Organised inpatient (stroke unit) care for stroke: network meta‐analysis

Peter Langhorne; Samantha Ramachandra; Stroke Unit Trialists' Collaboration

doi:10.1002/14651858.CD000197.pub4

. 2020 Apr 23;2020(4):CD000197. doi: 10.1002/14651858.CD000197.pub4

Organised inpatient (stroke unit) care for stroke: network meta‐analysis

Peter Langhorne ^1,^✉, Samantha Ramachandra ²; Stroke Unit Trialists' Collaboration³

Editor: Cochrane Stroke Group

PMCID: PMC7197653 PMID: 32324916

Abstract

Background

Organised inpatient (stroke unit) care is provided by multi‐disciplinary teams that manage stroke patients. This can been provided in a ward dedicated to stroke patients (stroke ward), with a peripatetic stroke team (mobile stroke team), or within a generic disability service (mixed rehabilitation ward). Team members aim to provide co‐ordinated multi‐disciplinary care using standard approaches to manage common post‐stroke problems.

Objectives

• To assess the effects of organised inpatient (stroke unit) care compared with an alternative service.

• To use a network meta‐analysis (NMA) approach to assess different types of organised inpatient (stroke unit) care for people admitted to hospital after a stroke (the standard comparator was care in a general ward).

Originally, we conducted this systematic review to clarify:

• The characteristic features of organised inpatient (stroke unit) care?

• Whether organised inpatient (stroke unit) care provide better patient outcomes than alternative forms of care?

• If benefits are apparent across a range of patient groups and across different approaches to delivering organised stroke unit care?

Within the current version, we wished to establish whether previous conclusions were altered by the inclusion of new outcome data from recent trials and further analysis via NMA.

Search methods

We searched the Cochrane Stroke Group Trials Register (2 April 2019); the Cochrane Central Register of Controlled Trials (CENTRAL; 2019, Issue 4), in the Cochrane Library (searched 2 April 2019); MEDLINE Ovid (1946 to 1 April 2019); Embase Ovid (1974 to 1 April 2019); and the Cumulative Index to Nursing and Allied Health Literature (CINAHL; 1982 to 2 April 2019). In an effort to identify further published, unpublished, and ongoing trials, we searched seven trial registries (2 April 2019). We also performed citation tracking of included studies, checked reference lists of relevant articles, and contacted trialists.

Selection criteria

Randomised controlled clinical trials comparing organised inpatient stroke unit care with an alternative service (typically contemporary conventional care), including comparing different types of organised inpatient (stroke unit) care for people with stroke who are admitted to hospital.

Data collection and analysis

Two review authors assessed eligibility and trial quality. We checked descriptive details and trial data with co‐ordinators of the original trials, assessed risk of bias, and applied GRADE. The primary outcome was poor outcome (death or dependency (Rankin score 3 to 5) or requiring institutional care) at the end of scheduled follow‐up. Secondary outcomes included death, institutional care, dependency, subjective health status, satisfaction, and length of stay. We used direct (pairwise) comparisons to compare organised inpatient (stroke unit) care with an alternative service. We used an NMA to confirm the relative effects of different approaches.

Main results

We included 29 trials (5902 participants) that compared organised inpatient (stroke unit) care with an alternative service: 20 trials (4127 participants) compared organised (stroke unit) care with a general ward, six trials (982 participants) compared different forms of organised (stroke unit) care, and three trials (793 participants) incorporated more than one comparison.

Compared with the alternative service, organised inpatient (stroke unit) care was associated with improved outcomes at the end of scheduled follow‐up (median one year): poor outcome (odds ratio (OR) 0.77, 95% confidence interval (CI) 0.69 to 0.87; moderate‐quality evidence), death (OR 0.76, 95% CI 0.66 to 0.88; moderate‐quality evidence), death or institutional care (OR 0.76, 95% CI 0.67 to 0.85; moderate‐quality evidence), and death or dependency (OR 0.75, 95% CI 0.66 to 0.85; moderate‐quality evidence). Evidence was of very low quality for subjective health status and was not available for patient satisfaction. Analysis of length of stay was complicated by variations in definition and measurement plus substantial statistical heterogeneity (I² = 85%). There was no indication that organised stroke unit care resulted in a longer hospital stay. Sensitivity analyses indicated that observed benefits remained when the analysis was restricted to securely randomised trials that used unequivocally blinded outcome assessment with a fixed period of follow‐up. Outcomes appeared to be independent of patient age, sex, initial stroke severity, stroke type, and duration of follow‐up. When calculated as the absolute risk difference for every 100 participants receiving stroke unit care, this equates to two extra survivors, six more living at home, and six more living independently.

The analysis of different types of organised (stroke unit) care used both direct pairwise comparisons and NMA.

Direct comparison of stroke ward versus general ward: 15 trials (3523 participants) compared care in a stroke ward with care in general wards. Stroke ward care showed a reduction in the odds of a poor outcome at the end of follow‐up (OR 0.78, 95% CI 0.68 to 0.91; moderate‐quality evidence).

Direct comparison of mobile stroke team versus general ward: two trials (438 participants) compared care from a mobile stroke team with care in general wards. Stroke team care may result in little difference in the odds of a poor outcome at the end of follow‐up (OR 0.80, 95% CI 0.52 to 1.22; low‐quality evidence).

Direct comparison of mixed rehabilitation ward versus general ward: six trials (630 participants) compared care in a mixed rehabilitation ward with care in general wards. Mixed rehabilitation ward care showed a reduction in the odds of a poor outcome at the end of follow‐up (OR 0.65, 95% CI 0.47 to 0.90; moderate‐quality evidence).

In a NMA using care in a general ward as the comparator, the odds of a poor outcome were as follows: stroke ward ‐ OR 0.74, 95% CI 0.62 to 0.89, moderate‐quality evidence; mobile stroke team ‐ OR 0.88, 95% CI 0.58 to 1.34, low‐quality evidence; mixed rehabilitation ward ‐ OR 0.70, 95% CI 0.52 to 0.95, low‐quality evidence.

Authors' conclusions

We found moderate‐quality evidence that stroke patients who receive organised inpatient (stroke unit) care are more likely to be alive, independent, and living at home one year after the stroke. The apparent benefits were independent of patient age, sex, initial stroke severity, or stroke type, and were most obvious in units based in a discrete stroke ward. We observed no systematic increase in the length of inpatient stay, but these findings had considerable uncertainty.

