Delivery, dose, outcomes and resource use of stroke therapy: the SSNAPIEST observational study

Article history

The research reported in this issue of the journal was funded by the HS&DR programme or one of its preceding programmes as project number 14/198/09. The contractual start date was in April 2016. The final report began editorial review in February 2019 and was accepted for publication in November 2019. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The HS&DR editors and production house have tried to ensure the accuracy of the authors’ report and would like to thank the reviewers for their constructive comments on the final report document. However, they do not accept liability for damages or losses arising from material published in this report.

Declared competing interests of authors

Andy Vail, Audrey Bowen, Benjamin Bray and Sarah Tyson declare research grant funding from the National Institute for Health Research. Audrey Bowen was a member of the Intercollegiate Stroke Working Party (2002–16). In addition, Audrey Bowen’s university salary is part funded by a personal award from the Stroke Association and stroke research grants from the National Institute for Health Research. Sarah Tyson is currently a member of the Intercollegiate Stroke Working Party (2019–present).

Copyright statement

© Queen’s Printer and Controller of HMSO 2020. This work was produced by Gittins et al. under the terms of a commissioning contract issued by the Secretary of State for Health and Social Care. This issue may be freely reproduced for the purposes of private research and study and extracts (or indeed, the full report) may be included in professional journals provided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising. Applications for commercial reproduction should be addressed to: NIHR Journals Library, National Institute for Health Research, Evaluation, Trials and Studies Coordinating Centre, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.

2020 Queen’s Printer and Controller of HMSO

Report structure and analytical strategy

The report is split into nine chapters; the first provides brief background information on the report, the stroke field and the study proposed, including its objectives. Chapter 2 describes the data set, along with any key methodological issues encountered and some brief summary statistics of the data. Chapters 3–8 describe the analysis performed to answer the objectives of the study (see Objectives), before describing, interpreting and discussing the associated results and limitations, and, when appropriate, make recommendations. Chapter 9 provides a brief summary and discussion of the results obtained.

Owing to the complex nature of the data set and analysis employed, it is useful to briefly explain the analytical strategy employed throughout this project. The study analysed an observational records-based data set with a number of characteristics, which meant that we needed to be careful with respect to what analysis was performed, but also when an analysis was performed. The primary pre-planned analysis is reported in Chapters 6 and 7, where we have used complex modelling techniques to account for as much measured and unmeasured confounding as possible, while also investigating the relationship between factors on a hypothesised likely causal time frame (e.g. therapy during inpatient care results in altered health status at discharge). Earlier in the document (see Chapters 4 and 5), common groups of patients were identified based on stroke or care characteristics. We understand that it may be desirable to compare patient or stroke characteristics across these groups using simple statistical tests. We have refrained from doing so and simply reported descriptive statistics. This is due to the complex nature of confounding within the data, which meant that any hypothesis test here would be difficult to justify or interpret, and would potentially result in spurious conclusions. Similarly, the analytical results reported in Chapters 6 and 7 are reported without p-values, as is common practice in most epidemiology journals.

Background introduction

In the UK, > 100,000 people have a stroke per year1 and two-thirds of those who survive are left with a long-term disability, such that there are > 1.2 million stroke survivors in England alone. 1 Consequently, stroke is the most common cause of severe adult disability in the UK. 2

Treatment in a specialist stroke unit is the cornerstone of stroke care, with meta-analyses showing that it reduces death and disability. 3 A fundamental element of this care is assessment and treatment by specialist stroke therapists working within a multidisciplinary team (MDT). 3 There is substantial evidence that stroke therapy is effective, but needs to be provided intensively. 4–7 However, most stroke patients receive little therapy and most spend most of their time inactive and alone,8–12 particularly in the UK. 13,14 Research comparing stroke rehabilitation outcomes in four European countries showed that UK stroke patients received less therapy and had poorer outcomes than those in Germany and Switzerland, even when confounding variables (such as stroke severity) were controlled. 15 The differences in outcome were attributed to the amount, rather than the type, of therapy and the UK’s low dose of therapy was due to poor organisation, rather than lower staffing levels. 16–19

For many years, stroke has been recognised as an NHS priority area, which led to a National Stroke Improvement Programme, including instigation of a national stroke register and audit programme called the Sentinel Stroke National Audit Programme (SSNAP). 13,20 Full details on SSNAP can be found elsewhere21 and are summarised in Chapter 2. One of the standards measured in SSNAP concerns the amount of therapy that patients receive, based on the national clinical guideline for stroke’s recommendation,18 that ‘People with stroke should accumulate at least 45 minutes of each appropriate therapy every day, at a frequency that enables them to meet their rehabilitation goals, and for as long as they are willing and capable of participating and showing measurable benefit from treatment’. However, this amount of therapy is rarely achieved. 13 It is acknowledged that therapy levels are inadequate for most stroke survivors,8–12 but little is known about the factors influencing the provision and effectiveness of stroke therapies or different therapy pathways, as most research focuses on individual interventions rather than service organisation. These are important issues, as lack of therapy and inequity of access are major causes of service users’ dissatisfaction with stroke services. 22,23 The importance of this issue has been recognised by the national clinical guideline for stroke, which highlighted that ‘research into the intensity of therapy and how to deliver it should be a high priority’. 16

Our aim was therefore to investigate why insufficient therapy is provided, by interrogating national audit data from SSNAP for insights into how inpatient and community-based stroke therapy is organised and delivered in the UK, and the association that this may have with patient- and service-related outcome, quality of care and cost. We wanted to find out not only who receives stroke therapy, but also what limits the amount that they receive. This information will provide rigorous, relevant evidence on factors influencing the quality, equity and organisation of stroke therapies, costs and outcomes, which is needed to enable evidence-based service developments to improve outcomes and equity of access, and may, ultimately, reduce therapy costs.

As the impairments caused by an individual’s stroke are a key factor influencing the therapy a patient requires, we also aimed to explore the frequency and severity of stroke-related impairments and how they cluster together to form stroke subgroups. The associations between subgroups and patient demographics, other stroke-related characteristics, the therapy received and outcomes, were investigated. This information will equip us with realistic NHS data with which to design future research trials of stratified stroke rehabilitation pathways and/or service redevelopment.

We investigated four therapies: physiotherapy (PT), occupational therapy (OT), speech and language therapy (SLT) and psychology (Psych).

Objectives

The overall objective of this project was to investigate how inpatient and community-based stroke therapy is organised and delivered in the England, Wales and Northern Ireland, and the associations this may have with patient and organisational factors, outcome and cost.

We will:

• Describe the stroke therapy delivered, the quality of processes of care and the stroke therapy workforce, and quantify variation in therapy provision (see Chapter 3).

  • Research questions: how much stroke therapy is provided? How many stroke services include community-based stroke therapy? How many services have access to the wider MDT? What are stroke therapy staffing levels and working hours? What is the quality of therapy-related processes of care? How much variation exists in the amount of each therapy received?

• Identify the different therapy pathways, characterise them and the patients who follow them, and calculate their costs (see Chapter 4).

  • Research questions: which stroke therapy pathways are used? What are the characteristics of patients who follow them? What therapies do they receive in each pathway? What does each pathway cost?

• Explore patient- and stroke-related characteristics and their association with recovery, to identify important subgroups of stroke survivors (see Chapter 5).

  • Research questions: what is the frequency of stroke-related impairments? Which impairments are commonly comorbid and to what extent? Do patients with common comorbidities receive different amounts of therapy or achieve different clinical outcomes?

• Identify the factors associated with therapy provision (see Chapter 6).

  • Research questions: which organisational- and patient-related factors are associated with the amount of therapy provided?

• Explore therapy and outcomes (see Chapter 7).

  • Research questions: how is therapy provision associated with patient- and service-related outcomes?

• Explore costs of stroke therapy (see Chapter 8).

  • Research questions: how is the amount of stroke therapy associated with resource use during inpatient stroke care?

Patient and public involvement

Members of the patient and public involvement panel of the University of Manchester’s Stroke Research Centre have contributed throughout the project. The panel consists of stroke survivors and their families and carers who provide a patient and public involvement perspective for stroke research in Manchester. It was founded by the North West Stroke Research Network and continued by the University of Manchester Stroke Research Group after the stroke research network’s demise. It is led by a stroke survivor. The panel has > 30 members of all ages, types and severity. The panel supported the project, highlighting that difficulty accessing appropriate therapy to meet their needs was a cause of great concern for many stroke survivors and a major cause of dissatisfaction with stroke services. They have contributed to the interpretation of the results and layperson’s summaries for stroke survivors and lay audiences.

The data extracted

Data for adults (aged ≥ 18 years) admitted to hospital with a stroke in England, Wales and Northern Ireland between July 2013 and July 2015 who were recorded in the SSNAP were extracted. To focus on the patients who were most likely candidates for stroke therapy, further inclusion criteria were to have survived and still be an inpatient after 72 hours but not be receiving end-of-life care during this period. Patients with intracerebral haemorrhage were included, but those with other types of haemorrhage (subarachnoid, subdural or extradural) and also patients who had their stroke more than 28 days before admission to hospital are not recorded in SSNAP and thus excluded from this project.

Although the clinical database collects information for each stroke team at every stage of the stroke care pathway, the organisational audits provide information regarding the care provided in a hospital or trust, which may include several different stroke teams. For example, a hospital or trust may include a team providing hyperacute or acute care, and a separate stroke rehabilitation team. To link the two data sets together, the SSNAP provided a codebook so that stroke teams and hospital trusts could be matched.

Missing data

The SSNAP involves rigorous data quality control so the data received were expected to be clean and complete, but nonetheless data checks were performed for missing data, extreme outliers and other inconsistencies, which were screened and clarified with the SSNAP team before analysis.

Unlike the acute organisational audit, participation in the post-acute organisational audit was voluntary. Thus, the post-acute organisational audit contains a relatively high degree of missing data and the stroke teams who reported to the SSNAP (who could be considered ‘early adopters’) during the data extraction period, may not be representative of all community-based stroke teams. Thus, we treated results regarding the community-based stroke services with caution.

Missing data were present in two key variables: (1) social deprivation and (2) stroke severity on admission. Data regarding social deprivation were based on the patients’ postcode, using data linkage with Index of Multiple Deprivation national statistics for patients living in England. Thus, much of the missing deprivation data were patients living in Wales and Northern Ireland. To maximise a complete cases analysis and adequately adjust for social deprivation as a key confounder, missing values for social deprivation were set as an ‘unknown’ category. Missing social deprivation scores were unlikely to be missing ‘completely at random’. As they may be more likely in lower (or higher) deprivation groups, the interpretation of any social deprivation results was minimised.

Stroke severity was measured on admission to hospital using the National Institutes of Health Stroke Scale (NIHSS). 28,29 This is a quick, simple, psychometrically robust (if relatively crude) measure of overall stroke severity, which assesses the number and severity of stroke-related impairments. It contains 15 items that measure level of consciousness (consciousness, orientation, ability to follow commands); cognition (language and neglect); vision (motor visual-field loss and extraocular movement); motor control (weakness of the limbs, ataxia and dysarthria); and sensory loss. A trained observer rates the patient’s ability to answer questions and perform activities. They score whether or not the impairment is present and its severity on 3-, 4- or 5-point scales. These are summed to give a total score, with a maximum of 42. A score of zero indicates the absence of stroke symptoms (sufficient to limit function) and the higher the score, the more severe the stroke. Only one item of the NIHSS is mandatory in the SSNAP (level of consciousness), so patients may have missing values for the remaining 14 items. If missing data were present, then the summed NIHSS score would be artificially low. For example, if the level of consciousness was recorded as 3 (severe unconsciousness and so a severe stroke) and all others were missing, then the total NIHSS score would also be 3, which indicates a mild stroke. To account for this, rather than excluding all patients without complete NIHSS data (which could cause a selection bias), level of consciousness was used as a proxy measure of stroke severity when other data were missing. The level of consciousness scores (i.e. 0, 1, 2 and 3) were mapped on to stroke severity scores from the NIHSS as follows:

• level of consciousness = 0 = mild stroke (NIHSS score < 5)

• level of consciousness = 1 = moderate stroke (NIHSS score 5–14)

• level of consciousness = 2 = severe stroke (NIHSS score 15–20)

• level of consciousness = 3 = very severe stroke (NIHSS score > 20).

If other items were recorded, they were added to the patient’s score, thus the patient’s corresponding severity category was defined as the total of the patient observed score plus the adjustment value for level of consciousness. Patients were excluded only if level of consciousness was recorded as 0 (i.e. alert), but all other NIHSS assessments were missing, as they were felt to be a special set of cases.

Defining stroke teams

The type of stroke team from whom the patient received care may impact on the care provided and, possibly, the recovery seen. In the 2014 organisational acute care audit,25 the SSNAP classified inpatient stroke teams as:

• routinely admitting (stroke) team (RAT) – stroke teams that regularly and directly admit stroke patients for acute and/or hyperacute stroke care

• non-routinely admitting (stroke) team (NRAT) – teams that do not generally admit stroke patients directly, but provide acute care and/or rehabilitation for patients repatriated from a hyperacute stroke team

• non-admitting inpatient (stroke) team (NAIT) – teams that do not admit stroke patients, but provide inpatient stroke rehabilitation.

These were modified in an attempt to distinguish between RATs providing acute (and sometimes hyperacute) care, and those that combined acute care and rehabilitation. This was performed according to each team’s median length of stay (LOS). The SSNAP defines acute stroke care as lasting up to 7 days,25 so median LOS < 7 days identified RATs that were working as a routinely admitting acute/hyperacute stroke team (RATa), whereas a median LOS of ≥ 7 days defined a RAT team which provided combined (i.e. acute care and rehabilitation) care [routinely admitting combined acute and rehabilitation team (RATc)].

In addition, teams providing community-based stroke care are defined by the SSNAP as:

• early supported discharge (ESD) team – patient was discharged to a MDT that typically co-ordinate early discharge from hospital and provide short-term continued rehabilitation for patients with mild–moderate stroke30

• community rehabilitation team (CRT) – a MDT providing community-based rehabilitation for stroke patients with any level of severity (the time scale over which treatment is provided varies but is generally longer than an ESD team)

• integrated CRT (ESD/CRT) – a team that provided both ESD and community rehabilitation.

