Population characteristics have not been adequately adjusted for in phase 1

Interviewees identified a range of population characteristics as important factors driving avoidable admissions. They mentioned very elderly people, deprivation, comorbidities, recent immigrants, high expectations of quick access to care and lack of awareness of how to use emergency and urgent care services. To a large extent these were included in the phase 1 regression. Age was adjusted for in the SAAR, including > 85 years old, and deprivation was the most explanatory factor in the phase 1 regression. The interviews helped to explain how deprivation may also be measuring other factors such as morbidity and high expectations of fast access. There seems little utility in pursuing further measurement of any population characteristics because these were measured in phase 1.

Acute trusts have different coding practices for admissions which affect variation in the standardised avoidable admissions rate

Some acute trusts have assessment units or ‘holding areas’ where patients can wait for tests or be prepared for discharge without staying in the ED and breaching the 4-hour wait. There was evidence in our six case studies that different coding practices were applied to these patients. This was explicit and extreme in two cases. In HU1, with a very high avoidable admission rate which we underpredicted, interviewees from one of the acute trusts were explicit that any attendances to their assessment units were classed as admissions, even though some junior staff did not realise this. In contrast, in MO6, with a medium admission rate which we overpredicted, interviewees in the acute trust were explicit that attendances to their assessment units were not classed as admissions, even if they stayed overnight. This distinction was less clear for the other four case studies. Even so, it appears that coding practices could explain further variation in the SAAR.

In the phase 1 regression two factors were included in the multiple regression which might be related to coding practices: conversion rate and percentage of short-stay patients. Looking at factors in the phase 1 regression for the acute trusts in HU1 and MO6, we can see little difference in poverty and the population use of their EDs (Table 31). However, there is a stark difference in the conversion rates from attendance to admissions, suggesting that this variable may be capturing coding practices as well as issues such as differences in presence of senior review at an acute trust. Surprisingly, there is not much difference in the proportion of emergency admissions staying for less than a day between these two acute trusts; both trusts have high rates of short-stay patients. Short stays can be influenced by approaches to early discharge as well as coding practices, but nevertheless a larger difference between the two trusts would be expected.

TABLE 31

Comparison of factors in phase 1 regression for HU1 and MO6 (acute trust-based systems)

Integrated systems have lower standardised avoidable admissions rates

Integration of different services was identified by interviewees as important for avoiding admissions. Interviewees in each of the six cases identified some integration within their system and some problems they perceived there to be with integration. We did not measure integration in the phase 1 regression. In our case studies there was evidence of considerable variation in integration, based on interviewees’ perceptions of their systems. We ordered the six systems by level of integration from best to poorest, a subjective assessment based on the interviews (Table 32). If integration is likely to explain further variation in the SAAR, the pattern we would expect to see is the two systems with underpredicted SAARs at the poor end of the spectrum. This pattern was not consistent in that MO6 is at the poor end of integration. However, the pattern is strong enough to support the hypothesis and we will test this in the phase 3 regression.

TABLE 32

Levels of integration in the six case studies, ordered from best to poorest

Proactive approach to admission avoidance gives low standardised avoidable admissions rate

Some systems had initiatives with a remit of admission avoidance which were described by interviewees as proactive and in operation during the time of the SAAR. These involved proactive senior review by medical staff, multidisciplinary teams actively seeking out hospital attendances who could be safely sent home with support, or teams actively checking on vulnerable people at home. We ordered the cases by the level of proactiveness around admission avoidance (Table 33). Again, this was a subjective assessment. There was a pattern of systems with underpredicted SAARs lacking proactive admission avoidance initiatives.

TABLE 33

Cases ordered by level of proactiveness of their admission avoidance initiatives

Hospital-centric systems have higher standardised avoidable admissions rates

One system was described as hospital-centric by interviewees (HU1), with little investment in community services. Interviewees did not describe others systems in this way. One way of avoiding admissions is to offer community support in community beds or patients’ own homes. There seemed to be variation in the amount of community services available in different systems in terms of community beds and community matrons, based on interviewees’ views. We had partial information about numbers of community services so we recontacted interviewees and searched websites to obtain these numbers (Table 34). In the phase 1 regression the number of acute beds per 1000 population was included in the acute trust-based analysis but not the geographically based analysis. This factor explained variation in the final multiple regression. In the geographically based system case studies, systems could have two acute trusts with very different numbers of acute beds. We took the average of these and found that our cases did not differ very much by this factor. When we compared community bed and matron numbers in each system, they were low in HU1 compared with the other five systems, supporting the interviewees’ perceptions in HU1. However, if this were another explanatory factor of the SAAR then we would expect both HU1 and MU4 to have low numbers, and this was not the case (Table 34).