Plain language summary

Organised inpatient (stroke unit) care

Review question Does organised inpatient (stroke unit) care improve the recovery of people with stroke in hospital compared with conventional care in general wards?

Background Organised inpatient (stroke unit) care is a form of care provided in hospital by nurses, doctors, and therapists who specialise in looking after people with stroke. They aim to work as a co‐ordinated team to provide the most appropriate care tailored to the needs of individual people with stroke.

Study characteristics We identified 29 trials involving 5902 participants (search completed 2 April 2019). Participants who were recruited had had a recent stroke and required admission to hospital. Organised inpatient (stroke unit) care was provided in a variety of ways including stroke ward (care provided in a discrete stroke ward), mixed rehabilitation ward (setting seeking to improve care for people with stroke within a mixed rehabilitation ward), and mobile stroke team (peripatetic team looking after people with stroke across a range of wards).

Key results At an average of 12 months after their stroke, people who received organised inpatient (stroke unit) care were more likely to be alive (an extra two people surviving for every 100 receiving stroke unit care; moderate‐quality evidence) and living at home (an extra six patients for every 100 receiving stroke unit care; moderate‐quality evidence). They also were more likely to be independent in daily activities (an extra six patients for every 100 receiving stroke unit care; moderate‐quality evidence). The apparent benefits were seen in men and women, older and younger patients, and people with different types of stroke and different stroke severity. Benefits were most obvious when the stroke unit was based in a discrete stroke ward.

Quality of the evidence We downgraded the quality of evidence to 'moderate' for the main outcomes because it was impossible to hide the treating service from participants or healthcare workers. These conclusions were not dependent on trials judged to be of lower quality because of poor design or missing data. More information was missing for some of the other outcome measures and analyses, which we have downgraded to low‐quality evidence.

Conclusion People with stroke who receive organised inpatient (stroke unit) care are more likely to be alive, living at home, and independent in looking after themselves one year after their stroke. Apparent benefits were seen across a broad range of people with stroke. Various types of stroke units have been developed. The best results appear to come from stroke units based in a dedicated stroke ward.

Summary of findings

Background

Description of the condition

Stroke is now the third leading cause of disability (Murray 2012), and it is the second leading cause of mortality worldwide (Lozano 2012). The global disease burden of stroke increased by 19% between 1990 and 2010 (Murray 2012), and current projections estimate the number of deaths worldwide will rise to 6.5 million in 2015 and to 7.8 million in 2030 (Strong 2007). Interventions that are applicable to a majority of people with stroke and that aim to reduce associated mortality and disability are essential.

During their initial illness, people with stroke are frequently admitted to hospital, where they can receive care in a variety of ways and in a range of settings. Traditionally, care for people with stroke was provided within departments of general (internal) medicine, neurology, or medicine for the elderly, where they would be managed alongside a range of other patient groups. A more focused approach to the treatment of people with stroke in hospital has been developed.

Description of the intervention

Organised inpatient (stroke unit) care is the term used to describe focused care for people with stroke in hospital under a multi‐disciplinary team of individuals who specialise in stroke management (SUTC 1997a). This concept is not new, and its value has been debated for more than 30 years (Ebrahim 1990; Garraway 1985; Langhorne 1993; Langhorne 1998; Langhorne 2012). In essence, the debate has concerned whether the perceived effort and cost of focusing the care of people hospitalised with stroke within specially organised units would be matched by tangible benefits for the people receiving that care. In particular, would more people survive and make a good recovery as a result of organised inpatient (stroke unit) care?

Why it is important to do this review

A systematic review of all available trials previously described the range of characteristics of stroke unit care and addressed the question of whether improving the organisation of inpatient stroke care can bring about improvements in important patient outcomes (SUTC 1997a). This review continues to be extended and updated within the Cochrane Library (SUTC 2001; SUTC 2007; SUTC 2013). Since the last update, network meta‐analysis (NMA) has become established as an approach for handling multiple comparisons. We have added NMA to our updated review.

Objectives

To assess the effects of organised inpatient (stroke unit) care compared with an alternative service (usually contemporary conventional care)
To use a network meta‐analysis (NMA) approach to assess different types of organised inpatient (stroke unit) care for people admitted to hospital after a stroke (the standard comparator was care in a general ward)

Originally, this systematic review was conducted to address four broad questions.

What are the characteristic features of organised inpatient (stroke unit) care?
Can organised inpatient (stroke unit) care provide better patient outcomes than alternative forms of care?
Are any benefits apparent across a range of patient groups?
Are any benefits apparent across different approaches to delivering organised stroke unit care? In particular, we hypothesised that organised care would be more effective than care provided in general medical wards, but that different forms of organised care would achieve similar outcomes.

Within the current version of this review, we wished to establish whether previous conclusions were altered by the inclusion of new outcome data from recent trials and further analysis via NMA.

Methods

Criteria for considering studies for this review

Types of studies

We included all randomised controlled clinical trials that compared an organised system of inpatient (stroke unit) care with an alternative form of inpatient care. This was usually contemporary conventional care but could include an alternative model of organised inpatient (stroke unit) care (see Types of interventions). Previous versions of this review included trials with quasi‐random treatment allocation (such as bed availability or date of admission) (SUTC 1997a; SUTC 2001; SUTC 2007). However, in an effort to ensure that this ongoing systematic review focuses on data from trials with strict randomisation procedures, we excluded all quasi‐randomised trials in the previous update (SUTC 2013). We would have included cluster‐randomised trials, but we identified none. We excluded cross‐over trials because of cross‐over of effects.

Types of participants

Any person admitted to hospital who had suffered a stroke was eligible. We recorded the delay between stroke onset and hospital admission but did not use this as an exclusion criterion. We used a clinical definition of stroke: focal neurological deficit due to cerebrovascular disease, excluding subarachnoid haemorrhage and subdural haematoma.

Types of interventions

Organised inpatient (stroke unit) care can be considered a complex organisational intervention comprising multi‐disciplinary staff providing a complex package of care to people with stroke in hospital. In the original version of this review, the first question was whether organised inpatient (stroke unit) care could improve outcomes compared with contemporary conventional care (usually in general wards) (SUTC 1997a). We then had to modify the analyses in a minor way to reflect the emerging pattern of service organisation and to allow the comparison of 'more organised' versus 'less organised' services (for which the latter was usually contemporary conventional care). We did this because some recent trials have addressed new questions and included comparisons of two services, both of which met the basic definition of organised (stroke unit) care. In the original service descriptions used in this review (SUTC 1997a), service organisation was considered as a hierarchy comprising the following.