Results: the SSNAPIEST cohort

During the data extraction period (July 2013 and June 2015), 149,560 stroke patients were admitted to hospital and entered into the SSNAP clinical audit. A total of 41,706 patients were excluded as they had a LOS < 3 days (due to death or early discharge), or received palliative care. A further 12,949 patients were excluded if level of consciousness on admission was recorded as zero (i.e. ‘alert’), but all other NIHSS items were incomplete, leaving 94,905 patients in the study. This included 314 patients who were recorded as being readmitted to inpatient stoke care after being discharged to a community-based stroke service.

There were slightly more women than men in the cohort (women n = 49,199, 51.8%), mean age was 76 [standard deviation (SD) 13.2, minimum : maximum = 1 : 114] years, 89% of patients were white, social deprivation was evenly distributed and 79% (n = 75,101) were independent before their stroke [pre-morbid modified Rankin Scale (mRS) ≤ 2]. 31,32 Over one-quarter (27.9%) had suffered a previous stroke/transient ischaemic attack (TIA) and two-thirds had one or two stroke-related comorbidities [median 1, interquartile range (IQR) 1–2]. Eleven per cent suffered an intracerebral haemorrhage rather than infarction. The median stroke severity (NIHSS score on admission) was 6 (IQR 3–12) and 40% had a moderately severe stroke (NIHSS on admission 5–14). Eight-one per cent (n = 76,585) were fully alert (NIHSS level of consciousness score = 0 on admission). Further details of the patient characteristics are in Table 1.

Fourteen per cent (n = 13,504) of patients died while under the care of an inpatient or community-based stroke service (having survived the initial 72 hours after admission). However, only 9227 (68.3% of those whom died) were recognised as needing palliative care. Fifty-seven per cent (n = 53,720) of patients were discharged home, 14.2% (n = 13,461) were discharged to a care home and 12% (n = 11,213) were transferred elsewhere (to another clinical service, for example) or were still an inpatient 6 months after their stroke. Median length of inpatient stay was 11 (IQR 6–27, minimum : maximum = 3 : 804) days for the whole cohort and for those who survived it was 16.1 (IQR 9–40) days. Although the median LOS was short, there were individual patients with a much longer LOS: > 9 months for people admitted to a RAT and up to 2 years in total. These patients form an interesting subgroup who are likely to use a high degree of resource. Further work to characterise them, their care and the resources used is needed.

At discharge from hospital, 36.7% (n = 42,255) were independent (mRS ≤ 2). Of the 53,721 patients who were discharged home, 15,233 (28.4%) required assistance with everyday activities and 15,058 (99%) patients received it. For 8108 patients (53% of those who received help), this was provided by formal (paid) carers; 4017 patients (26.7% of those who received help) were assisted by informal (unpaid) carers; and 2843 patients (18.9%) received help from both formal and informal carers. Some patients refused the help that they were considered to need by clinical teams (n = 228) and, for a few, the help needed was not available (n = 37). A total of 15,292 (16.11%) patients were known to live alone on discharge. Of patients living at home, 8582 (16.0%) received support from social services. The median number of visits per week was 14 (IQR 7–28, minimum : maximum = 0 : 56). Of the 13,461 patients who were discharged to a care home, 4078 (46.4%) were a resident in the care home before their stroke. For 14.1% (n = 1905) of patients, their stay in the care home was intended to be temporary.

A total of 15,861 patients (19.4% of those who survived and 29.5% of those who were discharged home) were transferred to community-based therapy on discharge from hospital. Sixty-six per cent (n = 10,577) were transferred to an ESD team: 3662 (23%) patients to a CRT and 1622 (10.2%) patients to an integrated therapy team. The median LOS/duration of treatment with community-based therapy was 41 (IQR 20–65) days: 36 (IQR 17–48) days for an ESD team and 68 (IQR 34–120) days with a CRT.

One hundred and eighty-three hospital trusts responded to the acute organisational audit, representing 197 acute stroke teams (some hospital trusts submitted responses from more than one stroke team). When matched with the clinical audit data, 82 teams were identified as having a hyper/acute stroke team (RATa), 78 had a combined stroke team (RATc) and 37 had a NRAT. The mean number of stroke beds in each service in the acute organisational audit (RATa, RATc and NRATs) was 28.7 (SD 12.9, minimum : maximum = 6 : 76). Thrombolysis was provided in 151 (82.5%) of acute stroke services. Most patients (n = 92,286, 70.4%) had access to an ESD team (whether stroke specific or not) and a CRT (n = 76,067, 58.0%). The median waiting time for initial contact with a community-based team after discharge was 1 (IQR 1–3) day and median time between discharge and the start of treatment was 2 (IQR 1–5) days. Both waiting times were shorter for ESD (both 1 day) than for community rehabilitation (median of 3 and 6 days, respectively) and integrated teams (median of 2 and 5 days, respectively).

Two hundred and one responders to the post-acute organisational audit were matched. Eighty-nine were specialist inpatient rehabilitation units (NAITs) and 201 were community-based teams (101 were ESD teams, 75 were CRTs and 25 integrated rehabilitation teams). Most (62%) of the community-based services that responded to the audit were stroke specific. However, this was skewed by the ESD services, which were almost exclusively for stroke survivors. Of the CRTs, approximately one-third each were stroke specific, combined stroke and neurological rehabilitation or generic. The median number of patients referred in the previous week for community-based teams was 19 (IQR 9–30). This was higher for ESD teams (median 10, IQR 10–28) in the last week than in CRTs (median 21.5, IQR 6–35).

Assessment

Table 2 shows the completeness of assessments. PT and OT assessments were completed for most patients (87.7% and 77.3%, respectively), whereas swallow and communication assessments were completed for < 50% of patients. Less than half of patients had a formal assessment of continence and a management plan in place, whereas approximately three-quarters of patients had a screening assessment of their mood and cognition at some point during their care.

Therapy requirements and provision

Table 3 shows patients’ requirements for each therapy and whether or not they received it during their stay. Nearly all patients were considered to require inpatient PT (92%) and OT (88%), whereas just over half required SLT, but only 5% were considered to require Psych (at any point during their inpatient stay). The proportion of the patients’ inpatient stay in which they received therapy was low, ranging from 40% for PT and 5% for Psych. However, only ≈ 40% of patients required each therapy for their whole admission (39.6% for PT, 39.1% for OT, 40.9% for SLT and 39.8% for Psych). For those who did not require therapy for their whole admission, the median days on which they required therapy were 20 (IQR 41–72) days for PT; 20 (IQR 41–70) days for OT; 18 (IQR 38–68) days for SLT and 25 (IQR 49–82) days for Psych. Thus, patients received therapy on only 20–60% of the days on which they needed it. As reflects the multidisciplinary nature of stroke rehabilitation, most patients (87%) required input from two or three therapy disciplines.

The average duration of an inpatient treatment session (see Table 3) varied from 31 minutes for SLT to 42 minutes for Psych. However, inpatients received therapy infrequently. On average, patients received PT on only 5 days, averaging 14 minutes per day of inpatient stay, and only one session of Psych, which amounted to an average of 2 minutes per day of inpatient stay.

During community-based stroke therapy (see Table 3), the proportion of patients deemed to require each therapy was lower in PT, OT and SLT than for inpatient therapy, but higher for Psych. For all therapies, patients tended to receive community-based therapy less often (18% of stay for PT, 14% for OT, 9% for SLT and 3% for Psych); however, the average duration of a treatment session (47–51 minutes) and the average amount of therapy per day of ‘stay’ were similar, ranging from 12.0 minutes per day of ‘stay’ for Psych to 13.8 minutes per day of ‘stay’ for PT, indicating that patients tended to receive longer, less frequent treatment in the community than as an inpatient.

There were also differences in the average amount of therapy per day of stay between stroke team types (Table 4), with NRATs and NAITs generally providing a greater amount of therapy per day of stay than the RATs (RATa and RATc).

There were marked differences in the average amount of therapy per day of stay provided in different types of community-based stroke teams (see Table 4), with ESD teams providing much more therapy per day of stay than community rehabilitation or integrated rehabilitation team, for whom the amounts of therapy per day of stay were similar. The average amount of therapy per day of stay provided by ESD teams was less than that provided during inpatient stroke therapy.

Processes of care

The SSNAP records some therapy-related processes of care regarding MDT meetings, goal-setting and discharge planning. These are detailed in Table 5, after they have been matched to the relevant patient-level data and split into inpatient and community-based teams. Note, there was a high degree of missing data (as reporting was voluntary), so the value represents the number of the patients in inpatient teams for whom ‘yes’ was recorded (i.e. these processes of care took place). All but 2% had a MDT meeting to discuss and plan their care, with just under 100% indicating a physiotherapist, occupational therapist, doctor and nurse attended meetings, and 88%, 60% and 75% reporting SLT, social worker and ESD member were present, respectively. Only 38% had a psychologist present at the meetings. Of these meetings for community care teams, most included a physiotherapist, occupational therapist, speech and language therapist, nurse and a member of the ESD team (87%, 88%, 76%, 52% and 53%, respectively); however, psychologists, social workers and doctors attended less frequently (31%, 9.8% and 17%, respectively). Of the patients for whom it was recorded, 77% had rehabilitation goals identified and 74.8% had a joint health and social care plan in place.

The therapy workforce and models of service delivery

Details of the therapy workforce were extracted from the acute and post-acute organisational audits, which do not collect the same information for all types of stroke team. Thus, we cannot present the same information regarding therapy workforce for both inpatient and community-based teams.

For acute inpatient teams, that is RATa, RATc and NRATs, nearly all (n = 178, 97.3%) had access to a social worker within 5 days of referral: less than two-thirds (n = 112, 61.2%) had access to clinical Psych and almost three-quarters had access to an ESD and community stroke rehabilitation team [n = 135 (74%) and n = 128 (70.0%), respectively]. Most teams (n = 100, 56.4%) only provided therapy on weekdays: 11% (n = 21) provided one therapy in an extended service (i.e. over 6 or 7 days per week) and one-third of teams (n = 62, 34%) provided an extended service of two or more therapies. The therapies provided and the days on which they are available are not specified in SSNAP.

Inpatient therapy and nursing staffing levels reported in the 2014 acute organisational care audit24 are shown in Table 6. There are wide variations between the minimum and maximum staffing levels for each profession and the deployment of therapy and support workers.

Staffing levels for community-based teams are shown in Table 7 and are quoted as WTEs per 100 referrals, like inpatients they showed considerable variation in staffing levels but the numbers of doctors and psychologists were low (or non-existent) in most teams. Most teams appeared to include physiotherapists, occupational therapists and speech and language therapists. The staffing levels for nurses, social workers and support staff in community-based stroke teams are not recorded in SSNAP.

Discussion

In this chapter we provided a detailed description of inpatient and community-based stroke therapy, the quality of processes of care and the stroke therapy workforce which will act as a benchmark for service evaluation and development. We found that timely therapy assessments were the norm for PT, OT and SLT, and nearly all patients who required therapy received some therapy. However, although approximately three-quarters of patients received a screening assessment for their mood and cognition at some point during their inpatient stay, and given that the incidence of emotional and cognitive disorders after stroke is around 30%,33,34 the number who were thought to need and then received Psych was implausibly low at 5%. This suggests that other members of the MDT may underestimate the need for Psych input when services are not available. When patients did receive Psych, this was generally a single treatment session lasting around 45 minutes when an inpatient, and two treatment sessions when community based. This indicates that most patients received a detailed assessment without ongoing treatment. This is clearly a suboptimal situation. We also found staffing levels for psychologists were extremely low (approximately half of the levels recommended for hyper/acute stroke units in the national clinical guideline for stroke35) and few had any access to Psych services. This lack of access and inadequate staffing levels are an obvious explanation for the low level of recognition of need and Psych input. Another possibility is that other members of staff are providing this treatment. For example, occupational therapists often assess and treat cognitive problems. Patient feedback and recent policy initiatives have highlighted the impact of emotional and cognitive difficulties on ‘life after stroke’, the frequency with which their needs are unmet, and the need for increased access to support for emotional and cognitive difficulties. 36 The results of this project further illustrates this need and suggests that improving Psych staffing levels and access to services may be one way to address this issue. Further research is needed to establish the most effective way to treat emotional and cognitive difficulties, and the most effective way to provide these services.

Very few patients received therapy daily, or even on every weekday, and so the amount of therapy per day of stay was well below the recommended levels of 45 minutes of each relevant therapy per day (according to need and capacity). 35 The number of days on which patients received therapy was also low (ranging from 1 day for inpatient Psych to 7 days for community-based PT). Given that the average LOS for inpatients was 11 days and 41 days for community-based therapy, then most patients received very little therapy, not only because the treatment was too short but also too infrequent.

This was further illustrated by the paucity of ‘extended’ therapy services, that is provision of therapy outside the usual working week (i.e. 7 hours/weekday). Fewer than half of stroke teams offered an extended service and when this did occur, only one profession was available in most cases. The SSNAP does not record the professions involved or the nature of the therapy provided in extended services. There is some evidence that weekend PT and/or OT can reduce LOS, but does not appear to improve recovery in terms of disability. 37 In addition, centralised hyperacute stroke services, which generally involve extended therapy provision, provide better quality care in terms of rapid therapy assessment than un-centralised services, which do not generally provide an extended service. 38,39 However, we also noted differences between types of stroke teams in the amount of therapy provided per day of stay. Teams which did not, or rarely admitted stroke patients (NRATs and NAITs) appeared to provide more inpatient therapy on average than routinely admitting teams (RATa and RATc), and ESD teams provided more therapy than community or integrated rehabilitation teams, but less than inpatient teams. A key element of ESD, which is considered an important contributor to the superiority of ESD services over rehabilitation in hospital,40,41 is that ESD services should provide a similar amount of therapy (and other care) as would be provided in hospital. 42 The results presented here suggest that this may not be achieved when ESD is implemented in real-world practice, and the amount of therapy per day of stay fell far below the levels recommended in national guidance. 35 The ReAcT (Why do stroke survivors not receive Recommended amounts of AcTive therapy?) study has recently investigated the factors influencing therapy provision during inpatient care, highlighting the impact of organisational factors, such as therapists’ time management and information exchange. 43 Our results suggest that a similar approach to investigate how therapy is organised, what is actually delivered during ESD and community rehabilitation, and the factors influencing the amount of therapy is delivered, is warranted.

