and emotional disorders with onset usually occurring in childhood and adolescence (F90–F98), but this was based on the costs solely of one individual. Ethnicity was not selected for the best-fitting model.

Insights into the quality of life of psychiatric decision unit service users

The primary function of mental health services is to improve health outcomes and not just impact on resource use and costs. Although the study did not intend to directly look at changes in clinical outcomes, such as levels of mental distress, it has been possible in the study to collect some self-report data on quality-of-life scores, using the EuroQOL EQ-5D-5L quality-of-life instrument.101 The instrument has five domains: mobility, self-care, ability to perform usual activities, pain and depression/anxiety levels, with a score of 1 indicating perfect quality life and a score of 0 being equivalent to death. The five response levels for each domain are: no problems, slight problems, moderate problems, severe problems and extreme problems. The instrument is widely used to generate estimates of quality adjusted life-years used to help assess the cost-effectiveness of health system interventions.

In our analysis quality-of-life data can inform scenario modelling to help indicate the potential plausibility for improving quality of life of PDU service users. This can be useful given that quality of life is the primary outcome measure used in economic analysis to inform decision-making in England, where judgements are typically concerned with assessing whether the additional costs incurred by a service are justified by the quality-of-life gains made.

PDU service users in the cohort study in two of the study areas, LPFT and STHFT were invited to complete the EQ-5D-5L at baseline and at 9-month follow-up. Individuals were able to post their completed instruments back to the PDU and these questionnaires could then be linked to individual service user identifiers. Complete responses were received from 148 participants, 107 from LPFT and 41 from STHFT.

Currently there is no accepted value set in England for the five-level version of EQ-5D, so in line with current recommendations from the National Institute for Health and Care Excellence,102 we have used the crosswalk mapping function developed to map five-level values to the existing English three-level value set103 to generate utility scores. Figure 15 shows the distribution of EQ-5D-5L scores at baseline. The average utility score of just 0.14 is extremely low, being for example, much lower than average values reported for young people in England who have self-harmed of 0.68.104 41% of these scores were less than zero, indicating that these individuals had utility scores that could be viewed as worse than death; these very low values in a high proportion of individuals might be expected, given that many individuals attending PDUs may have been experiencing suicidal thoughts.

FIGURE 15 Distribution of EQ-5D-5L scores in reporting participants at baseline in LPFT and SHSCFT. (Figure not reproduced here; see the original article.)

While this sample is small, the data suggest that there is potential for substantial changes in utility values with appropriate treatment; if PDUs do lead to such improvements then the economic case for investment will be strengthened. Although only 28 follow-up responses at 9 months on utility scores were returned to the study team, average utility scores in these 28 individuals were much greater at 0.41, compared with 0.10 for the 28 at baseline. While too small to have clinical meaning, nonetheless pairwise analysis indicated that this 0.30 improvement in utility scores was significant (p = 0.002).

Regression analysis

The association of quality-of-life scores pre-PDU visit, individual characteristics and total costs were also explored by using multiple regression analyses. GLM was used, testing for the distributional forms for the best-fitting models. All individual characteristics, pre-PDU quality-of-life scores and pre-PDU total costs were tested for inclusion in the models, using a p-value of 0.05.

A gamma log function model was best fitting. Better quality of life was associated with less use of services. As Table 29 shows the ED-5D-5L utility scores at baseline were significantly associated with total post-PDU costs (p = 0.041). For each one-unit increase in utility scores, there was a 15% decrease in total costs. Being a woman was associated with 13% higher costs than being a man (p = 0.000). Having an unknown housing situation was also associated with 76% lower post-PDU costs than those in mainstream housing (p = 0.000). Living in a health facility at baseline was associated with 71% higher post-PDU costs but this only concerned two individuals. Relative to a primary diagnosis of a mood disorder (ICD-10 codes F30–F39), being diagnosed with psychoactive substance use was associated with 43% higher post-PDU costs (p = 0.027). Similarly, a diagnosis of schizophrenia, schizotypal, delusional, and other non-mood psychotic disorders (F20–F29) was associated with 201% higher costs (p = 0.000), while anxiety, dissociative, stress-related, somatoform and other non-psychotic mental disorders (F40–F48) were associated with 134% higher costs than those with mood disorders (p = 0.007).