Stroke ward: where a multi‐disciplinary team including specialist nursing staff based in a discrete ward cares exclusively for people with stroke. This category included the following subdivisions.
- Acute stroke units that accept patients acutely but discharge early (usually within seven days); these appear to fall into three broad subcategories.
  - 'Intensive' model of care with continuous monitoring, high nurse staffing levels, and the potential for life support.
  - 'Semi‐intensive' model of care with continuous monitoring, high nurse staffing, but no life support facilities.
  - 'Non‐intensive' model of care with none of the above.
- Rehabilitation stroke units that accept patients after a delay, usually of seven days or longer, and focus on rehabilitation.
- Comprehensive (i.e. combined acute and rehabilitation) stroke units that accept patients acutely but also provide rehabilitation for at least several weeks if necessary. Both the rehabilitation unit model and the comprehensive unit model offer prolonged periods of rehabilitation.
Mixed rehabilitation ward: where a multi‐disciplinary team including specialist nursing staff in a ward provides a generic rehabilitation service but not exclusively caring for people with stroke.
Mobile stroke team: where a peripatetic multi‐disciplinary team (usually excluding specialist nursing staff) provides care in a variety of settings.
General medical ward: where care is provided in an acute medical or neurology ward without routine multi‐disciplinary input.

For the NMA eligibility assessment, we considered the transitivity (similarity) of trials included in the network (see Data synthesis), which requires all interventions to be legitimate alternatives. All of the four main categories have been used to provide care for unselected acute stroke patients and can be considered broadly comparable for the purpose of an NMA. The general medical ward group was the reference group,

Types of outcome measures

Primary outcomes

In the previous version of this review, the primary analyses examined death, dependency, and the requirement for institutional care at the end of scheduled follow‐up of the original trial (SUTC 2013). We categorised dependency into two groups, where we took 'independent' to mean that an individual did not require physical assistance for transfers, mobility, dressing, feeding, or toileting. We considered individuals who failed any of these criteria 'dependent'. The criteria for independence were approximately equivalent to a modified Rankin score of 0 to 2 or a Barthel Index greater than 18 out of 20 (Wade 1992). We took the requirement for long‐term institutional care to mean care in a residential home, a nursing home, or a hospital at the end of scheduled follow‐up.

In view of changes in reporting standards, for this update we have now provided a single composite primary outcome: poor outcome: death or dependency or requiring institutional care (if dependency data were not available). This allowed us to keep the primary focus of the review (i.e. the focus on independent survival as an outcome), while optimising the quantity of data available.

Secondary outcomes

Secondary outcome measures now include:

death;
death or institutional care;
death or dependency;
patient subjective health status (measured using tools such as the Nottingham Health Profile, EuroQol, Short Form‐36);
patient and carer satisfaction (recorded on a Likert scale or as responses to statements); and
duration of stay in hospital or institution or both.

Outcomes are reported at the end of scheduled follow‐up. Some trials subsequently provided supplementary extended follow‐up data, which are presented separately.

Search methods for identification of studies

See the methods for the Cochrane Stroke Group Specialised register. We searched for trials in all languages and arranged the translation of relevant papers published in languages other than English.

Electronic searches

We searched the trials registers of the Cochrane Stroke Group (2 April 2019). In addition, in collaboration with the Cochrane Stroke Group Information Specialist, we searched the Cochrane Central Register of Controlled Trials (CENTRAL; 2019; Issue 4), in the Cochrane Library (searched 2 April 2019) (Appendix 1); MEDLINE Ovid (1946 to 1 April 2019) (Appendix 2); Embase Ovid (1974 to 1 April 2019) (Appendix 3); and the Cumulative Index to Nursing and Allied Health Literature (CINAHL) EBSCO (1982 to 2 April 2019) (Appendix 4).

We searched the following registers of ongoing trials using the keyword 'stroke'.

US National Institutes of Health Ongoing Trials Register ClinicalTrials.gov (www.clinicaltrials.gov; searched 2 April 2019).
World Health Organization International Clinical Trials Registry Platform (apps.who.int/trialsearch; searched 2 April 2019).
CenterWatch Clinical Trials Listing Service (www.centerwatch.com; searched 13 August 2018).
Community Research & Development Information Service (of the European Union) (cordis.europa.eu/en/home.html; searched 13 August 2018).
South African National Clinical Trial Register (www.sanctr.gov.za; searched 13 August 2018).
The Internet Stroke Center ‐ Stroke Trials Registry (www.strokecenter.org/trials;searched 13 August 2018).
Clinical Trials Results register (www.clinicaltrialresults.org; searched 2 April 2019).

Searching other resources

In an effort to identify further published, unpublished, and ongoing trials, we:

performed citation tracking using Web of Science Cited Reference Search for all included studies;
searched the reference lists of included trials and all relevant articles;
obtained further information from individual trialists; and
contacted other researchers in the field and publicised our preliminary findings at stroke conferences in UK, Scandinavia, Germany, Netherlands, Switzerland, Spain, Canada, Brazil, Argentina, Australia, Belgium, USA, India, Sri Lanka, Singapore, Italy, and Hong Kong.

Data collection and analysis

Selection of studies

For this updated review, one review author (PL) read the titles and abstracts of records obtained through the electronic searches and excluded obviously irrelevant studies. We obtained full copies of the remaining studies, and two review authors (PL and SR) independently selected studies for inclusion based on the following eligibility criteria.

Randomised controlled trial.
Service intervention providing a form of organised inpatient (stroke unit) care.
Service aim to improve functional recovery and survival after stroke.
Trial of stroke patients.

We tried to establish the characteristics of unpublished trials through discussion with the Cochrane Stroke Group Information Specialist before analysing the results.

Data extraction and management

If possible, the principal review author (PL) obtained descriptive information about the service characteristics of organised inpatient (stroke unit) care and conventional care settings through a structured interview or correspondence conducted with the trial co‐ordinators (n = 19). We obtained additional information from published sources. We then allocated trials to service subgroups. We confirmed outcome data from published sources and supplemented them with unpublished information provided by the co‐ordinator of each individual trial. We asked trialists to provide information on the number of participants who were dead or dependent and the number requiring institutional care or missing at the end of scheduled follow‐up. For this updated review, two review authors (PL, SR) independently extracted information using a standard data extraction form.