An obvious candidate to influence the amount of therapy provided is staffing levels. There is, however, little evidence to guide recommendations regarding therapy staffing levels. The 2012 and 2016 national clinical guidelines for stroke18,35 make recommendations for acute and hyperacute stroke units (HASUs) (based on the work to centralise hyperacute stroke services in London and Manchester, which showed improved outcomes compared with non-centralised services). 44,45 We found that average staffing levels were around the recommended levels for OT, PT and SLT, but also found wide variations for each profession and the deployment of therapy and support workers, suggesting that some services were understaffed and some may be considered overstaffed. The national clinical guidelines for stroke18,35 do not make any recommendations for staffing levels for rehabilitation or community-based teams. We have been unable to find any source of recommendation for specialist inpatient teams or CRTs, but an international consensus group for ESD services agreed recommendations for staffing in ESD teams. 30 They recommend one WTE per 100 referrals for physiotherapists and OT, 0.4 WTE per 100 referrals for speech and language therapists and 0.1 WTE per 100 referrals for doctors. They do not make a recommendation for psychologists. The median figures in the current results are similar to these recommendations, but, like inpatient staffing levels, they are varied; thus, some teams were understaffed and some may be considered overstaffed.

The relationship between staffing levels and the amount of therapy provided is not as straightforward as many would assume. Over a decade ago, a comparison of stroke rehabilitation in four European countries showed that UK stroke patients received less therapy and had poorer outcomes but higher therapy staffing levels than those in Germany and Switzerland, even when confounding variables were controlled. 9,17,46 The UK’s low dose of therapy was considered to be due to poor organisation, rather than lower staffing levels. 16–19 More recently, the ReAcT study highlighted that, although therapy staffing levels undoubtedly played a part in the amount of therapy delivered, they were not the main determinant. 43 The most significant factor influencing the amount and frequency of therapy was the way therapy was organised, specifically the time spent in information exchange and use of individual patient therapy timetables. Units that delivered (relatively) high doses of therapy had reorganised their service specifically to increase the amount of therapy provided. Further research and service improvement work needs to focus not only on increasing the amount of therapy patients receive, but also on the frequency with which they receive it, and the best way of organising and resourcing the therapy workforce to deliver it. Further research is needed to establish how to achieve this most effectively within each patient’s needs and capabilities.

Interestingly, few patients were considered to require therapy for all of their inpatient stay and one would expect a patient who needed therapy to be discharged once it was no longer necessary. The apparent gap between no longer needing therapy and discharge may be because other matters were preventing discharge such as medical problems, delays to getting community-based care in place, ineffective discharge planning, or therapists’ lack of ambition or low expectations of their patients’ potential abilities. Further research is needed to investigate the causes of this apparent delay in discharge and how to overcome it.

The SSNAP’s records regarding therapy and rehabilitation processes of care were often incomplete. However, the teams which recorded these items showed good completion routes: all inpatient teams set rehabilitation goals; and approximately three-quarters assessed continence and created a management plan, developed joint health and social care plans, and held regular MDT meetings to monitor and plan care. However, it should be noted that the SSNAP merely notes whether or not these activities took place, but not their content or efficiency. It is also possible that the proportion of teams that did not record these activities were not completing them, so if the whole stroke population is considered, access to these aspects of rehabilitation may be lower.

Introduction

Having made a detailed description of the Sentinel Stroke National Audit Programme: Investigating Stroke Therapy (SSNAPIEST) cohort and therapy provided, our next objective was to explore the stroke therapy pathways. We aimed to identify the different stroke therapy pathways and characterise them and the patients who followed them, and calculate their costs. The research questions were as follows: Which stroke therapy pathways are used? What characteristics do the patients who follow them have? How much therapy do patients receive in each pathway? What does each pathway cost?

Real-world stroke services typically involve several teams providing different stages of care (such as acute, rehabilitation and community-based teams), which can be configured in a variety of ways. It was anticipated that the route patients took through these services would be varied and could depend not only on personal factors, such as the severity of a patient’s stroke, but also on the way that stroke therapy was organised and the way that services were configured, hence our desire to investigate those routes. Our initial intention was to define stroke therapy pathways according to detailed information about therapy service delivery, such as whether or not an ‘extended’ service was available, staffing levels, involvement of therapy support staff and availability of community-based therapy. However, the information recorded in the SSNAP, particularly for post-acute stroke services, was insufficiently detailed and complete, and the services provided were too varied to enable this. Instead, we defined the pathways according to the type of stroke team(s) that treated the patient.

Method

The SSNAP database contains one or more ‘entries’ per patient to represent their ‘admission’ to the initial hospital stroke team, ‘transfers’ to other inpatient stroke team(s) and ‘discharge’ from inpatient care, which may be followed by referral to an ESD team or CRT. The ‘route’ a patient experienced through stroke services is the distinct combination of types of inpatient and community-based teams from which ‘pathways’ were identified.

The route each patient took through their stroke services was described using the terminology outlined in Chapter 2 for each inpatient admission and transfer(s) with or without community rehabilitation. Patients taking similar routes were grouped to identify common pathways. To identify common routes, the study team, advisory group and external experts from all relevant professions with clinical and academic expertise in stroke care identified a series of rules. Patients:

• who had been discharged from hospital were collapsed into two groups: those who did and did not receive community-based rehabilitation

• whose first admission was not to a RAT (1.1%) were collapsed and referred to as ‘other’ admissions

• with two or more inpatient entries were collapsed into a ‘one or more transfers’ group.

Having identified the most common routes, the patients following these were characterised in terms of their demographics, stroke characteristics, therapy received and outcomes using simple standard descriptive statistics.

The average costs of each route were calculated using the NHS Reference Costs 2014 to 2015,48 to compute the costs associated with inpatient stroke care, the costs associated with community-based rehabilitation were computed using the SSNAP cost and cost-effectiveness analysis. 48,49 The reference cost collection is the single national collection of service costs within the NHS. This collection reports the average unit cost to the NHS provider for each currency or spell of health care in England in a given financial year. 50 They include direct, indirect and overhead costs, and emphasise the cost of delivering the service. They do not provide information on the variation of costs between patients receiving the same health-care activity, nor the location of the service or the funding streams used to recover these costs. 50,51

To calculate the costs of inpatient stroke care, the average cost of non-elective stays, LOS with each stroke team and the pathway were used for each patient. It is important to note that when a patient was transferred to a new hospital, this is considered a new spell of stroke care and its respective average cost was added to the total average cost for that patient. However, transfers between different stroke teams within the same hospital were considered part of the same spell of care. If the patient was discharged and then readmitted to inpatient care, this was considered a new spell of care. The costs associated with community-based stroke care were calculated from the type and amount of therapy received by patients treated by an ESD team or CRT (see Appendix 1) and the average cost per patient applied. 49 Information on the cost per visit or per hour was taken from the SSNAP cost-effectiveness analysis and assuming that patients had one visit of each therapy on the days they received treatment. 49 The cost for psychological therapy was computed per hour. 49 Inpatient and community care costs were computed for each patient using the information in Appendices 1 and 2, and averaged per pathway.

Results

Defining the pathways

The data included 115,247 individual entries by the 94,905 patients. Eighteen per cent of patients had more than one inpatient entry, 15.6% had two inpatient entries and one patient had eight inpatient entries, the maximum number observed.

Eight hundred and seventy-four distinct routes were initially identified, of which 75% of patients were located in the 20 most common routes and 500 routes involved five patients or fewer. In 59.9% of cases, stroke patients were admitted to a RATa and 39% to a RATc. Initial collapsing of patient groups resulted in 42 routes (see Appendix 3), further consolidation using the rules detailed above identified nine common routes, which formed four pathways. Figure 1 depicts the flow of stroke patients through stroke services.

FIGURE 1.

The flow of patients through stroke services, illustrating the main stroke care pathways and routes (note, the size of the arrows depict the number of patients in each pathway).

The common pathways involved patients who were:

• discharged directly from their admitting stroke team, referred to as a the direct discharge pathway [n = 48,972 (52%), routes 1 and 2]

• transferred to community-based rehabilitation on discharge from hospital (from either the admitting or transferred inpatient team), referred to as the community rehabilitation pathway [n = 44,978 (47.2%), routes 3–6]

• transferred from the admitting team to further inpatient team(s), referred to as the inpatient transfer pathway [n = 17,766 (18.7%), routes 5–8].

An additional pathway involving all the remaining other routes was referred to as the other pathway [n = 1155 (1.2% of the whole cohort), route 9]. (Note, routes 5 and 6 appear in both the community rehabilitation and the inpatient transfer pathways.)

The routes are detailed in Table 8.

TABLE 8 - Characteristics of the final common pathways and routes

Note

Routes 5 and 6 appear in both the community rehabilitation and the inpatient transfer pathways.

Pathway	Route number	First inpatient admission	Number of transfers to further inpatient care	Community rehabilitation after discharge	Number of patients (%)
Direct discharge pathway	1	RATa	0	No	21,647 (22.8)
	2	RATc	0	No	19,779 (20.8)
Community rehabilitation pathway	3	RATa	0	Yes	17,308 (18.2)
	4	RATc	0	Yes	17,250 (18.2)
	5	RATa	≥ 1	Yes	8410 (8.9)
	6	RATc	≥ 1	Yes	1810 (1.9)
Inpatient transfer pathway	7	RATa	≥ 1	No	6479 (6.8)
	8	RATc	≥ 1	No	1067 (1.1)
Other pathway	9	All others	≥ 1	Yes or no	1155 (1.2)
Total					94,905

Table 9 summarises the patients who followed the direct discharge pathway. Similar numbers followed route 1 i.e. were admitted to an acute stroke team, n = 21,647, 22.8% of the whole cohort) or route 2 (admitted to a combined stroke team n = 19,779, 20.8% of the whole cohort). Patients who followed routes 1 and 2 had similar demographics, stroke characteristics, need for stroke therapy, amounts of therapy, mortality and cost of care. However, the average LOS differed. It was a median 8.7 (IQR 5–25) days for patients who followed route 1 and a median of 10.2 (IQR 5–31) days for those who followed route 2.

TABLE 9 - Description of patients who followed the direct discharge pathways

Note

Percentages refer to the percentage of the whole cohort.

Characteristic	Route 1 (acute admitting team)	Route 2 (combined admitting team)
Age (years), mean (SD)	78.0 (13.1)	77.5 (13.1)
Sex, n (%)
Female	11,867 (24.1)	10,658 (21.7)
Male	9780 (21.4)	9121 (20.0)
Ethnicity, n (%)
Asian	424 (15.9)	340 (12.7)
Black	200 (14.7)	130 (9.5)
Mixed	44 (15.0)	48 (16.3)
Unknown	1211 (26.2)	487 (10.5)
Other	205 (18.0)	80 (7.0)
White	19,563 (23.1)	18,694 (22.0)
Social deprivation (lowest–highest quartile), n (%)
1 (least deprived)	4519 (20.6)	4199 (19.2)
2	5478 (24.5)	3950 (17.7)
3	5402 (24.0)	4515 (20.0)
4 (most deprived)	4805 (23.4)	4517 (22.0)
Missing	1443 (19.0)	2598 (34.3)
Stroke severity (NIHSS), median (IQR)	6 (3–15)	6 (2–14)
Mild (< 5), n (%)	7499 (20.6)	7699 (21.2)
Moderate (5–14), n (%)	7978 (21.3)	6767 (18.0)
Severe (15–20), n (%)	2833 (27.0)	2450 (23.3)
Very severe (> 20), n (%)	3337 (31.8)	2863 (27.3)
Independence pre stroke (mRS ≤ 2), n (%)	15,657 (20.9)	14,653 (19.5)
Number (%) who had a haemorrhage	19,119 (22.8)	17,580 (20.9)
Number (%) who needed therapy
PT	19,195 (88.7)	17,549 (88.7)
OT	17,655 (81.6)	16,073 (81.3)
SLT	11,888 (54.9)	10,633 (53.8)
Psych	661 (3.1)	848 (4.3)
Amount of therapy (minutes/day of stay), mean (SD)
PT	12.2 (9)	12.4 (10)
OT	11.2 (10)	11.3 (9)
SLT	6.9 (6)	7.3 (6)
Psych	3.0 (4)	1.9 (3)
LOS (days) if survived, median (IQR)	8.7 (5–25)	10.2 (5–31)
Mortality (for those surviving > 3 days), n (%)	5853 (27.0)	5393 (27.3)
Average inpatient costs (£/patient)	5461.40	5300.80

Severity of stroke showed a different pattern in this pathway, compared with the other pathways (Tables 10 and 11). The median stroke severity was 6 (on NIHSS), indicating that, on average, patients had a moderate stroke. However, this pathway involved a higher proportion of patients who had a severe or very severe stroke, as well as more people with mild strokes than the other pathways. Subsequently, a somewhat smaller proportion of patients in this pathway required therapy than the other routes, and those who needed therapy tended to receive less. The mortality rates were also higher than in the other pathways (see Tables 9–11). The average cost of inpatient stroke care with an acute team (route 1) was a little higher than for those admitted to a combined team (route 2), which may reflect that a higher proportion was receiving hyperacute care.

TABLE 10 - Description of patients who followed the community rehabilitation pathway

Notes

Percentages refer to the percentage of the whole cohort.

Routes 5 and 6 fall in to both the inpatient transfer and the community rehabilitation pathways.