TABLE 34

Hospital to community beds ratio in the six case studies

Systems with support services out of hours have lower standardised avoidable admissions rates

Interviewees described how having community nursing, mental health and social services available OOH (in evenings, at night and at weekends) could help to avoid admissions. Interviewees in every system wanted more services available for longer regardless of the amount of OOH availability they had. Nonetheless, some systems appeared to have more services available OOH than others, based on the interviews. Perceptions of OOH services differed by system, with a pattern that systems with underpredicted SAARs were perceived to have poorer access (Table 35).

TABLE 35

Perceptions of availability of OOH provision (based on interviews)

The GP OOH service is a significant service in the system and one on which we had few data in the phase 1 regression. We were able to test the population’s knowledge of how to access this service and their perceptions of how easy it was to access. However, we had no data on availability or quality. In the case studies, the views about this service differed by system. Interviewees in both rural systems felt that covering a large geographical area exacerbated problems with filling shifts and providing a good-quality service (LO2, MU4). In one system a much larger number of interviewees expressed negative views, particularly expressing concerns about the extent to which the service failed to cover shifts (MU4).

We searched websites and documents for information on the opening times of different services relevant to avoiding admissions. We included availability of radiography because access to diagnostics OOH had been raised by a number of interviewees. We mapped these onto the system configuration diagrams, shading services darkest where they were open longer. It was not possible to see a pattern between different systems that supported the perceptions in Table 35 (see Figures 12–17).

FIGURE 12

HU1 OOH service opening hours and radiography availability. MIU, minor injuries unit; WIC, walk-in centre.

FIGURE 17

MO6 OOH service opening hours and radiography availability.

FIGURE 13

LO2 OOH service opening hours and radiography availability. MIU, minor injuries unit; WIC, walk-in centre.

FIGURE 14

MO3 OOH service opening hours and radiography availability. MIU, minor injuries unit.

FIGURE 15

MU4 OOH service opening hours and radiography availability. MIU, minor injuries unit; WIC, walk-in centre.

FIGURE 16

LO5 OOH service opening hours and radiography availability. WIC, walk-in centre.

Well-functioning emergency departments can help to avoid admissions

In the interviews there was some evidence that EDs with relationships with services within their hospital and in the community could avoid admissions, as could those which were not suffering from low ED consultant presence. No ED interviewees said that they were fully staffed and had the full complement of support services they needed. However, in one ED in each of our underpredicted systems (HU1 and MU4) there was discussion about significant difficulties recruiting consultants. We did have some variables in the phase 1 regression that are likely to capture what happens in the ED, particularly the conversion rate from attendance to admission and the length of stay < 1 day.

The conversion variable in the geographically based system phase 1 regression may be incorrect

As we developed the case personalities we started to doubt the data used about EDs for the geographically based systems. The data for geographically based systems sometimes looked very different from those for acute trust-based systems and did not tally with interviewees’ comments. We specifically wondered if systems with large numbers of UCCs, walk-in centres and minor injury units (known as type 3 and 4 EDs in national data sets) had sometimes been incorporated with EDs led by consultants (type 1 and type 2). The data for the two variables about ED use and conversion rates had come from the Atlas of Variation,³⁵ which had issued a caution about the quality of its variables. Therefore we used HES A&E 2009/10 data to calculate the demand for EDs (types 1 and 2 only) and the conversion rate for our six geographically based systems. The data from the two sources were different for some of our cases (emboldened), but especially for HU1, where only 47% of the attendances in HES A&E were for type 1 and 2 EDs compared with over 90% for the other five cases (Table 36). The new data were more in line with interviewees’ comments and the acute trust data. The data were also different for MO6 ED attendance rates.

TABLE 36

Comparison of old and new ED variables for six cases

Systems which are not the primary population for an acute trust struggle with integration and therefore have high standardised avoidable admissions rates

HU1 had a population which was not the primary population of either of the two acute trusts it used. This appeared to mean that integration between the acute trusts and community services was difficult. This may have been the case for one acute trust in LO5.

Systems with low resources cannot fund services to avoid admissions

The case studies highlighted that our phase 1 regression did not take into account the resources available within systems. There was some evidence that systems facing financial difficulties reduced their admission avoidance schemes. We did not include a resource variable in the regression because the one on which we found data – resource allocation – was based on emergency admission rates.

Conclusions

The 14 conditions on which the SAAR was based had credibility among our interviewees. Interviewees offered views which supported and explained some of the findings in the phase 1 regression. For example, social deprivation was identified as driving avoidable admissions and the reasons for this were high levels of morbidity but also how people from deprived communities used services. There was also some indication of why rural areas might have lower SAARs: that the difficulty in offering services over a large geographical was an incentive to keep people out of hospital. The conversion rate from ED attendance to admission could indicate the use of proactive initiatives such as senior review or rapid assessment teams, or the way in which admissions were coded. The case studies also identified further factors that might explain variation in the SAAR, for example the level of integration within a system or the availability of support services OOH.

Factors to test in phase 3

• Level of integration between services in the system.
• Amount of proactive admission avoidance initiatives.
• New data on attendance and conversion rates at EDs.
• Availability of services OOH.
• Amount of resource within systems.