We sought subgroup information on potential effect modifiers primarily for the combined outcome of death or requiring institutional care. We obtained unpublished aggregated data for a majority of trials, but insufficient quantities of individual patient data were available to allow a comprehensive individual patient data analysis.

We obtained subgroup data regarding the following participant groups (see SUTC 1997a for details).

Age: up to 75 years or 75 years or older.
Sex: men or women.
Stroke severity: dependency at the time of randomisation (usually within one week of the index stroke).
- Mild stroke: equivalent to a Barthel Index of 10 to 20 out of 20 during the first week.
- Moderate stroke: equivalent to a Barthel Index of 3 to 9 out of 20 during the first week.
- Severe stroke: equivalent to a Barthel Index of 0 to 2 out of 20 during the first week.
Stroke type: ischaemic or haemorrhagic based on neuroimaging.

Assessment of risk of bias in included studies

We assessed risk of bias using Cochrane's 'Risk of bias' tool, as described in Chapter 8 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). Two review authors identified the method of random sequence generation, the method of concealment of treatment allocation, blinding of participants and personnel, the presence of blinding of outcome assessment, completeness of follow‐up, and evidence of selective reporting. We used these potentially important factors in sensitivity analyses, but we did not use them as exclusion criteria. The principal review author then used this information to inform the categorisations within the 'Summary of findings' tables and the GRADE allocations.

Measures of treatment effect

When our analyses of poor outcome, death, dependency, or institutionalisation at the end of scheduled follow‐up were reported, we analysed these using the odds ratio (OR) and the 95% confidence interval (CI) for an adverse outcome.

We aimed to record length of stay in hospital or in an institution as the mean and standard deviation (SD). When only medians were available, we assumed these were approximate to the mean. When no other data were provided with the mean value, we inferred the SD as being at least as large as those in comparable trials using the same measure. Because length of stay was reported in a variety of ways, we checked the results obtained with the mean difference (MD) using the standardised mean difference (SMD) and the 95% CI.

We anticipated that measures of subjective health status would be analysed as mean differences, and measures of satisfaction would be analysed as odds ratios for particular responses.

Unit of analysis issues

We anticipated that most trials would have a simple parallel‐group design in which each individual was randomised to one of two treatment groups. When a trial had three (or more) treatment groups, we planned to analyse each treatment arm as a separate study. We have not included cross‐over trials because of the likelihood or carryover effects.

Dealing with missing data

When data were missing for the outcome of death, dependency, or institutionalisation, we assumed the participant to be alive, independent, and living at home. We aimed to explore the implications of these assumptions in sensitivity analyses.

Assessment of heterogeneity

We planned to determine heterogeneity by visually inspecting the forest plot and by using the I² statistic. We defined significant heterogeneity as I² greater than 50%. When significant heterogeneity occurred, we explored potential sources using pre‐planned sensitivity analyses. Assessments of transitivity and consistency are discussed in the NMA section under Data synthesis.

Assessment of reporting biases

We employed a comprehensive search strategy in an effort to avoid reporting biases. To identify unpublished studies, we searched trial registers and contacted trialists and other experts in the field. We planned to inspect funnel plots if enough studies were available.

Data synthesis

Pairwise comparisons

When we did not have access to individual patient data, two review authors (PL, SR) extracted data from published reports. When individual patient data were available, we checked them for internal consistency and consistency with published reports. One review author entered data into the Review Manager software (RevMan 2014), and a second review author checked the entries. We analysed binary outcome data using OR and 95% CI. We analysed continuous outcome data using SMD and 95% CI. By default, we used a fixed‐effect model first, but we corroborated results by using a random‐effects model if heterogeneity was significant.

When data were available, we carried out subgroup analyses for age, sex, stroke severity, and stroke type. Through subgroup analyses, we considered the degree of interaction between subgroups (Higgins 2011).

Network meta‐analysis

In this version of the review, we include a newer approach to meta‐analysis in the form of a network meta‐analysis (NMA) of trial data. The original aim of this review was to compare the effects of organised inpatient (stroke unit) care versus conventional care. We expected that within this broad definition, the included trials would comprise a range of treatment comparisons (which could include stroke wards, mobile stroke teams, mixed rehabilitation wards) with conventional care in general wards. In addition, later trials have addressed newer questions comparing different forms of organised inpatient (stroke unit) care (e.g. stroke ward versus mobile stroke team).

We have retained the previous analysis, which was organised in a hierarchical manner (organised stroke care versus alternative service; organised stroke care versus general ward; different systems of organised care). However, we now include an NMA to explore, when possible, the impact of different systems of stroke care. We used MetaInsight software, which uses a frequentist approach and is designed specifically for this role ‐ to enable us to conduct our NMA (Owen 2019).

An NMA uses information from both direct and indirect estimates of treatment effect (Tonin 2017). Direct estimates are provided by a head‐to‐head comparison (e.g. treatment A versus treatment B). Indirect estimates are provided by two or more head‐to‐head comparisons that share a common comparator (e.g. when A versus B is the comparison of interest, then trials with A versus C and with B versus C are used). A trial network is then formed, using trials that allow, through direct and indirect comparisons, calculation of the relative effects of all treatments versus each other (or versus a single comparator). The results of such analyses are usually presented as comparisons against a common comparator group (e.g. general ward). It is also possible to present a rank analysis, which ranks treatment groups on the likelihood of being most/least effective.

A key assumption in NMA is that of transitivity (or similarity). This concerns the validity of making indirect comparisons and assumes that treatment effects are 'exchangeable' across the included trials, and that all treatments are 'jointly randomisable'. In other words, all treatment categories could feasibly be randomised in the same trial, and those that are not treatment arms in any given trial are 'missing at random' (Lu 2006). As this assumption cannot be formally tested statistically, it was judged through consideration of trial settings and characteristics, patient characteristics, and treatment mechanisms, and to investigate if any differences would be expected to modify relative treatment effects. Two review authors independently extracted the relevant information, and the principal review author made the final judgement.

A second important assumption (known as the consistency assumption) assumes that it is feasible to make indirect comparisons between two treatments, and that the indirect evidence is consistent with the direct evidence (Lu 2006). The consistency assumption can be evaluated statistically by comparing the difference between the direct estimate and the indirect estimate for each loop of evidence. Therefore, we examined for any important differences in numerical results between direct, indirect, and network results.