Characteristic	Discharged from admitting team		Transferred to another inpatient stroke team
	Route 3 (acute admitting team)	Route 4 (combined admitting team)	Route 5 (acute admitting team)	Route 6 (combined admitting team)
Age (years), mean (SD)	74.0 (13.1)	74.2 (12.9)	73.8 (13.4)	74.1 (12.7)
Sex, n (%)
Female	8478 (17.2)	8449 (17.2)	4033 (8.2)	885 (1.8)
Male	8830 (19.3)	8801 (19.3)	4377 (9.6)	925 (2.0)
Ethnicity, n (%)
Asian	474 (17.8)	379 (14.2)	570 (21.4)	15 (0.6)
Black	177 (13.0)	94 (6.9)	408 (29.9)	8 (0.6)
Mixed	49 (16.7)	50 (17.0)	45 (15.3)	4 (1.4)
Unknown	951 (20.6)	554 (12.0)	677 (14.7)	76 (1.6)
Other	140 (12.3)	75 (6.6)	331 (29.0)	9 (0.8)
White	15,517 (18.3)	16,098 (19.0)	6379 (7.5)	1698 (2.0)
Social deprivation (lowest–highest quartile), n (%)
1 (least deprived)	4179 (19.1)	4466 (20.4)	2185 (10.0)	342 (1.6)
2	4289 (19.2)	3757 (16.8)	2295 (10.3)	397 (1.8)
3	4267 (18.9)	4055 (18.0)	2022 (9.0)	423 (1.9)
4 (most deprived)	3929 (19.2)	3745 (18.3)	1592 (7.8)	339 (1.7)
Missing	644 (8.5)	1227 (16.2)	316 (4.2)	309 (4.1)
Stroke severity (NIHSS), median (IQR)	5 (2–9)	5 (2–9)	7 (4–13)	7 (4–14)
Mild (< 5), n (%)	7895 (21.7)	8003 (22.0)	2626 (7.2)	501 (1.4)
Moderate (5–14), n (%)	7300 (19.5)	6908 (18.4)	3922 (10.5)	866 (2.3)
Severe (15–20), n (%)	1229 (11.7)	1307 (12.4)	1041 (9.9)	258 (2.5)
Very severe (> 20), n (%)	884 (8.4)	1032 (9.8)	821 (7.8)	185 (1.8)
Independence pre stroke (mRS ≤ 2), n (%)	15,042 (20.0)	14,783 (19.7)	7080 (9.4)	1620 (2.2)
Number (%) who had a haemorrhage	15,615 (18.6)	15,482 (18.4)	7127 (8.5)	1553 (1.9)
Number (%) who needed therapy
PT	16,680 (96.3)	16,643 (96.4)	17,012 (94.1)	3571 (96.4)
OT	16,611 (95.9)	16,528 (95.8)	16,893 (93.4)	3453 (93.2)
SLT	9437 (54.5)	9742 (56.5)	11,854 (65.5)	2286 (61.7)
Psych	891 (5.1)	1242 (7.2)	2091 (11.6)	559 (15.1)
Amount of inpatient therapy (minutes/day of stay)
PT	16.3 (10)	16.1 (11)	21.0 (10)	19.9 (9)
OT	17.8 (12)	16.5 (11)	20.3 (11)	16.4 (9)
SLT	9.1 (8)	9.1 (8)	11.9 (9)	9.1 (7)
Psych	3.4 (4)	2.4 (3)	5.0 (6)	2.6 (3)
Amount of community therapy (minutes/day of stay)
PT	12.6 (13)	11.2 (12)	11.9 (12)	12.1 (10)
OT	11.9 (14)	10.0 (13)	10.4 (14)	9.0 (9)
SLT	8.7 (11)	6.8 (8)	7.3 (12)	7.2 (8)
Psych	2.1 (3)	2.7 (4)	1.8 (3)	2.2 (2)
Length of inpatient stay (days), median (IQR)	8.6 (5–19)	13.5 (7–31)	30.9 (15–56)	54.2 (33–79)
Mortality	93 (0.5)	123 (0.7)	112 (0.13)	13 (0.07)
Average inpatient costs (£/patient)	5082.80	5025.20	11,516.50	12,730.50

TABLE 11 - Description of patients who followed the inpatient transfer pathways

For those who survived > 3 days.

Notes

Percentages refer to the percentage of the whole cohort and, therefore, do not add up to 100% in this table.

Routes 5 and 6 fall in to both the inpatient transfer and the community rehabilitation pathways.

Characteristic	Discharged with community rehabilitation		Discharged without community rehabilitation
	Route 5 (acute admitting team)	Route 6 (combined admitting team)	Route 7 (acute admitting team)	Route 8 (combined admitting team)
Age (years), mean (SD)	73.8 (13.4)	74.1 (12.7)	78.3 (13.1)	77.7 (12.7)
Sex, n (%)
Female	4033 (8.2)	885 (1.8)	3635 (7.4)	583 (1.2)
Male	4377 (9.6)	925 (2.0)	2844 (6.2)	484 (1.1)
Ethnicity, n (%)
Asian	570 (21.4)	15 (0.6)	367 (13.8)	12 (0.4)
Black	408 (29.9)	8 (0.6)	255 (18.7)	4 (0.3)
Mixed	45 (15.3)	4 (1.4)	35 (11.9)	3 (1.0)
Unknown	677 (14.7)	76 (1.6)	592 (12.8)	38 (0.8)
Other	331 (29.0)	9 (0.8)	274 (24.0)	5 (0.4)
White	6379 (7.5)	1698 (2.0)	4956 (5.8)	1005 (1.2)
Social deprivation (lowest–highest quartile), n (%)
1 (least deprived)	2185 (10.0)	342 (1.6)	1498 (6.8)	121 (0.6)
2	2295 (10.3)	397 (1.8)	1784 (8.0)	146 (0.7)
3	2022 (9.0)	423 (1.9)	1452 (6.4)	176 (0.8)
4 (most deprived)	1592 (7.8)	339 (1.7)	1222 (6.0)	151 (0.7)
Missing	316 (4.2)	309 (4.1)	523 (6.9)	473 (6.2)
Stroke severity (NIHSS), median (IQR)	7 (4–13)	7 (4–14)	9 (5–17)	8 (4–15)
Mild (< 5), n (%)	2626 (7.2)	501 (1.4)	1425 (3.9)	262 (0.7)
Moderate (5–14), n (%)	3922 (10.5)	866 (2.3)	2859 (7.6)	466 (1.2)
Severe (15–20), n (%)	1041 (9.9)	258 (2.5)	1080 (10.3)	190 (1.8)
Very severe (> 20), n (%)	821 (7.8)	185 (1.8)	1115 (10.6)	149 (1.4)
Independence pre stroke (mRS ≤ 2), n (%)	7080 (9.4)	1620 (2.2)	4548 (6.1)	870 (1.2)
Number (%) who had a haemorrhage	7127 (8.5)	1553 (1.9)	5537 (6.6)	922 (1.1)
Number (%) who needed therapy
PT	17,012 (94.1)	3571 (96.4)	12,523 (90.1)	1959 (89.5)
OT	16,893 (93.4)	3453 (93.2)	11,850 (85.3)	1809 (82.6)
SLT	11,854 (65.5)	2286 (61.7)	9433 (67.9)	1283 (58.6)
Psych	2091 (11.6)	559 (15.1)	1059 (7.6)	206 (9.4)
Amount of inpatient therapy (minutes/day of stay)
PT	21.0 (10)	19.9 (9)	17.2 (9)	14.9 (9)
OT	20.3 (11)	16.4 (9)	14.9 (10)	11.7 (9)
SLT	11.9 (9)	9.1 (7)	10.1 (8)	7.1 (7)
Psych	5.0 (6)	2.6 (3)	4.8 (6)	1.9 (2)
Length of inpatient staya (days), median (IQR)	30.9 (15–56)	54.2 (33–79)	38.2 (16–67)	57 (30–97)
Mortality,a n (%)	112 (0.13)	13 (0.7)	1543 (23.8)	174 (2.6)
Average inpatient costs (£/patient)	11,516.5	12,730.5	11,239.1	12,515.1

Community rehabilitation pathway

The community rehabilitation pathway involved 44,778 patients (47.2% of the whole cohort) and four routes. Patients who were discharged to community rehabilitation directly from their admitting team formed routes 3 [if discharged from an acute team, n = 17,308 (18.2%)] and 4 [if discharged from a combined team, n = 17,250 (18.2%)]. Patients who were transferred to another inpatient team before discharge with community rehabilitation formed routes 5 (n = 8410, 8.9%) and 6 (n = 1810, 1.9%). Overall, patients in this pathway had the lowest mortality rates, the mildest strokes (on average) and were most frequently independent before their stroke. This was particularly noticeable for patients in routes 3 and 4 (discharged to community rehabilitation from the admitting team). Patients following routes 5 and 6 (community rehabilitation after an inpatient transfer) were more severely affected.

Despite having relatively mild strokes, patients in the community rehabilitation pathway had the highest demand for therapy of all the pathways. Nearly all patients required PT and OT, whereas the demand for SLT and Psych was 54.5–56.5% and 5.1–15.1%, respectively. They also received relatively large amounts of therapy compared with the other pathways. This was not only because they received therapy after hospital discharge, but also while an inpatient. The amount of community therapy received was similar in all the routes and less than while an inpatient.

Again, there was a marked difference in length of inpatient stay for patients originally admitted to an acute or a combined stroke team. Routes 3 and 5 (admitted to an acute stroke team) had a median LOS of 8.6 and 30.9 days respectively, whereas for routes 4 and 6 median LOS was 14 and 54 days, respectively. A high proportion of patients who suffered an intracerebral haemorrhage followed route 5. The average costs of care in the community rehabilitation pathway (routes 3 and 4) were lower than equivalent routes without community (routes 1 and 2), despite a longer LOS. The costs of community rehabilitation plus acute care (route 3, £5083) was slightly higher than that from a combined team (route 4, £5025), but the reverse was seen in the routes that also included transfer to another inpatient team (routes 5 and 6), here being originally admitted to a combined team involved higher cost (£11,517 and £12,731, respectively).

Inpatient transfer pathway

Nineteen per cent of the cohort (n = 17,766) were treated by more than one inpatient stroke team and followed the inpatient transfer pathway, 80% of whom (n = 14,213) were transferred only once before discharge. The inpatient transfer pathway involved four routes in which patients were discharged from stroke care with (route 5 if originally admitted to an acute team and route 6 if originally admitted to a combined team) or without community rehabilitation (routes 7 and 8, respectively). Route 5 involved 8410 patients (8.9% of the whole cohort); route 6 involved 1810 patients (1.9% of the whole cohort); route 7 involved 6479 patients (6.8% of the whole cohort) and route 8 involved 1067 patients (1.1% of the whole cohort). Routes 5 and 6 appeared in both the community rehabilitation and the inpatient transfer pathways.

The patients’ characteristics for the routes in the inpatient transfer pathway are detailed in Table 11. Little difference was seen in the patients’ demographics and stroke characteristics between the routes, except that patients following routes 7 and 8 (discharged without community rehabilitation after inpatient transfer) tended to be older and have a more severe stroke than those following routes 5 and 6 (discharged with community rehabilitation after inpatient transfer). Patients following routes 7 and 8 less frequently required any of the therapies (expect SLT) and, for those who needed it, received less therapy than routes 5 and 6. Like the other pathways, patients who were originally admitted to a combined stroke team (routes 6 and 8) had a much longer LOS than those admitted to an acute team (routes 5 and 7). This was regardless of whether they received community rehabilitation on discharge from hospital, or not. For routes 5 and 7 the LOS was 4–5 weeks (31 days and 38 days, respectively),whereas for routes 6 and 8 it was approximately 7–8 weeks (54 and 57 days, respectively). There was a marked difference in mortality between the routes in the inpatient transfer pathway. It was low in routes 5 and 6 (in which the patient received community rehabilitation), but higher in routes 7 and 8, particularly route 7 (23.8%), which had a similar mortality to routes 1 and 2 (the direct discharge pathway).

The average costs of all routes are shown in Table 12, with further details broken down by severity of stroke in Report Supplementary Material 1. Differences across pathways were explained by differences in stroke severity between the pathways, number of transfers between inpatient teams (which is zero for routes 1–4) and the amount of community therapy (zero for routes 1, 2, 7 and 8). The multiteam, more complex pathways had a higher proportion of severely disabled patients and cost more than the simpler pathways with patients with milder strokes.

TABLE 12 - Average costs of inpatient and community-based stroke care costs per pathway

Note

Routes 5 and 6 fall in to both the inpatient transfer and the community rehabilitation pathways.

Pathway		Route number	Number of patients	% of patients who were independent (mRS ≤ 2) at final discharge	Average inpatient cost/patient (£)	Average community therapy cost/patient (£)	Total average cost/patient (£)
Direct discharge pathway		1	21,646	38	5461.40	0.00	5461.40
		2	19,779	40	5300.80	0.00	5300.80
Community rehabilitation pathway		3	17,308	58	4617.00	465.80	5082.80
		4	17,250	54	4646.10	379.10	5025.20
	Inpatient transfer pathway	5	8410	36	10,601.70	914.80	11,516.50
		6	1810	34	11,774.60	955.90	12,730.50
		7	6479	20	11,239.10	0.00	11,239.10
		8	1067	25	12,515.40	0.00	12,515.40
Other pathways		9	1155	36	5769.80	972.60	6742.40

Discussion

To address our objective to examine and compare different stroke care pathways, we defined models to deliver stroke services based on admission rates, LOS and access to community stroke services. Even with this simple approach, > 800 different routes were recorded; however, nine common routes and four pathways emerged. Patients were most likely to be admitted to either an acute stroke team (which might include a hyperacute team) or a combined acute and rehabilitation team. Most stroke patients stayed with only one inpatient team. Almost half of these patients were referred to a ESD team or a CRT. Of the remaining patients, most were transferred from the admitting stroke team to another inpatient stroke team. Typically, this was a NRAT (a combined acute and rehabilitation unit which received patients from a hyperacute unit but also other acute admissions), or a specialist stroke rehabilitation unit (or non-admitting inpatient unit, NAIT), before discharge, with or without community-based rehabilitation (routes 5 and 6, and 7 and 8, respectively). The final pathway (route 9) included the 1% of patients who were treated by more stroke teams (up to eight).

There was a high proportion of patients in the direct discharge pathway with mild strokes and a short LOS. There is growing recognition that people with ‘mild’ strokes often suffer enduring impairments, but the limitations they impose on activity and participation only become apparent once patients are discharged and attempt to function within their own environment. 52–54 This highlights the need for all stroke survivors to be actively monitored and supported after discharge, with easy, rapid access to rehabilitation services when needed.

Patients who followed the community rehabilitation pathway had the mildest strokes and were least frequently dependent before their stroke. There may be several possible explanations for this. First, patients who were dependent pre morbidly, may already have the facilities in place to provide care compared with those who were previously independent and, hence, have less need for community rehabilitation. Second, as pre-morbid disability is a predictor of poor recovery after stroke,55 they may not have been referred for community rehabilitation, as they were thought to have little potential for further recovery. 56

Patients following the community rehabilitation pathway, despite having had a relatively mild stroke, were more likely than those following the other pathways to require therapy and they received the most therapy, even while an inpatient. Patients received less therapy during community-based rehabilitation than while an inpatient. One of the central tenets of ESD services is that the patients should receive similar amounts of therapy as they would have received if they remained as an inpatient. This does not appear to happen in practice. Furthermore, community rehabilitation, particularly ESD, is said to reduce length of inpatient stay and thus costs (which are driven by LOS). 40–42 Thus, one might expect the length of inpatient stay and costs of the community rehabilitation pathway to be less than the other pathways. This was not the case; LOS was similar and costs were slightly less in the direct discharge pathway, and the community rehabilitation pathway was longer and more costly than the inpatient transfer pathway. This could be because patients who required community rehabilitation were subjected to delays in availability of the community team, or a care package which extended their LOS. Further research is needed to understand how community-based rehabilitation, particularly ESD services, is delivered in real life and to establish its cost-effectiveness.