Individual service user scenarios

Decision has shown that PDUs support individuals with a wide range of differing personal and social circumstances, as well as differing mental health problems. Our analysis of service use in the 9 months after initial visit to the PDU for the longitudinal cohorts in all four sites indicates this is associated with significantly greater levels of both community and inpatient mental health service utilisation than received in the 9 months pre-PDU visit.

A key question is whether service pathways followed after PDU visits are more appropriate to service user needs and lead to better outcomes than would otherwise be the case. The time that PDU staff have to influence service pathways is very limited, usually no more than 48 hours per stay, at which point a decision must be made as to whether a service user is simply discharged home, referred or signposted to a local crisis team or other community mental health service, or transferred to an acute psychiatric ward.

Ideally, this question is best addressed through longitudinal analyses comparing the experiences of individual service users who come into contact with mental health services as a result of crises and are referred to PDUs, with those service users who do not have this option. While we do not have such data, we can illustratively look at the level of impact that a PDU would potentially have to have to be considered cost-effective. This can be done by creating simple scenarios looking at potential experiences and outcomes for hypothetical individual service users.

We can construct some scenarios drawing on information on the cost of PDU service provision, together with observed patterns of short-term service use drawn from the cohort analysis, as well as quality-of-life utility scores we have collated from 148 PDU service users. In addition, we also refer to qualitative research in Chapter 7 on whether PDUs are perceived as helpful in recovery.

We have illustrated the potential economic case for 16 hypothetical PDU service users, 4 in each of our PDU sites. Sex in the cohort analysis was associated with increased service use costs in the 9 months post PDU use, so our scenarios include two men and two women from each site. Our cohort analyses also indicate that costs are highly skewed, with a small number of individuals having substantial inpatient mental health stays in each area, while the majority of individuals only use community services post PDU visit. To reflect this cost variation simply, four of the scenarios look at median incremental costs post PDU visit compared with pre-PDU service use, as there was no use of inpatient services (‘low cost’), while the other four scenarios apply mean costs identified in each of the cohort sites which do account for inpatient costs (‘high cost’).

We do not know what impacts PDUs have on quality of life, but we can estimate the minimum level of quality adjusted life-year (QALY) gain that would be needed for the PDU intervention to have a cost per QALY below the accepted cost-effectiveness threshold of £30,000 for each of the 16 scenarios. Some service users in qualitative interviews also pointed to the role that PDUs can play in signposting towards support to help stay or return to employment, for instance signposting to recovery colleges and employment service-related information. One individual also expressed an interest in getting a job in future in a PDU. Equally, some service users did not have good experiences in post-PDU support to obtain employment and also mentioned the practical difficulties sometimes faced when wanting to return to an existing job, particularly after a long period of absences. The desire to maintain or return to employment was an issue raised in interviews so we also considered how varying the proportion of individuals who are in employment at 9 months post PDU might also impact on cost per QALY gained from a societal perspective.

As Table 30 indicates for the low-cost scenarios where service users do not spend time in inpatient care, then the QALY gains over 9 months required to produce a cost per QALY below £30,000, range from 0.06 for a woman or man in Birmingham to 0.19 for a woman in Sheffield. If on average over 9 months an additional 30 days were spent in work, valued conservatively at the minimum wage rate in 2020 for the over 25s, then the cost per QALY would fall further, in the case of Birmingham almost being cost neutral.

For the high-cost scenarios in Table 31 when service users do spend time in inpatient care, then the QALY gains required to produce a cost per QALY below £30,000 range from 0.25 for a man in SWL to 0.42 for a man in Sheffield. If on average individuals spend an extra 15 days in work over 9 months then the cost per QALY gained could range between £25,000 and £27,000.

These estimates are purely illustrative, but they do indicate, particularly for service users who do not need to make use of inpatient care, that only very modest improvement in quality of life might be required for the PDU intervention to be considered cost-effective. For high service uses the picture is more complex, with a much higher level of improvement in quality of life needed for the PDU to be considered cost-effective. Our earlier collection of data on quality of life, did however indicate that mean changes in quality-of-life utility values of 0.3 were seen in the very small number of individuals who provided these utility values at 9 months post PDU. These scenarios are also potentially conservative as they assume that there would be no increase in use of services for individuals who do not make use of PDUs; there will also be impacts on the use of acute hospital services that were not included in service use data in the cohort analyses. The scenarios only cover a 9-month period, if any quality of life and productivity benefits are sustained beyond this time period then the case for investment would be strengthened.