GRADE and 'Summary of findings'

We constructed 'Summary of findings' tables and used GRADE criteria to assess the quality of evidence. 'Summary of findings' tables included the new primary outcome (poor outcome) and the three main clinical outcomes included in previous reviews (death, death or requiring institutional care, death or dependency) plus subjective health status, patient satisfaction, and length of stay in a hospital or institution. All were recorded at the end of scheduled follow‐up.

This update included an NMA whereby different types of organised inpatient (stroke unit) care were compared with care provided in a general ward (see below). For the NMA, we used the approach of the GRADE group as outlined below (Brignardello 2018; Puhan 2014).

Present direct and indirect treatment estimates for each comparison of the evidence network.
Rate the quality of each direct and indirect effect estimate (downgrading for risk of bias, inconsistency, indirectness, imprecision, and publication bias).
Present the NMA estimate for each comparison of the evidence network.
Rate the quality of each NMA effect estimate (as above).

Subgroup analysis and investigation of heterogeneity

Subgroup analyses involved a re‐analysis stratified by participant or service subgroup using tabular subgroup data provided by trialists or obtained from published sources. We used a fixed‐effect approach unless heterogeneity was statistically significant, and all subgroup analyses considered the degree of interaction between subgroups (Higgins 2011). We applied subgroup analyses only to the main (first) comparison to minimise the risk of false‐positive results.

Sensitivity analysis

We planned sensitivity analyses around key aspects of risk of bias that we identified during our assessment of risk of bias (i.e. concealment of treatment allocation, blinding of outcome assessment, completeness of follow‐up, and a fixed period of follow‐up). We applied sensitivity analyses only to the main (first) comparison.

Results

Description of studies

This is the fifth update of this Cochrane Review. The key references are described in the following relevant tables: Characteristics of included studies; Characteristics of excluded studies; Characteristics of studies awaiting classification; Characteristics of ongoing studies.

Results of the search

The search strategy for previous versions of this review yielded 28 eligible trials (Included studies), seven awaiting classification (Studies awaiting classification), three ongoing studies (Ongoing studies), and 28 excluded studies (Excluded studies).

For this updated review, searches of Embase, CINHAL, MEDLINE, and CENTRAL revealed 16,562 records. Searches of Cochrane trials registers and other ongoing trials registers identified 432 new potentially eligible trials for consideration based on the four selection criteria (Figure 1). After exclusion of duplicate records and those that were obviously irrelevant, we were left with 32 abstracts for screening. Of these, one was a new record of a previously identified study (Goteborg‐Sahlgren 1994), one was not randomised, and 19 described interventions that did not match organised inpatient (stroke unit) care.

Flow diagram illustrating the results of updated searches.

Of the remaining 13 articles retrieved, we excluded 10: seven were not randomised (Akhtar 2015; Al‐Qahtany 2014; Fu 2006; He 2014; Inoue 2013; Rai 2016; Raiborirug 2017), and three did not meet the definition of stroke unit (Felix 2016; Janssen 2014; Middleton 2018). Two are ongoing studies (China (Wang) 2015; Russia 2017), and we included one new trial (New South Wales 2014).

Therefore, this updated review incorporates 29 randomised controlled trials with 5902 participants.

Included studies

Service characteristics within organised (stroke unit) care and conventional care settings

Descriptive information was available for all trials: for eight trials, we had access to published information (Birmingham 1972; Guangdong 2008; Guangdong 2009; Huaihua 2004; Hunan 2007; Illinois 1966; New South Wales 2014; New York 1962); for two trials, we had detailed unpublished information (Beijing 2004; Joinville 2003); and for the remaining 19 trials, we carried out a structured interview with the trial co‐ordinator to determine the service characteristics.

Our original publication outlined the features of stroke unit trials (SUTC 1997a). In summary, organised inpatient (stroke unit) care was characterised by:

co‐ordinated multi‐disciplinary rehabilitation;
staff with a specialist interest in stroke or rehabilitation, or both;
routine involvement of carers in the rehabilitation process; and
regular programmes of education and training.

Several factors indicating more intensive or more comprehensive input of care were also associated with the stroke unit setting. Various service models of care exist (Table 5), but core characteristics that were invariably included in the stroke unit setting were (1) multi‐disciplinary staffing, that is, medical, nursing, and therapy staff (usually including physiotherapy, occupational therapy, speech therapy, social work); and (2) co‐ordinated multi‐disciplinary team care incorporating meetings at least once per week (SUTC 1997a). When both of the compared services could satisfy the description of stroke unit care, the local conventional system of care was taken as the control service.

1. Typical characteristics of different models of organised stroke care.

Name	Structure	Patients	Service type	Admission	Discharge	Features
Stroke ward	Ward	Stroke	Various (see below)	Various (see below)	Various (see below)	Various (see below)
Acute stroke ward	Ward	Stroke	Acute	Acute (hours)	Days	Close physiological monitoring, often followed by care in separate rehabilitation ward if required
Comprehensive stroke ward^a	Ward	Stroke	Acute, rehabilitation	Acute (hours)	Days to weeks	Acute care and rehabilitation; conventional staffing levels
Rehabilitation stroke ward	Ward	Stroke	Rehabilitation	Delayed (days)	Weeks	Focus on rehabilitation
Mobile stroke team	Mobile team	Stroke	Mobile stroke team	Variable	Days to weeks	Peripatetic care to patients in general wards; medical and rehabilitation advice
Mixed rehabilitation ward	Ward	Mixed	Rehabilitation	Variable	Weeks	Mixed patient group; focus on rehabilitation
Intensive care	Ward	Mixed	Acute, intensive	Acute (hours)	Days	High nurse staffing; life support facilities

Trials	Participants	Index (stroke unit) care	Conventional care	Reference
15	3521	Stroke ward	General medical ward	Athens 1995, Beijing 2004, Dover 1984 (GMW), Edinburgh 1980, Goteborg‐Ostra 1988, Goteborg‐Sahlgren 1994, Guangdong 2009, Huaihua 2004, Joinville 2003, Nottingham 1996 (GMW), Orpington 1993 (GMW), Orpington 1995, Perth 1997, Svendborg 1995, Trondheim 1991
6	630	Mixed rehabilitation ward	General medical ward	Birmingham 1972, Helsinki 1995, Illinois 1966, Kuopio 1985, New York 1962; Newcastle 1993
2	438	Mobile stroke team (peripatetic care)	General medical ward	Manchester 2003, Montreal 1985
4	542	Stroke ward	Mixed rehabilitation ward	Dover 1984 (MRW), Nottingham 1996 (MRW), Orpington 1993 (MRW), Tampere 1993
1	304	Stroke ward (comprehensive)	Mobile stroke team	Orpington 2000
1	54	Stroke ward (acute)	Stroke ward (comprehensive unit)	Groningen 2003
1	47	Stroke ward (comprehensive)	Stroke wards (acute+rehabilitation)	New South Wales 2014
2	366	Stroke ward (plus TCM)	Stroke ward (without TCM)	Guangdong 2008, Hunan 2007