Approximately one-fifth of patients followed the inpatient transfer pathway. Typically, the transfers were to stroke teams that focused on rehabilitation, such as NRATs or a specialist stroke rehabilitation team (or non-admitting inpatient unit). Patients in this pathway had, overall, the most severe strokes and the demand for therapy was high. Although the amount of therapy provided was well below that recommended in national clinical guidelines18 in all pathways, it was relatively high in this pathway. Unsurprisingly, LOS and inpatient care costs were higher than in the other pathways.

Inpatient LOS varied between the pathways: for the direct discharge pathways it averaged 9–10 days, for the community rehabilitation pathway it was 10 days, and for the inpatient transfer pathway it was approximately 1 month if initially admitted to an acute team and 2 months if admitted to a combined team. Most stroke services need to estimate patients’ discharge date soon after admission to the stroke team. These values can be used as a benchmark to guide such estimates for each pathway.

In each pathway, routes involving initial admission to a combined stroke team (routes 2, 4, 6 and 8) had a longer median length of inpatient stay than those initially admitted to an acute team (routes 1, 3, 5 and 7). There are no obvious differences in the patient or stroke characteristics, therapy demand or provision to explain this. It suggests that whether a patient is treated by an acute or combined team is not driven by the patients’ needs (or clinical decision-making), but by managerial choices about how stroke teams are configured. Further research to investigate the causes of the difference in LOS and to compare other outcomes, such as quality of care, satisfaction and cost-effectiveness, is warranted to see whether or not the different configurations are equivalent. We hypothesise that providing all inpatient stroke care in a single combined unit may be preferable, as it could prevent potential delays and disruption caused by transfer between teams. Alternatively, a specific rehabilitation unit, which is separate from the acute stroke unit, may enable patients to receive more therapy (and other aspects of rehabilitation), without distraction from the demands for rapid assessment and discharge, which are given priority in acute stroke care. Further research is needed to investigate the clinical effectiveness and cost-effectiveness of different ways of configuring stroke rehabilitation services and to understand the impact of rehabilitation on the use of NHS resources.

We calculated the costs for the different pathways. These will be a useful baseline for calculating the costs of stroke therapy services in the UK. Unsurprisingly, the more complex multiteam pathways with a longer LOS were the most expensive, but adding community-based therapy to inpatient care appeared to reduce the costs of simple pathways and made a negligible difference to the costs of the complex ones. However, we emphasise that the information provided here regarding the costs of different stroke pathways should not be interpreted as a suggestion that one pathway is more cost-effective or represents an optimal use of resources. This was not a cost-effectiveness analysis, as the patients who followed different pathways were not necessarily comparable, as the pathway followed was not randomly assigned but depended on observable and unobservable characteristics that were not all accounted for in this project. Furthermore, the SSNAP does not report data on the European Quality of Life measure (EuroQol-5 Dimensions) needed to estimate quality-adjusted life-years or disability-adjusted life-years. However, this analysis represents an initial exploration of the resources used in each stroke pathway. Further research is needed to examine the cost-effectiveness of different models of stroke care. The other relevant limitations to this chapter were discussed in Chapters 2 and 3.

Introduction

Having described and defined the stroke therapy provided, the workforce, pathways and costs, our next objective was to identify subgroups of stroke survivors based on their impairments, as the impairment ‘profile’ is likely to influence the therapy that patients receive. For example, SLT will only be relevant for people with communication and/or swallowing problems. The objective addressed in this chapter was to identify and characterise important subgroups of stroke survivors. The research questions were, what is the frequency of stroke-related impairments? Which impairments are commonly comorbid and to what extent? Do patients with common comorbidities receive different amounts of therapy or achieve different clinical outcomes?

The most widely adopted way of classifying stroke is the Oxford Community Stroke Project classification, which defined stroke types according to the location and size of the stroke, and is used to define its severity and prognosis. 57,58 Although highly valuable to guide medical care, these classifications have less to offer when it comes to guiding rehabilitation, as rehabilitation is largely dependent on the impairments that have been caused by the stroke, rather than the original stroke pathology. Surprisingly, there is no widely adopted classification based on stroke-related impairments. In 1991, Sánchez-Blanco et al. 59 used data from 92 patients to classify stroke survivors as having a ‘motor-only’, ‘motor-sensory’ and ‘motor-sensory-hemianopic’ symptoms, and found that these patients formed distinct groups with respect to recovery of independence in activities of daily living, mobility, age, lesion size and location. However, cognition and communication were not included in the classification and the sample was restricted to people referred for rehabilitation. The Southampton Stroke Audit60 involved all people with stroke in a single health district over a 15-month period (1993–4, n = 203) and used a mixture of impairments and activity limitations to classify strokes as:

• very severe (unconscious or deeply drowsy)

• severe (dense hemiplegia causing loss of sitting balance and/or other system impairments without loss of consciousness)

• moderate (motor weakness only causing difficulties with self-care, mobility, continence or balance)

• mild (no or mild residual deficit).

Although a more comprehensive sample than Sánchez-Blanco et al. ,59 the classification did not include individual impairments. We developed stroke impairment categories (SICs) by further developing the groupings suggested by Sánchez-Blanco et al. 59 and the Southampton Stroke Audit60 using the NIHSS score on admission. 28

Method

The individual items from the NIHSS28 were grouped into body systems and, as our focus was on the presence or absence of the impairments rather than their severity, the scores were interpreted as follows:

Results

Figure 2 reports the frequency and percentage of the presence of each individual NIHSS factor on admission in the 94,905 patients still in hospital after 3 days. These were subsequently aggregated in order to define each patient’s stroke impairment type (Figure 3). It shows that the most common impairments were facial palsy (50%) and dysarthria (≈ 47%). Around 30% of patients had a weakness of each of the limbs, with similar numbers with upper and lower limb weakness. It should be noted, however, that SSNAP merely records whether or not there is a weakness in an individual limb. It does not record the patterns of weakness between the limbs, such as whether or not a hemiparesis was present.

FIGURE 2.

The number and percentage of stroke patients with each individual NIHSS factor at baseline admission in those still in hospital after 3 days. LOC, loss of consciousness.

FIGURE 3.

The number and percentage of stroke patients with each body system impairment, as identified by the NIHSS at baseline admission in those still in hospital after 3 days.

Figure 3 reports the frequency and percentage of body system impairments on admission, as identified by the NIHSS. In each case, we report the number for whom any of the impairments within each ‘body system’ were present (but not their severity or how many were present), and regardless of any overlap between different body system impairments. ‘Sensory’ included those with visual impairments; however, those with visual impairments were also reported for comparison. Note, a large amount of overlap was present between visual and sensory impairments. The graph shows the predominance of motor impairments (> 80%) and the high proportion of patients with cognitive and/or communication problems (> 50%). Loss of consciousness affected < 5% of patients.

The geometric coding revealed 31 different combinations of system impairments (detailed in Appendix 4), of which nine involved < 124 individuals (1% of the sample) and six involved < 10 patients. The final SICs were as follows:

• Loss of consciousness and another system impairments, referred to as a loss of consciousness stroke (Loss-Con) (n = 6034, 6.4%).

• Motor + cognitive + senses impairments, referred to as a motor-cognitive-senses stroke (Mo-Co-Se) (n = 28,226, 29.7%).

• Motor + cognitive impairments, referred to as a motor-cognitive stroke (Mo-Co) (n = 16,967, 17.9%).

• Motor + senses impairments, referred to as a motor-senses stroke (Mo-Se) (n = 9882, 10.4%).

• Motor impairments only, referred to as a motor-only stroke (Mot-O) (n = 20,471, 21.6%).

• Any system impairments except motor or loss of consciousness, referred to as a non-motor stroke (No-Mo) (n = 7498, 7.9%).

• No system impairments referred to as a no impairment stroke (None) (n = 5827, 6.1%).

The most common SIC was a Mo-Co-Se (which occurred in 29.7% of patients) and the least common was ‘no impairments’ (which occurred in 6.1% of patients). Details of the patients’ demographics and stroke characteristics in each SIC are detailed in Table 13. There was very little difference in ethnic origins or socioeconomic status in patients in the different categories, but there was a gradation in stroke severity: Loss-Con was most severe and, unsurprisingly, None was the least severe, with the other categories showing a gradation of stroke severity broadly according to the number of impairments (Loss-Con, Mo-Co-Se, Mo-Co, Mo-Se, Mot-O, No-Mo). Associated with this was also a broad gradation indicating that patients in more severe SICs tended to be older, more frequently suffered an intracerebral haemorrhage, were dependent before their stroke and have more comorbidities.

The completeness of screening for therapy during the first 72 hours after stroke was mixed in the different SICs (Table 14). People with the most and least severe SICs (Loss-Con and None, respectively) were least likely to be screened for therapy and were least likely to be considered to need each therapy. Interestingly, the numbers considered to require Psych were very low regardless of whether or not the patient had cognitive impairments. Nearly all patients were screened for, and required, PT and OT (slightly more for PT than OT). The numbers requiring PT tended to reduce slightly with the less severe SICs, but for OT it increased. Unsurprisingly, the most severe SICs (except Loss-Con) more frequently required input from multiple therapies than the less severe SICs.

The therapy received by patients differed in the different SICs and is detailed in Table 15. Overall, patients in the most severe (Loss-Con) and the mildest (None) SICs tended to receive the least inpatient and community-based therapy. Patients in these SICs received 21 and 25 minutes of ‘any therapy’ per day during inpatient and community-based therapy respectively, whereas patients in the other (more moderately severe) SICs received more: ≈30–35 minutes per day of stay as an inpatient and 19–23 minutes per day of stay of community-based therapy. Furthermore, relatively few patients in the most severe and mildest SICs (0.65% and 0.13%, respectively) were referred for community-based therapy. For the other SICs, the pattern of therapy provision varied. The amount of PT increased, with decreasing severity of SIC (i.e. milder SICs received more therapy), except patients with the mildest SICs (Mot-O and No-Mo) who received less therapy (6.5 and 11.7 minutes/day of stay, respectively) than the other SICs. For inpatient OT and Psych, this ‘bump’ in the amount of therapy per day of stay for moderately severe SICs was not seen: the amount of OT and Psych per day of stay increased with decreasing severity of SIC. During community-based OT and Psych, all SICs received similar average amount of therapy per day of stay, regardless of whether they had cognitive impairments. Patients with Mo-Co and No-Mo received the most SLT during both inpatient and community-based therapy, likely reflecting the inclusion of people with aphasia in these SICs.

Outcomes for patients in the different SICs differed and are reported in Table 16. The median length of inpatient stay was greater for those in more severe SICs (37.1 days for Loss-Con, 25.1 days for Mo-Co-Se and 8 days for No-Mo and None). Similarly, mortality was ≈5% for No-Mo and Mot-O, but 49% for Loss-Con. Overall, 14% of all patients were discharged to a care home; unsurprisingly this proportion was higher (≈20%) in the more severe SICs (Loss-Con, Mo-Co-Se and Mo-Co) and ≈ 10% for the other SICs. This was echoed by the proportion of patients who were dependent on discharge, but the differences were less marked. Although 90% of people with a Loss-Con and 77% of people with a Mo-Co-Se were dependent, ≈40% of people with the less severe SICs (Mot-O, No-MO and None) were also dependent. The destination on discharge also varied markedly between SICs, a similar proportion of surviving patients with a Loss-Con were discharged home or to a care home. This proportion shifted in favour of patients being discharged home, as the severity of the SIC reduced.

Discussion

In this chapter, we developed a new way of classifying stroke according to the patients’ stroke-related impairments. We identified seven distinct groups in terms of demographics and stroke characteristics, the therapy received and outcomes. As detailed in Chapters 6 and 7, we also demonstrate that the SICs are important factors associated with the dose of therapy provided and with outcome. There is growing recognition that stroke rehabilitation research and stroke services need a way to stratify patients, one which recognises their highly varied difficulties, needs and recovery patterns, so that more appropriate individualised care can be defined, developed, evaluated and benchmarked. 62 We believe the SICs are a feasible and meaningful way of doing this. The validity of the classification is illustrated by defining the patients, therapy and outcomes in each category; they form different groups. They also have face, content, construct and ecological validity. The reliability of their scoring is dependent on the scoring of the NIHSS, which is good. 63 Future work will examine the scope of using the SICs to predict recovery and outcome.

Descriptions of the therapy provided and outcomes for each SIC can act as a useful benchmark for clinical services. Unsurprisingly, the amount of PT showed a bell-shaped curve, with patients in moderately impaired SICs receiving most therapy, and the people in the most and least impaired SICs receiving less therapy, presumably because they either could not tolerate it or did not need it. However, the amount of SLT and Psych showed less distinct input to different categories. One might expect these professions to provide most input to SICs involving cognitive impairments; however, this was not the case. This may reflect speech and language therapists’ input to people with motor impairments (dysphagia and dysarthria), as well as those with aphasia. However, stroke SLT is dominated by treatment for aphasia, so this is unlikely. A similar pattern was seen in Psych. This may suggest that screening processes to identify people with communication, cognitive and emotional problems fail to identify those with most need. This is reflected in the relatively low number of patients who receive screening assessments for swallowing, communication, emotional and mood, and cognitive problems, compared with PT and OT assessments. Further research and service development is needed to establish, implement and evaluate effective screening and assessment processes. This will require workforce and staffing levels to be addressed for these professions, so that appropriate assessment and treatment can be provided.

A positive finding was the relatively low mortality in patients with the most severe (Loss-Con) SIC. Although almost half of patients in this group died, this is a huge improvement compared with the Southampton Stroke Audit,60 which evaluated outcome for a group with a similar definition, in which 99% of patients died. This reflects the improvements in stroke services, particularly acute care, in the last couple of decades. 64

Methods

Results

Discussion

In this chapter we investigated the factors associated with the amount of inpatient and community-based therapy that patients received per day of stay. We identified that patients with severe disability received less therapy per day of stay than more able patients during both inpatient and community-based therapy. However, we also found apparent disparities in the amount of inpatient therapy received by patients from ethnic minorities and socially deprived groups compared with white and affluent stroke patients, even when confounders are controlled for. This finding needs to be treated with caution, as the proportion of patients from ethnic minority backgrounds was small (≈ 20%) compared with the white population and there were many missing data regarding socioeconomic status. However, as we had access to information about postcodes only for patients in England, most of the missing data were for patients from Wales and Northern Ireland, and it is unlikely that the socioeconomic status of stroke patients from these countries is systematically different from England. The finding that people from some ethnic minorities appear to either have less access to inpatient therapy or take up the available therapy less often than the white population concurs with the body of evidence that shows people from ethnic minorities in the UK and USA often have a poorer outcome and experience poorer quality acute stroke care and rehabilitation than the white population. 68–76 This may be due to a number of factors, including language barriers or cultural differences in family support networks. Further research is needed to better understand the mechanisms behind these possible disparities and how to overcome them.

There is much evidence that more affluent people have a lower incidence of stroke and mortality, less severe strokes and some evidence that the socially deprived patients are less likely to receive stroke rehabilitation or have access to therapy, among other aspects of care. 77,78 Our results concur with and expand these findings, as they are based on population-level data rather than the single or small groups of stroke services involved in previous studies. They highlight the need for further research to improve equity of access to all aspects of effective stroke care.

We also found several modifiable organisational factors that influence the average amount of therapy per day of stay. The day and time of admission may not be considered modifiable, but the services provided throughout the day and week are. Patients admitted to hospital or starting community-based therapy during the main working day tended to receive more therapy per day of stay than those admitted ‘out of hours’. Those admitted to hospital at the beginning of the week or who started community therapy later in the week received the most therapy per day of stay. Several observational studies have reported a greater mortality from stroke and other conditions for people admitted over the weekend and at night (the so-called weekend effect);79,80 however, Bray et al. 81 found this to be an oversimplification and that the weekend effect for stroke was just one of several patterns of variation in the quality of stroke care during the week. The greater stroke mortality observed with out-of-hours admissions was consequently attributed to patients tending to have a more severe stroke at these times. 82 Furthermore, quality of out-of-hours care was attributed to nurse staffing levels, rather than the availability of specialist physicians. 83 Our findings further expand these patterns of care to include therapists.

We postulate that the pattern of variation during the week seen here is a reflection of the availability of therapists, which, for most services (66%), is limited to the normal working week (08.30 to 16.30, Monday to Friday). The presence of an extended (weekend) therapy service was associated with provision of more therapy per day of stay. There is moderate evidence that an extended weekend therapy service can reduce hospital LOS. 37 Our results suggest that it may be due to quicker therapy assessment and the provision of more therapy per day of stay. Further research is needed to establish the most effective model of extended service, including the number and types of therapists needed and the optimal presence during the weekend. A key question is whether or not an effective extended service requires an increase in therapy staffing levels to cover the weekend input or whether or not spreading the existing therapy workforce to cover the whole week, but more thinly, is adequate.

A further modifiable factor which influenced the amount of therapy per day of stay was inpatient staffing levels for both therapist and nurses, and for both qualified staff and support workers. As noted previously, staffing levels were highly varied. Over a decade ago, the CERISE (Collaborative Evaluation of Rehabilitation in Stroke across Europe) study showed that low amounts of therapy in the UK were related to poorer outcomes and were due to the way therapists’ workload was organised, with administration and non-direct contact often being given priority over face-to-face contact with patients. 9,15,17 Since then, NHS funding has fallen and staff shortages have risen,84 such that staffing levels may have been ‘cut to the bone’ and that therapists struggle to provide sufficient therapy. A recent ethnographic study found that although staffing levels influenced the amount of therapy provided, the way in which therapy services were organised also played a part. 43 It found that therapy teams that prioritised treatment over administrative tasks often provided more therapy, in some cases despite relatively low staffing levels. Surprisingly, staffing levels in community-based therapy teams were not associated with the amount of therapy. This may be because staffing levels are not as critically low in the community-based teams as for inpatient teams, or it could be an effect of the high degree of missing data for community-based teams.

How stroke rehabilitation teams operate day to day has received little research attention and there has been little work to investigate, develop and implement effective organisation-level interventions, as most research focuses on the clinical effectiveness of interventions on individual patients’ impairments and activities. Such research is clearly warranted.

Further to staffing levels, the type of stroke team was also associated with the amount of therapy per day of stay. Stroke teams that did not routinely admit stroke patients (NRATs and NAITs) provided more therapy per day of stay than routinely admitting teams (RATas and RATcs). This is to be expected and is probably due to the priority given to rapid assessment and organisation of early discharge over the provision of ongoing ‘rehabilitation-focused’ therapy during acute stroke care. This highlights the potential benefit of providing rehabilitation in a stand-alone specialist rehabilitation unit (i.e. a NAIT) for patients who require ongoing rehabilitation and for whom ESD is inappropriate. The implementation of specialist acute stroke services and centralised hyperacute services have improved outcomes and the quality of acute stroke care,44,45,85 but the provision of inpatient stroke rehabilitation has been neglected. 1,86 Further research is needed to investigate the optimal way to configure stroke rehabilitation services for patients with different needs and levels of ability.

Our findings indicate that the patients treated by an inpatient team with access to a community-based therapy team (ESD team or CRT) tended to receive more therapy per day of stay while an inpatient than those without access to community-based therapy. This may be considered counter-intuitive, as many therapists consider lack of access to the community-based therapy to be a reason to extend inpatient stay so that patients continue to receive the therapy they need. An alternative explanation is that services with access to community-based therapy may be well organised, and possibly better staffed, and so offer more inpatient therapy than teams who do not have access to community-based therapy.

We also examined the factors influencing the amount of community-based therapy per day of stay. The pattern of therapy provision was the same as for inpatients between sexes and for people with differing degrees of stroke severity. There were also some important differences, however. Unlike inpatient teams, patients from some ethnic minority backgrounds tended to receive more PT and OT per day of stay than white stroke survivors, and patients received more therapy per day of stay if they started treatment towards the end of the week rather than earlier. This may reflect that many ESD services follow an extended weekend working pattern. Surprisingly, we did not find that staffing levels or waiting list times for community-based therapy was a factor influencing the average amount of inpatient therapy per day of stay.

Finally, we found that the type of community-based stroke team and the way that they were organised was associated with the amount of therapy provided per day of stay. ESD teams tended to provide more therapy per day of stay than CRTs. This is unsurprising as ESD teams are intended to provide ‘intensive therapy’ for a (relatively) short period after discharge, whereas CRTs generally intend to provide less-intensive, longer-term support. Regular team meetings (twice/week vs. once/week) were associated with patients receiving more PT and OT per day of stay, but less SLT. There is an extensive body of research that shows ESD is effective; however, the detail of how they operate in the ‘real world’ has received less attention. 42,87–91 Further work is needed to develop and evaluate optimal models of delivering community-based therapy, to address the needs of patients with a range of difficulties, their caregivers and different economic and social care contexts.

Method

Only inpatient therapies were included in this analysis, as the sample size for community-based data was too small for the resulting models to be stable. The primary dependent variable to be modelled was disability on discharge from inpatient stroke care. Secondary outcomes were mortality, LOS and destination on discharge (i.e. home or residential care). Multilevel, mixed-effects regression modelled the association between the patient-, organisation- and therapy-related factors (detailed in Chapters 6) on these outcomes, while accounting for both measured and unmeasured (due to clustering) confounders. The average amount of therapy per day of stay was the primary predictor and was included as a linear therapy dose–response term.

Results

Additional exploratory investigation of therapy effect on outcomes

The analysis above showed that although the amount of OT, SLT and Psych per inpatient day of stay was associated with more positive outcomes, the amount of inpatient PT per day of stay was associated with greater disability at discharge and a lower odds of being discharged home. This was unexpected; however, a previous trial93 showed that very early PT, OT and nursing input in the form of early mobilisation within 24 hours of stroke had a negative impact on mortality for some groups of patients. We hypothesised that a non-linear relationship between the amount of therapy per day of stay and outcomes might be present. Therefore, an exploratory analysis was devised that used natural cubic splines to allow the regression model to more flexibly represent the relationship between therapy and outcome, as defined by the data.

Exploratory analysis: flexibly modelling the relationship between inpatient therapy and outcomes using natural cubic splines

Knot points were defined by Harrell95 percentile point and located at approximately 2.1, 7.7, 13.2, 20 and 36 minutes of PT per day of stay. This was consistent for OT, SLT and Psych. Point predictions and corresponding CIs were extracted and plotted for PT, OT, SLT and Psych in Figures 4–7, representing the OR of greater disability (i.e. higher mRS score) at discharge from inpatient stroke care was associated with a greater amount of PT per day compared with zero. A value < 1 indicated decreased odds of greater disability (i.e. less disability, a positive outcome), whereas a value > 1 indicated increased odds of greater disability (a negative outcome). Note, all relevant model fit checks were performed and an investigation of the outliers (99th percentile and above of PT/day of stay) were removed; however, the results remained consistent.

FIGURE 4.

Cubic splines plot of the ORs (95% CI) for disability on discharge per minute of PT per day of inpatient stay (referenced to zero), as produced from the fully adjusted, multilevel, mixed-effects ordinal regression. Dashed lines indicate 95% CI.

FIGURE 5.

Cubic splines plot of the ORs (95% CI) for disability on discharge (mRS) per minute of OT per inpatient day of stay (referenced to zero), as produced from the fully adjusted, multilevel, mixed-effects ordinal regression model. Dashed lines indicate 95% CI.

FIGURE 6.

Cubic splines plot of the ORs (95% CI) for disability on discharge (mRS) per minute of SLT per inpatient day of stay (referenced to zero), as produced from the fully adjusted, multilevel, mixed-effects regression model. Dashed lines indicate 95% CI.

FIGURE 7.

Cubic splines plot of the ORs (95% CI) of disability on discharge per minute of Psych per inpatient day of stay (referenced to zero), as produced from the fully adjusted, multilevel, mixed-effects ordinal logistic regression model. Dashed lines indicate 95% CI.

Modelling PT as a single linear term (see Inpatient therapy) had indicated that increasing the average amount of therapy per day of stay was associated with greater disability on hospital discharge (see Table 19). Flexible modelling indicated that disability on discharge tended to improve with up to 5–10 minutes of PT per day of stay compared with zero, after which there was a gradual decline in improvement with greater amounts of PT per day of stay. At approximately 35 minutes of PT per day of stay, the effect crossed the null, indicating increased odds of greater disability with more PT per day of stay. Note, up until 35 minutes of PT per day of stay, the patient would continue to improve (albeit more slowly) or remain stable. In addition, CIs were wide, particularly in comparison with the other therapies, indicating that the response to greater amounts of PT per day of stay were highly variable.

Increases in the average amount of OT, SLT and Psych per day of inpatient stay were all previously associated with a small decrease in disability at discharge (see Table 19). Similarly to PT, the natural cubic splines indicated that increasing the average amount of these therapies per day of inpatient stay, up to an average of 5–10 minutes per day of stay, was associated with less disability at discharge. However, unlike PT, patients continued to see improvements in disability with greater amounts of therapy per day of stay, albeit at a more gradual rate. Note, the number of people receiving Psych was much smaller than the other therapies, so the natural cubic spline is much more susceptible to random noise, causing, we believe, the sharp increase in the first few minutes (see Figure 7).

Figures 8–19 describe the results of the flexible natural cubic spline models investigating the relationship between the average amount of therapy per day of stay and the other outcomes: mortality, destination on discharge and LOS, respectively. In the primary analysis (see Inpatient therapy), including therapy per day of stay as a single linear term indicated that the odds of dying decreased for all four therapies (PT = 0.97, OT = 0.88, SLT = 0.94 and Psych = 0.84) as amounts of therapy per day of inpatient stay increased. Figure 8 shows that the odds of survival appeared to increase up to an average of 10–15 minutes of therapy per day of stay. This subsequently became more gradual as the amount of therapy per day of stay increased past 15 minutes. In all except PT, the odds stabilise or continue to improve with more therapy per day of stay. For PT, there is a very gradual return towards the null and although this continues to indicate an improvement on the odds of dying (i.e. better survival), it is slower than the other therapies.

FIGURE 8.

Cubic splines plot of the ORs (95% CI) for mortality per minute of PT per inpatient day of stay (referenced to zero), as produced from the fully adjusted multilevel mixed-effects logistic regression model. Dashed lines indicate 95% CI.

FIGURE 9.

Cubic splines plot of the ORs (95% CI) for mortality per minute of OT per inpatient day of stay (referenced to zero), as produced from the fully adjusted multilevel mixed-effects logistic regression model. Dashed lines indicate 95% CI.

FIGURE 10.

Cubic splines plot of the ORs (95% CI) for mortality per minute of SLT per inpatient day of stay (referenced to zero), as produced from the fully adjusted multilevel mixed-effects logistic regression model. Dashed lines indicate 95% CI.

FIGURE 11.

Cubic splines plot of the ORs (95% CI) for mortality per minute of Psych per inpatient day of stay (referenced to zero), as produced from the fully adjusted multilevel mixed-effects ordinal logistic regression model. Dashed lines indicate 95% CI.

FIGURE 12.

Cubic splines plot of the ORs (95% CI) for discharge home per minute of PT per inpatient day of stay (referenced to zero), as produced from the fully adjusted multilevel mixed-effects ordinal logistic regression model. Dashed lines indicate 95% CI.

FIGURE 13.

Cubic splines plot of the ORs (95% CI) for discharge home per minute of OT per inpatient day of stay (referenced to zero), as produced from the fully adjusted multilevel mixed-effects ordinal logisitic regression model. Dashed lines indicate 95% CI.

FIGURE 14.

Cubic splines plot of the ORs (95% CI) for discharge home per minute of SLT per inpatient day of stay (referenced to zero), as produced from the fully adjusted multilevel mixed-effects ordinal logistic regression model. Dashed lines indicate 95% CI.

FIGURE 15.

Cubic splines plot of the ORs (95% CI) for discharge home per minute of Psych per inpatient day of stay (referenced to zero), as produced from the fully adjusted multilevel mixed-effects ordinal logistic regression model. Dashed lines indicate 95% CI.

FIGURE 16.

Cubic splines plot of the IRRs (95% CI) for LOS per minute of PT per inpatient day of stay (referenced to zero), as produced from the fully adjusted multilevel mixed-effect negative binomial regression model. Dashed lines indicate 95% CI.

FIGURE 17.

Cubic splines plot of the IRRs (95% CI) for LOS per minute of OT per inpatient day of stay (referenced to zero), as produced from the fully adjusted multilevel mixed-effects negative binomial regression model. Dashed lines indicate 95% CI.