Organised inpatient (stroke unit) care compared with alternative service
Patient or population: adults with acute stroke Settings: hospital Intervention: organised inpatient (stroke unit) care Comparison: alternative service (contemporary conventional care)
Outcomes	*Illustrative comparative risks (95% CI)**		Relative effect (95% CI)	Number of participants (studies)	Quality of the evidence (GRADE)	Comments
	Assumed risk	Corresponding risk
	Alternative service	Organised inpatient (stroke unit) care
Poor outcome by the end of scheduled follow‐up (modified Rankin score 3 to 6 or requiring institutional care; median 12‐month follow‐up) (Analysis 1.1)	577 per 1000	517 per 1000 (497 to 547)	OR 0.77 (0.69 to 0.87)	5336 (26)	⊕⊕⊕⊝ moderate^a	Sensitivity analysis based on trial quality suggested no alteration of conclusions
Death by the end of scheduled follow‐up (median 12‐month follow‐up) (Analysis 1.2)	219 per 1000	199 per 1000 (179 to 209)	OR 0.76 (0.66 to 0.88)	5902 (29)	⊕⊕⊕⊝ moderate^a	Sensitivity analysis based on trial quality suggested no alteration of conclusions
Death or institutional care by the end of scheduled follow‐up (median 12‐month follow‐up) (Analysis 1.3)	405 per 1000	345 per 1000 (315 to 375)	OR 0.76 (0.67 to 0.85)	4887 (25)	⊕⊕⊕⊝ moderate^a	Sensitivity analysis based on trial quality suggested no alteration of conclusions
Death or dependency by the end of scheduled follow‐up (modified Rankin score 3 to 6; median 12‐month follow‐up) (Analysis 1.4)	609 per 1000	549 per 1000 (519 to 567)	OR 0.75 (0.66 to 0.85)	4854 (24)	⊕⊕⊕⊝ moderate^a	Sensitivity analysis based on trial quality suggested no alteration of conclusions
Subjective health status score Participant quality of life (Nottingham Health Profile; Quality of Life Scale)	There was a pattern of improved results among stroke unit survivors, with results attaining statistical significance in 2 individual trials		N/A	843 (3)	⊕⊝⊝⊝ very low^a,b,c	Data from 3 trials only High rate of missing data
Patient satisfaction or preference	We could find no systematically gathered information on patient preferences		N/A	N/A	N/A	No data available
Length of stay (days) in a hospital or institution (Analysis 1.5)	Mean length of stay across the control groups ranged from 12.1 to 123 days	Mean length of stay for the intervention groups was, on average, 4.3 days less (7.9 days less to 0.7 days more)	SMD 0.16 lower (0.33 lower to 0.01 higher)	4162 (19)	⊕⊕⊝⊝ low^a,b	Different definitions and imprecise measures of length of stay were reported
The basis for the assumed risk* (e.g. median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: confidence interval; N/A: not applicable; OR: odds ratio; SMD: standardised mean difference.
GRADE Working Group grades of evidence. High quality: further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: we are very uncertain about the estimate.

Mobile stroke team care compared with general medical ward care for stroke
Patient or population: adults with acute stroke Settings: hospital Intervention: mobile stroke team care Comparison: general medical ward care
Outcomes	*Illustrative comparative risks (95% CI)**		Relative effect (95% CI)	Number of participants (studies)	Quality of the evidence (GRADE)	Comments
	Assumed risk	Corresponding risk
	General medical ward care	Mobile stroke team care
Poor outcome by the end of scheduled follow‐up (modified Rankin score 3 to 6 or requiring institutional care; median 12‐month follow‐up) (Analysis 3.1)	712 per 1000	672 per 1000 (582 to 752)	OR 0.80 (0.52 to 1.22)	438 (2)	⊕⊕⊝⊝ low^a,b	As dependency data were complete, these are the same data as for death or dependency
Death by the end of scheduled follow‐up (median 12‐month follow‐up) (Analysis 3.2)	259 per 1000	279 per 1000 (189 to 359)	OR 1.08 (0.71 to 1.65)	438 (2)	⊕⊕⊝⊝ low^a,b
Death or institutional care by the end of scheduled follow‐up (median 12‐month follow‐up) (Analysis 3.3)	481 per 1000	531 per 1000 (451 to 611)	OR 1.27 (0.84 to 1.93)	438 (2)	⊕⊕⊝⊝ low^a,b
Death or dependency by the end of scheduled follow‐up (modified Rankin score 3 to 6; median 12‐month follow‐up) (Analysis 3.4)	712 per 1000	672 per 1000 (582 to 752)	OR 0.80 (0.52 to 1.22)	438 (2)	⊕⊕⊝⊝ low^a,b	As dependency data were complete, these are the same data as for poor outcome
Subjective health status score Participant quality of life (EuroQol)	No apparent differences between groups		N/A	308 (1)	⊕⊝⊝⊝ very low^a,b	Data from 1 trial only
Patient satisfaction or preference	We could find no systematically gathered information on patient preferences		N/A	N/A	N/A	No data available
Length of stay (days) in a hospital or institution (Analysis 3.5)	No data available		N/A	N/A	N/A	No data available
The basis for the assumed risk* (e.g. median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: confidence interval; N/A: not applicable; OR: odds ratio.
GRADE Working Group grades of evidence. High quality: further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: we are very uncertain about the estimate.