FIGURE 18.

Cubic splines plot of the IRRs (95% CI) for LOS per minute of SLT per inpatient day of stay (referenced to zero), as produced from the fully adjusted multilevel mixed-effects negative binomial regression model. Dashed lines indicate 95% CI.

FIGURE 19.

Cubic splines plot of the IRRs (95% CI) for LOS per minute of Psych per inpatient day of stay (referenced to zero), as produced from the fully adjusted multilevel mixed-effects negative binomial regression model. Dashed lines indicate 95% CI.

In Inpatient therapy we reported a slight decrease in the odds of being discharged home with increasing PT per day of stay, but slight increases with greater amounts of OT, SLT and Psych per day of stay. Similar to the results for disability on discharge, the odds of being discharged home (Figures 12–15) increased in all four therapies for the first 5–10 minutes per day of stay before stabilising. The odds of discharge home continued to increase gradually, with increasing amounts of OT per day of stay; however, PT, SLT and Psych showed a decrease in the rate of improvement. This was gradual for SLT and Psych and does not return to the null effect (OR 1.0). PT was much more pronounced with the null line crossed between 15 and 20 minutes of PT per day of stay.

Figures 16–19 report the incidence rate ratios (IRRs) associated with LOS. An IRR < 1 indicated that, compared with zero minutes, the rate of expected LOS (days) would decrease. In all four therapies, the IRR indicated a decrease in LOS with increasing amounts of therapy per day of stay. The reduction in LOS was greatest with up to ≈ 10 minutes of therapy per day of stay, after which although the LOS continued to decrease (it was at a more gradual rate).

Discussion

In this chapter we considered the association between the amount of inpatient therapy per day of stay and outcome, in terms of disability and destination on discharge, mortality and LOS. Increasing the amount of all types of inpatient therapy per day of stay were associated with less mortality and LOS. Increasing amounts of inpatient OT, SLT and Psych per day of stay was also associated with less disability and institutionalisation. Large amounts of inpatient PT were associated with greater disability and institutionalisation at discharge. The results of the natural cubic splines indicated that for all professions, improved outcomes were associated with more therapy up to an average of 5–10 minutes of each therapy per day of stay, after which diminishing returns were observed, such that when PT exceeded ≈ 35 minutes per day of stay, poorer odds were seen for disability and institutionalisation destination on discharge. We also found that factors related to the patients’ level of disability were associated with outcome, whereas organisational factors were less so; however, the day and time of admission and nurses staffing levels were again implicated.

With some important caveats, especially with respect to PT, these results support the widely held view that ‘the more therapy, the better’, but also suggest that in some instances this simple mantra might be harmful. 93 As the patient-related factors associated with improved outcome were largely unmodifiable and organisational factors had little influence, the results highlight the potential importance of increasing the amount of therapy per day of stay to improve outcomes. Further research is needed to develop and evaluate the implementation of interventions and models of therapy delivery to increase the amount of therapy. Several interventions, such as circuit classes or group exercise, independent and semisupervised practice, and use of technology, have been found to be feasible and acceptable and are potential candidate interventions for further testing. 16,37,96–101

The results regarding the association between the amount of PT per day of stay and outcome in terms of disability on discharge and the proportion of patients discharged home were unexpected. Our additional analysis using cubic splines indicated that increasing the average amount of PT per day of stay was associated with less disability and increased odds of being discharged home up to 5–10 minutes of PT per day of stay. The association then levelled off until ≈ 35 minutes of therapy per day of stay, at which point increasing the average amount of PT per day of stay was associated with a poorer outcome. It should be noted that this ’35-minute’ figure does not refer to treatment sessions lasting 35 minutes. It refers to an average of 35 minutes of PT on every day of their stay in hospital. This may have comprised several treatment sessions per day, but this is unlikely as other studies have found this rarely happens in practice43,102 and the results reported in Chapter 3 showed that few patients receive treatment every day, or even every weekday. Thus, to receive an average of 35 minutes of PT per day of stay, patients would need to receive much longer treatment sessions on the days it was received. The SSNAP does not record details of the therapy received, other than the number of minutes of treatment per day, nor were we able to extract information about which days patients received treatment (due to concerns about data protection), consequently we are unable to analyse this finding further. In particular, we were unable to investigate the proportion of patients who received an average of 35 minutes of PT per day of stay to characterise them and the teams which delivered this relatively larger amount of therapy, nor were we able to investigate the content of the therapy provided.

Clinical interpretation of these results needs to be treated with great caution. They do not indicate that therapists only need provide 5–10 minutes of therapy per day of stay for maximum benefit, nor that providing > 35 minutes of PT per day of stay is harmful. They do suggest that the simple mantra – the more therapy, the better – is an oversimplification and large doses of therapy may not be beneficial for all patients. A similar conclusion was drawn from the A Very Early Rehabilitation Trial for Stroke (AVERT) trial, which found that very early mobilisation (within the first 24 hours) by therapists and nurses could be harmful, particularly for patients with severe strokes. It is notable that much of the research showing that ‘more therapy is better’ was undertaken with patients in the subacute and chronic stages post stroke, whereas (like AVERT) the current project included patients in the acute stages post stroke. The results from Chapter 3 indicate that PT is provided to a greater proportion of patients, earlier, and for a greater proportion of patients’ LOS, than the other therapies. This may indicate that physiotherapists are more involved with the subgroup of patients for whom large doses of therapy is unbeneficial than the other therapy professions, or who have a floor or ceiling effect on the outcomes measured in the SSNAP. Further research is clearly needed to better understand the optimal dose of therapy for different types of stroke patients and the most effective content of the therapies. Given the range of patients’ problems and potential for recovery, this will require the development of clinically feasible ways to stratify patients and individualise treatment algorithms.

We cannot fully explain the apparently detrimental effect of increasing amounts of PT. We adjusted analyses for all recorded confounders and the general conclusion remained robust to different methods of analysis, including linear and cubic splines. The same analytic methods applied to other forms of therapy resulted in estimates of effect in the anticipated direction: improved outcome with more therapy. Nevertheless, it is not possible to rule out the possibility that the apparent detriment is an artefact, possibly connected to different case mix for PT compared with the other professions. This is supported by the indication that increasing the amount of Psych per day of stay is associated with implausibly large improvements in outcome. Psych could be considered to have a case mix, which is the ‘opposite end of the scale’ to PT: very few stroke survivors receive Psych and those who do only receive small amounts (often only amounting to a one-off assessment) relatively late in their rehabilitation. Understanding of these findings can only be achieved through prospective research with more detailed data collection.

Data

Data from all patients admitted with stroke and reported to the SSNAP from June 2013 to July 2015 plus the 2014 acute organisational audit were included in the analysis. 113 This data set differed from that used in other chapters. The rationale for this decision was that the proportion of patients discharged within the first 3 days of admission to hospital was higher in some units than in others and, therefore, the resources used by stroke teams to provide care to these patients may affect their relative performance. Inpatient stroke teams were categorised as RATs, NRATs and NAITs. 114 In addition, the eight HASUs operating in London at the time were also identified. Although HASUs were also in operation elsewhere, the London HASUs were the only teams to submit data separately from other stroke teams within their organisation. As care procedures and the way stroke teams are organised differ, separate analyses were conducted for each type of stroke team.

The main outcome variable was LOS in the stroke team, which was defined as the time (in days) between admission to the stroke team and discharge or transfer to another stroke team. LOS in the SSNAP is recorded in minutes; however, for ease of interpretation, LOS was expressed in full days here by transforming the SSNAP record into days and rounding it to the closest integer.

Length of stay distributions are highly skewed, which may influence estimation of the relative influence of each hospital on its patients’ LOS. For this reason, right-tailed LOS outliers were excluded. Outliers were identified for each type of stroke team with a threshold based on three times the SD of the LOS distribution for that type of stroke team. 104 The cut-off point to define outliers for each case was calculated by first computing the number of days exceeding the national average LOS by 3 SDs and then rounding this number to the next complete month (30 days/month). For example, for NRATs the 3 SD threshold was 113 days, hence the cut-off point is rounded to 120 days to match 4 complete months. An observation was classified as outlier if it’s LOS exceeded 30 days for HASUs, 90 days for RATs, 120 days for NRATs and 150 days for NAITs. Consequently, 2410 admissions were identified as LOS outliers and dropped from the final sample.

Completion of the SSNAP was not mandatory for community-based therapy teams, hence the ≈ 20,000 observations describing community-based therapy were probably subject to selection bias and were thus not included in the analysis. The analysis presented in this chapter focused only on resource use and stroke care performance during inpatient care. Observations from stroke teams with < 24 admissions (i.e. one per month) during the study period were also dropped from the analysis (282 admissions). Additional criteria to exclude observations were admissions with no stroke impairments (NIHSS = 0), errors in the collection of LOS data (23 cases when LOS in a stroke unit was longer than the overall inpatient LOS and 774 observations when LOS in the stroke team could not be recovered from the SSNAP), and missing values in any of the admission-specific covariates. The final sample used in this analysis included 145,396 admissions from 256 stroke care teams.

Methods

A two-step approach suggested by Laudicella et al. ,115 Gaughan et al. 106 and Street et al. 105 was used to analyse variations in inpatient resource use and stroke care performance within the different types of stroke teams (HASU, RAT, NRAT, NAIT). 104,106,115 In particular, this chapter identified the main drivers of resource use for inpatient stroke care, estimated the influence of each stroke team on patients’ LOS over and above patient and treatment characteristics, and explored the extent to which organisational factors explain the variation in relative performance across stroke care teams.

The first stage specified a multilevel model. This considered that stroke admissions (level 1) were clustered within stroke teams (level 2) and estimated the team influence on LOS purged of patient- and admission-specific characteristics. Each team effect was then interpreted as a measure of relative performance, with higher values implying that a team used more resources than other teams to treat stroke patients. 116,117

Length of stay is considered count data as it tends to take a limited set of low values, with many patients having short LOS and relatively few staying for longer periods. 105 Therefore, Poisson and negative binomial regression models were used as LOS was overdispersed (variance higher than the mean, as reported Appendix 5). 118 The associated probability of observing the count y_ik in the most common version of the negative binomial model, is:

where Γ(.) is the gamma function. The first two conditional moments are:

where ∝ is a constant overdispersion parameter to be estimated with the rest of the parameters; y_ik is the LOS (number of days) of patient i in team k; X_ik is a vector of patient and treatment characteristics, used as proxy for case-mix; and h_k represents the deviation of team k from the grand mean. u_k is therefore the estimated team effect on LOS (performance measure), which is analogous to a fixed effect in linear models. 119

The X_ik vector includes demographic characteristics, such as the patient’s age group, a dummy variable indicating whether or not the patient was male and five dummy variables identifying the patient’s ethnicity. X_ik also captures the patient’s social deprivation status and whether or not he/she suffered a previous stroke or comorbidities (congestive heart failure, hypertension, atrial fibrillation and/or diabetes) before their stroke.

Stroke severity was taken into account by including the following factors in X_ik: three dummy variables indicating whether the stroke was moderate, severe or very severe, taking mild as reference (according to the NIHSS score on admission); a binary variable that identifies whether or not the stroke was a primary intracerebral haemorrhage; four dummy variables, each indicating if the patient required OT, PT, SLT or Psych in each stay; and disability (mRS) before arrival (whether by or from another stroke team) to the stroke team in question, expressed as a categorical variable taking the value of 2 (slight disability) as reference. The latter is defined as the mRS at discharge of the previous team if the patient was transferred in from another stroke team. Nearly all patients were first admitted to a HASU or RAT (99.3% and 97.9%, respectively). Therefore, the mRS score prior to arrival to the stroke team in question was not included in the model for HASUs and RATs.

In addition to patient characteristics, X_ik also includes treatment factors, such as the interaction of the need for each therapy with the average amount of therapy per day of stay per patient. The order of the stroke team in the patient’s pathway, the day of first admission and whether or not he/she was referred from a HASU were also considered. To capture the presence of adverse events (often used as a proxy for care quality), a dummy variable indicates whether or not the patient developed an urinary tract infection in the first 7 days following admission. 106 Finally, a variable indicating whether or not the patient died during their inpatient stay was also included.

The second stage analysed the variation in the estimated stroke team effects. Team- and hospital-level characteristics were used to estimate generalised least squares model with weights proportional to the inverse of the squared standard errors and Efron robust standard errors to correct for potential heteroscedasticity:104,120

The number of patients treated by the stroke team (in hundreds) was one of the explanatory variables included in z_k, to investigate if economies of scale were associated with stroke care performance (i.e. whether or not volume increases are associated with decreasing average costs, using LOS as a proxy for costs). The mortality rate per 100 patients observed by each team was also included in z_k, to explore whether or not stroke teams with high mortality had shorter average LOS. In addition, the rate per 100 patients who left the stroke team in a dependent status, using their mRS at discharge, was included to examine the extent to which systematically discharging patients in this condition was related to the team’s performance measure. The rate of urinary tract infections per 100 patients was included as a covariate to study the association between quality of care and the measure of stroke care performance. The numbers of WTE, qualified OT, PT, SLT, Psych, dietetics and nursing staff per 100 admissions were also included as covariates in the second stage. Dummy variables indicating whether or not thrombolysis was available on site and whether or not the team had access to community-based stroke therapy were also included in z_k. Finally, the geographical region where the team is based was also considered.

The second stage analysis was only conducted for RATs and NRATs, as there were only eight HASUs and team-level data for NAITs were not available because they reported to the 2015 post-acute organisational audit using different items.

Results

The wider selection criteria for this analysis generated a cohort of 145,396; however, their characteristics were similar to that of main SSNAPIEST cohort (detailed in Appendix 5).

Results from the negative binomial models estimated in the first stage are reported in Table 22.

In general, after conditioning for demographic characteristics, the estimated associations between LOS and stroke severity had the same (positive) direction, but different magnitude across types of stroke team (they were higher in RATs than other teams). For example, patients with a moderate stroke (compared with those with a mild stroke) were, on average, hospitalised for 13% more days in HASUs, 18% more days in NAITs, 21% more days in NRATs and 53% more days in RATs. The magnitude of these associations increased noticeably in all types of stroke team for patients with severe and very severe strokes. In particular, patients admitted to a NRAT with a very severe stroke stayed, on average, 84% longer than those admitted with a mild stroke. An intracerebral haemorrhage was also significantly associated with longer inpatient stays in HASUs, RATs and NRATs than an infarct.