Mixed rehabilitation ward care compared with general medical ward care for stroke
Patient or population: adults with acute stroke Settings: hospital Intervention: mixed rehabilitation ward care Comparison: general medical ward care
Outcomes	*Illustrative comparative risks (95% CI)**		Relative effect (95% CI)	Number of participants (studies)	Quality of the evidence (GRADE)	Comments
	Assumed risk	Corresponding risk
	General medical ward care	Mixed rehabilitation wardcare
Poor outcome by the end of scheduled follow‐up (modified Rankin score 3 to 6 or requiring institutional care; median 12‐month follow‐up) (Analysis 4.1)	574 per 1000	474 per 1000 (404 to 554)	OR 0.65 (0.47 to 0.90)	630 (6)	⊕⊕⊝⊝ low^a,b	As dependency data were complete, these are the same data as for death or dependency
Death by the end of scheduled follow‐up (median 12‐month follow‐up) (Analysis 4.2)	171 per 1000	161 per 1000 (101 to 211)	OR 0.91 (0.58 to 1.42)	630 (6)	⊕⊕⊝⊝ low^a,b
Death or institutional care by the end of scheduled follow‐up (median 12‐month follow‐up) (Analysis 4.3)	451 per 1000	371 per 1000 (291 to 451)	OR 0.71 (0.51 to 0.99)	578 (5)	⊕⊕⊝⊝ low^a,b
Death or dependency by the end of scheduled follow‐up (modified Rankin score 3 to 6; median 12‐month follow‐up) (Analysis 4.4)	574 per 1000	474 per 1000 (404 to 554)	OR 0.65 (0.47 to 0.90)	630 (6)	⊕⊕⊝⊝ low ^a,b	As dependency data were complete, these are the same data as for poor outcome
Subjective health Status score	No data available		N/A	N/A	N/A	No data available
Patient satisfaction or preference	We could find no systematically gathered information on patient preferences		N/A	N/A	N/A	No data available
Length of stay (days) in a hospital or institution (Analysis 4.5)	Mean length of stay across control groups ranged from 30.5 to 129.5 days	Mean length of stay for the intervention groups was, on average, 3.9 days more (13.5 days less to 21.5 days more)	MD 0.08 more (0.21 lower to 0.37 higher)	387 (3)	⊕⊝⊝⊝ very low^a,b,c	Different definitions and imprecise measures of length of stay were reported
The basis for the assumed risk* (e.g. median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: confidence interval; MD: mean difference; N/A: not applicable; OR: odds ratio.
GRADE Working Group grades of evidence. High quality: further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: we are very uncertain about the estimate.

Stroke ward
15 trials (3521 participants)	General medical ward
4 trials (542 participants)	6 trials (630 participants)	Mixed rehabilitation ward
1 trial (304 participants)	2 trials (438 participants)	‐‐	Mobile stroke team

Comparison	Number of studies	Log direct comparison OR	Log indirect comparison OR	Log difference (95% CI) between direct and indirect comparisons	P value of difference between direct and indirect comparisons
Mixed rehabilitation ward vs GMW	6	‐0.413	‐0.266	‐0.147 (‐0.776 to 0.483)	0.65
Mobile stroke team vs GMW	2	‐0.238	0.032	‐0.270 (‐1.111 to 0.571)	0.53
Stroke ward vs GMW	13	‐0.273	‐0.478	0.205 (‐0.317 to 0.728)	0.44
Mixed rehabilitation ward vs mobile stroke team	0	N/A	‐0.237	N/A	N/A
Mixed rehabilitation ward vs stroke ward	4	0.022	‐0.124	0.146 (‐0.483 to 0.776)	0.65
Mobile stroke team vs stroke ward	1	0.321	0.051	0.270 (‐0.571 to 1.111)	0.53

Intervention organised inpatient (stroke unit) care	Comparison GMW	Number of studies (participants) with direct comparison evidence	Direct comparison evidence OR (95% CI)	Quality of the evidence (GRADE) for direct comparisons	Direct plus indirect evidence in NMA OR (95% CI)	Quality of the evidence (GRADE) for NMA
Poor outcome at end of scheduled follow‐up
Stroke ward	GMW	14 (3321)	0.78 (0.68 to 0.91)	Moderate^a	0.74 (0.62 to 0.89)	Moderate^a
Mobile stroke team	GMW	2 (438)	0.80 (0.52 to 1.22)	Low^a,b	0.88 (0.58 to 1.34)	Low^a,b
Mixed rehabilitation ward	GMW	6 (630)	0.65 (0.47 to 0.90)	Moderate^a	0.70 (0.52 to 0.95)	Low^a,b
Death at end of scheduled follow‐up
Stroke ward	GMW	15 (3523)	0.75 (0.63 to 0.90)	Moderate^a	0.62 (0.47 to 0.82)	Moderate^a
Mobile stroke team	GMW	2 (438)	1.08 (0.71 to 1.65)	Low^a,b	1.23 (0.67 to 2.27)	Low^a,b
Mixed rehabilitation ward	GMW	6 (630)	0.91 (0.58 to 1.42)	Low^a,b	1.20 (0.73 to 1.99)	Low^a,b
Death or institutional care at end of scheduled follow‐up
Stroke ward	GMW	13 (2924)	0.74 (0.63 to 0.87)	Moderate^a	0.72 (0.62 to 0.83)	Moderate^a
Mobile stroke team	GMW	2 (438)	1.27 (0.64 to 1.27)	Low^a,b	1.46 (1.03 to 2.05)	Low^a,b
Mixed rehabilitation ward	GMW	5 (578)	0.71 (0.51 to 0.99)	Low^a,b	0.75 (0.58 to 0.96)	Low^a,b
Death or dependency at end of scheduled follow‐up
Stroke ward	GMW	12 (2839)	0.75 (0.64 to 0.88)	Moderate^a	0.71 (0.58 to 0.86)	Moderate^a
Mobile stroke team	GMW	2 (438)	0.80 (0.52 to 1.22)	Low^a,b	0.87 (0.57 to 1.32)	Low^a,b
Mixed rehabilitation ward	GMW	6 (630)	0.65 (0.47 to 0.90)	Moderate^a	0.69 (0.51 to 0.93)	Low^a,b

Date	Event	Description
2 April 2019	New search has been performed	This update incorporated (1) a revised literature search (updated to 2 April 2019), (2) 1 new included trial, (3) a single primary outcome (poor outcome: death or dependency or requiring institutional care) in addition to the previous outcomes, (4) a revision of the data presentation, (5) a new network meta‐analysis, and (6) new 'Summary of findings' tables with GRADE quality of evidence classifications. The review now includes 29 studies involving 5902 participants
2 April 2019	New citation required but conclusions have not changed	The conclusions have not changed and remain similar to the previous version