The need for all types of therapy was strongly associated with longer LOS in all types of stroke teams. For example, needing OT was associated with 131% longer stays in RATs and with 70%, 78% and 108% longer LOS in NAITs, HASUs and NRATs, respectively. The need for PT was associated with a 76% longer LOS in HASUs, 75% in RATs, 72% in NRATs and 69% in NAITs. The range of association of the need of SLT with longer LOS ranged between 34% in NAITs to 112% in RATs. Finally, patients needing Psych at some point of their stay in RATs had a 132% longer LOS than patients in RATs who did not need this kind of therapy (this figure was 92% for patients in NRATs, 61% for patients in HASUs and 40% for patients in NAITs).

Conditional on needing the therapy in question, the average daily dose of therapy received across all types of stroke team was associated with a shorter LOS, with a similar magnitude across all types of stroke team (Table 23). For example, an additional minute of OT daily was associated with a 1.5% shorter LOS in HASUs, a 2.3% shorter LOS in RATs and a 1.4% shorter LOS in NRATs and NAITs. The strongest negative association was found between LOS and the average amount of Psych per day in RATs: an additional minute of Psych per day of stay was associated with a 4.9% shorter stay. When expressed in days, if all else was unchanged, an additional minute of OT or SLT daily in a RAT was associated with a 0.30-day (432-minute) shorter LOS and an additional minute of either PT or Psych daily were associated with a 90-minute shorter LOS (see Table 21).

Among other significant results, it is important to highlight the following: the association between transferring from a HASU to another inpatient stroke team was only significant when transferred to a NRAT or a NAIT. This was associated with a 18.4% and 24.2% shorter stay, respectively. Patients transferred to another inpatient team (i.e. the inpatient transfer and ‘other’ pathways) were associated with a longer LOS in HASUs and RATs, but a shorter stay in NRATs and NAITs. Dying in hospital was associated with a 42% longer LOS in HASUs, but with a significantly shorter LOS with other stroke teams.

To ease interpretation of stroke team effects (purged from patient- and admission-specific characteristics) as measures of relative performance, Figure 20 plots team effects standardised by the national average LOS for each type of team. 104 In this sense, a standardised team effect of 1.5 means that patients in the team in question have 50% longer LOS than the average for all teams of that type (not being due to the factors included in X_ik). Team effects are ranked by their deviation from the national average from left to right, with those on the left-hand side having shorter LOS. It can be seen that even after conditioning on measurable patient and treatment characteristics, there are large variations in the influence of stroke teams on their patients’ LOS in all types of stroke team. This difference is most noticeable among RATs (partly explained by the higher number of teams in this category), for which the plot of standardised team effects follows an S-shaped distribution, with clearly identified groups of teams having both significantly lower and significantly higher effect on the LOS of its patients with respect to the national average. The influence of the stroke team on resource use ranged from –19% to 20% of the national average for HASUs, from –52% to 107% of the national average for RATs, from –39% to 128% of the national average for NRAT and from –53% to 121% of the national average for NAITs.

FIGURE 20.

Unexplained variation in stroke teams’ average LOS for (a) HASUs (June 2013–July 2015); (b) RATs (June 2013–July 2015); (c) NRATs (June 2013–July 2015); and (d) NAITs (June 2013–July 2015).

The results of the second stage of analysis showed that lower team effects meant shorter than average LOS and, consequently, a negative coefficient in the second stage was interpreted as a positive association with stroke care performance. In general, the results of the second stage show that the variation in performance remains highly unexplained, as only few significant associations are found and the adjusted R² in both cases is < 0.25 (see Table 23).

Discussion

By estimating four separate multilevel count data models using inpatient LOS as the dependent variable and conditioning for patient and treatment characteristics, this analysis identified the main drivers of resource use in stroke care, estimated stroke care team effects on the LOS of its patients, interpreted these team effects as a performance measure and then estimated two linear models to analyse the variation in stroke care performance across two categories of stroke care teams.

We showed that the need for therapy was the main driver of resource use, even after conditioning for stroke severity; patients who needed therapy stayed in hospital longer and thus used more resources than those who did not. All the associations with need for therapy were significantly different from zero at the 1% significance level.

After conditioning for need for therapy, we also found that the average amount of therapy per day of stay significantly influenced resource use: more therapy led to shorter LOS and thus less resource use in all types of stroke team and all therapies, except PT in NAITs. As these findings are not showing a causal relationship they should be interpreted with caution, but they should motivate future research into the economic consequences of rapid assessment of therapy needs and the amount of therapy provided to those stroke patients in need.

We also noted a huge variation in resource use for all types of stroke team, which was unexplained by the patient-, stroke- or team-related characteristics available in the SSNAP. Therefore, these findings deserve further operational and financial analyses to unmask the cause(s) of the variation.

As indicated by the adjusted deviance R² (a measure of goodness of fit for count data models),121 the patient-level (stage 1) analysis explained between 33% (NAITs) and 48% (RATs) of the variation in stroke LOS. A similar study analysing stroke LOS in England during 2007/8 explained up to 33% of the variation in LOS. 106 One explanation for the better fit of the models in this chapter could be the use of the SSNAP data that report clinical information at the stroke admission level (e.g. stroke severity, pre-morbid disability, type and amount of therapy/day received with each stroke team), which is not commonly recorded in the administrative databases most used in the UK.

The analysis presented here is subject to two main limitations. First, we did not analyse cost variations across stroke teams, as cost information is not available in the SSNAP. A previous study using European data found that stroke costs and inpatient LOS are highly correlated;107 however, a study using English data found that this did not necessarily apply for this country. 106 The latter used data from 2007/8 and many stroke services in the UK have been reorganised in recent years, particularly with the advent of centralised hyperacute stroke care. Therefore, the types of stroke teams defined here cannot be applied to the earlier work (and vice versa). Linking the SSNAP with other databases reporting cost data at the individual level would allow the present analysis to be complemented with a similar one using patient and treatment costs as the outcome variables to explain resource use. This would represent an opportunity to explore not only if the relationship between costs and LOS has changed over time (compared with the Gaughan et al. 106 study), but also if this relationship varies across types of stroke team. However, this was not possible due to data protection issues.

The second limitation was that further hospital admissions with a different primary diagnosis, as well as specialist outpatient clinic, general practitioner visits following discharge and other community care were not taken into account in this study, given that the SSNAP only includes details of inpatient admissions for stroke. Consequently, the present analysis was restricted to analysing resource use by inpatient stroke teams and, as such, is only a partial analysis of resource use. Future studies should exploit health-care utilisation information included in other administrative data sets not readily linked to the SSNAP (e.g. Hospital Episodes Statistics, Clinical Practice Research Datalink) to analyse the extent to which reduction in the use of resources in the stroke team is associated with overall health-care resource use by stroke patients.

Limitations

The findings reported in this study need to be considered in the light of several important limitations. First, they were drawn from an observational data set of clinical records. Therefore, the associations reported do not indicate causation. It is unlikely that the clinical database is completely accurate and so misclassification may have occurred; however, we feel that this is unlikely to be systematic. We included all the relevant variables available to us from the SSNAP database in the regression modelling, but not everything can be measured so it is probable that there is some residual confounding in the results due to factors which are not, or cannot, be measured. Please note that we have assessed the association between therapy and resource use, and reported the costs of stroke therapy. We have not undertaken an evaluation of cost-effectiveness.

Implications for practice

The large body of information describing stroke therapy, the pathways and the SICs can be used to describe, define, benchmark and develop services. The SICs may prove useful to develop personalised treatment protocols in the future. The LOS identified for each pathway can be used as a benchmark to estimate discharge date after admission. On average, LOS is 9–10 days for an acute or combined stroke team, whether or not the patient is discharged to community rehabilitation. If the patient is transferred to another inpatient stroke team, the LOS was approximately 1 month if initially admitted to an acute team and 2 months if admitted to a combined team.

Given higher (therapy and nurse) staffing levels and an extended (weekend) therapy service were associated with more therapy, and more therapy was associated with improved LOS, resource use and mortality, clinical services should consider the feasibility of increasing staffing levels and extending their availability. Clinical interpretation of the complex associations between the amount of therapy and the other outcomes needs to be treated with great caution. They do not indicate that therapists only need to provide 5–10 minutes of therapy per day of stay for maximum benefit, nor that providing > 35 minutes of PT per day of stay is harmful. They do suggest that the simple mantra – the more therapy, the better – is an oversimplification and that large doses of therapy may not be beneficial for all patients.

Recommendations for further research

The findings of this project indicate that further research is needed to address the following questions:

• Is the surprising finding regarding the association between dose of inpatient PT and outcomes confirmed by prospective research, or is it an artefact?

• What is the optimal dose (in terms of amounts, intensity, frequency and duration) of each therapy for different types of stroke patients?

• What are the most effective interventions to deliver an optimal dose of therapy to different patient groups?

• What is the most effective way to configure, organise and resource stroke therapy and rehabilitation services (including the type of stroke team, staffing levels, the ratio of qualified and support staff, and working hours) to optimise the amount of therapy provided? A key question is whether an effective extended service requires an increase in therapy staffing levels to cover the weekend input or whether spreading the existing therapy workforce to cover the whole week, but more thinly, is adequate. In addition, whether or not patients who require ongoing rehabilitation may be best served by a standalone, specialist rehabilitation team.

• What is the most effective way to configure stroke rehabilitation services in terms of type of stroke team and stroke care pathway for patients with different impairments and levels of disability?

• What are the economic and clinical consequences of increasing the amount of therapy?

• What are the reasons for the observed variation in resource use during stroke care?

• Can SICs be used to predict recovery and outcome, and to develop more personalised treatment algorithms for individual patients?

• How great is the need for psychological therapy and what is the most effective way to provide these services?

• How can the apparent inequity of access to therapy for disadvantaged group be overcome?

• How are community-based therapy service organised? What is the content of the therapy they deliver? How can it be optimised?

• Why is there often an interval between completing therapy and discharge from hospital? Should services aim to remove or overcome it? If so, what is the most effective way to do so?

• What are the long-term needs of people with mild stroke who are discharged from hospital very quickly? How can they be provided with easy access to stroke rehabilitation services if needed?

• What are the reasons behind the extremely long LOS for a small number of stroke survivors? What can be done to reduce these?

Contributions of authors

Dr Matthew Gittins (https://orcid.org/0000-0002-9888-1197) (Lecturer in Biostatistics, University of Manchester) managed the day-to-day delivery of the project, developed the statistical plan, prepared the data, undertook the statistical analyses, contributed to the interpretation of the results, wrote the report and disseminated of the results.

Dr David Lugo-Palacios (https://orcid.org/0000-0003-2621-5627) (Research Associate, University of Manchester, at the time of the project) undertook the health economics section of the project, conceived the analysis, prepared the data, undertook the analysis, interpreted the results, wrote the relevant chapters and disseminated the results.

Professor Andy Vail (https://orcid.org/0000-0001-8274-2726) (Professor of Biostatistics, University of Manchester) line-managed the statistical analyses and contributed to the all aspects of the project, development of the grant application, conceptualisation of the project, development of the statistical plan, analyses, interpretation of results, production of the report and dissemination of the results.

Professor Audrey Bowen (https://orcid.org/0000-0003-4075-1215) (Professor of Neuropsychological Rehabilitation, University of Manchester) contributed to the development of the grant application, conceptualisation of the project, interpretation of results, production of the report and dissemination of the results.

Ms Lizz Paley (https://orcid.org/0000-0002-3879-5377) (Intelligence Manager, SSNAP, at the time of the project) contributed to the preparation of the data, the analyses, interpretation of the results, production of the report and dissemination of the results.

Dr Benjamin Bray (https://orcid.org/0000-0002-8983-7530) (Research Director, SSNAP, at the time of the project) contributed to the development of the grant application, conceptualisation of the project and the analyses, interpretation of results, production of the report and dissemination of the results.

Professor Brenda Gannon (https://orcid.org/0000-0002-3672-4030) (Professor of Health Economics, University of Queensland) contributed to the health economics of the project, development of the grant application, conceptualisation of the project, development of the analysis plan, interpretation of the results, production of the report and dissemination of the results.

Professor Sarah F Tyson (https://orcid.org/0000-0001-6301-8791) (Professor of Rehabilitation, University of Manchester) was the principal investigator and budget holder, and was involved in all aspects of the project.

Publications

Hammerbeck U, Gittins M, Vail A, Paley L, Tyson SF, Bowen A. Spatial neglect in stroke: identification, disease process and association with outcome during inpatient rehabilitation. Brain Sci 2019;9:374.

Lugo-Palacios DG, Gannon B, Gittins M, Vail A, Bowen A, Tyson SF. Variations in hospital resource use across stroke care teams in England, Wales and Northern Ireland. BMJ Open 2019;9:e030426.

Gittins M, Lugo-Palacios DG, Paley L, Bray B, Bowen A, Vail A, Gannon B, Tyson S. How do patients pass through stroke services? Identifying stroke care pathways using national audit data [published online ahead of print 6 March 2020]. Clin Rehabil 2020.

Data-sharing statement

This project involved secondary analysis of anonymised, routinely collected clinical data. The data were made available under a data-sharing agreement between the University of Manchester, SSNAP and the Health Care Quality Improvements Partnership. A condition of that agreement was that all the data would be destroyed at the end of the project. Therefore, we are unable to make them available to others. Further information can be obtained from the corresponding author.

Patient data

This work uses data provided by patients and collected by the NHS as part of their care and support. Using patient data is vital to improve health and care for everyone. There is huge potential to make better use of information from people’s patient records, to understand more about disease, develop new treatments, monitor safety, and plan NHS services. Patient data should be kept safe and secure, to protect everyone’s privacy, and it’s important that there are safeguards to make sure that it is stored and used responsibly. Everyone should be able to find out about how patient data are used. #datasaveslives You can find out more about the background to this citation here: https://understandingpatientdata.org.uk/data-citation.

Disclaimers

This report presents independent research funded by the National Institute for Health Research (NIHR). The views and opinions expressed by authors in this publication are those of the authors and do not necessarily reflect those of the NHS, the NIHR, NETSCC, the HS&DR programme or the Department of Health and Social Care. If there are verbatim quotations included in this publication the views and opinions expressed by the interviewees are those of the interviewees and do not necessarily reflect those of the authors, those of the NHS, the NIHR, NETSCC, the HS&DR programme or the Department of Health and Social Care.