Date	Event	Description
6 January 2017	Feedback has been incorporated	Feedback has been incorporated
14 July 2014	Feedback has been incorporated	Feedback has been incorporated and numerical error in the Abstract has been corrected with no change to the conclusions
29 January 2013	New search has been performed	This updated review identified 4 new trials (763 participants). We have excluded 7 previously included quasi‐randomised prospective controlled clinical trials. This review now incorporates an individual patient data meta‐analysis of 28 randomised controlled trials (5855 participants). More recent stroke unit trials have addressed different ways of providing organised care. This update contains data from trials comparing stroke unit care with care given in general medical wards and comparing 2 different forms of organised (stroke unit) care
29 January 2013	New citation required but conclusions have not changed	The conclusions of the review have not changed
9 September 2008	Amended	The review has been converted to new review format
28 November 2006	New search has been performed	New data on 2027 participants from 8 new trials (Athens, Beijing, Cape Town, Groningen, Joinville, Manchester, Osaka, and Pavia) have become available. More recent stroke unit trials have addressed different ways of providing organised care. This update contains new information and data from trials comparing stroke unit care with care provided on general medical wards and comparing 2 different forms of organised (stroke unit) care

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1.1 Poor outcome by the end of scheduled follow‐up	29	5336	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.77 [0.69, 0.87]
1.1.1 Stroke ward vs general medical ward	14	3321	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.78 [0.68, 0.91]
1.1.2 Mixed rehabilitation ward vs general medical ward	6	630	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.65 [0.47, 0.90]
1.1.3 Mobile stroke team vs general medical ward	2	438	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.80 [0.52, 1.22]
1.1.4 Stroke ward vs mixed rehabilitation ward	4	542	Peto Odds Ratio (Peto, Fixed, 95% CI)	1.01 [0.68, 1.50]
1.1.5 Stroke ward vs mobile stroke team	1	304	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.73 [0.46, 1.14]
1.1.6 Stroke ward vs stroke ward	2	101	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.47 [0.21, 1.04]
1.2 Death by the end of scheduled follow‐up	32	5902	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.76 [0.66, 0.88]
1.2.1 Stroke ward vs general medical ward	15	3521	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.75 [0.63, 0.90]
1.2.2 Mixed rehabilitation ward vs general medical ward	6	630	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.91 [0.58, 1.42]
1.2.3 Mobile stroke team vs general medical ward	2	438	Peto Odds Ratio (Peto, Fixed, 95% CI)	1.08 [0.71, 1.65]
1.2.4 Stroke ward vs mixed rehabilitation ward	4	542	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.82 [0.54, 1.24]
1.2.5 Stroke ward vs mobile stroke team	1	304	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.35 [0.19, 0.65]
1.2.6 Stroke ward vs stroke ward	4	467	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.36 [0.14, 0.94]
1.3 Death or institutional care by the end of scheduled follow‐up	27	4887	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.76 [0.67, 0.85]
1.3.1 Stroke ward vs general medical ward	13	2924	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.74 [0.63, 0.87]
1.3.2 Mixed rehabilitation ward vs general medical ward	5	578	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.71 [0.51, 0.99]
1.3.3 Mobile stroke team vs general medical ward	2	438	Peto Odds Ratio (Peto, Fixed, 95% CI)	1.27 [0.84, 1.93]
1.3.4 Stroke ward vs mixed rehabilitation ward	4	542	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.90 [0.64, 1.27]
1.3.5 Stroke ward vs mobile stroke team	1	304	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.40 [0.23, 0.68]
1.3.6 Stroke ward vs stroke ward	2	101	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.39 [0.16, 0.93]
1.4 Death or dependency by the end of scheduled follow‐up	27	4854	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.75 [0.66, 0.85]
1.4.1 Stroke ward vs general medical ward	12	2839	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.75 [0.64, 0.88]
1.4.2 Mixed rehabilitation ward vs general medical ward	6	630	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.65 [0.47, 0.90]
1.4.3 Mobile stroke team vs general medical ward	2	438	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.80 [0.52, 1.22]
1.4.4 Stroke ward vs mixed rehabilitation ward	4	542	Peto Odds Ratio (Peto, Fixed, 95% CI)	1.01 [0.68, 1.50]
1.4.5 Stroke ward vs mobile stroke team	1	304	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.73 [0.46, 1.14]
1.4.6 Stroke ward vs stroke ward	2	101	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.47 [0.21, 1.04]
1.5 Length of stay (days) in a hospital or institution or both	20	4162	Mean Difference (IV, Random, 95% CI)	‐4.28 [‐7.86, ‐0.71]
1.5.1 Stroke ward	17	3775	Mean Difference (IV, Random, 95% CI)	‐4.76 [‐8.46, ‐1.05]
1.5.2 Mixed rehabilitation ward	3	387	Mean Difference (IV, Random, 95% CI)	3.85 [‐13.49, 21.18]
1.6 Length of stay (days) in a hospital or hospital plus institution	20		Mean Difference (IV, Random, 95% CI)	Subtotals only
1.6.1 Acute hospital stay only	7	1817	Mean Difference (IV, Random, 95% CI)	‐2.78 [‐6.13, 0.56]
1.6.2 Hospital and institution stay	13	2345	Mean Difference (IV, Random, 95% CI)	‐4.84 [‐14.52, 4.84]
1.7 Poor outcome at 5‐year follow‐up	2	535	Odds Ratio (M‐H, Random, 95% CI)	0.54 [0.22, 1.34]
1.8 Death at 5‐year follow‐up	3	1139	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.74 [0.59, 0.94]
1.9 Death or institutional care at 5‐year follow‐up	2	535	Odds Ratio (M‐H, Random, 95% CI)	0.59 [0.33, 1.05]
1.10 Death or dependency at 5‐year follow‐up	2	535	Odds Ratio (M‐H, Random, 95% CI)	0.54 [0.22, 1.34]
1.11 Poor outcome at 10‐year follow‐up	2	535	Odds Ratio (M‐H, Random, 95% CI)	0.70 [0.27, 1.80]
1.12 Death at 10‐year follow‐up	3	1139	Odds Ratio (M‐H, Random, 95% CI)	0.66 [0.43, 1.03]
1.13 Death or institutional care at 10‐year follow‐up	2	535	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.57 [0.37, 0.88]
1.14 Death or dependency at 10‐year follow‐up	2	535	Odds Ratio (M‐H, Random, 95% CI)	0.70 [0.27, 1.80]

*Study characteristics*
Methods	RCT Sealed envelopes Unblinded follow‐up
Participants	People with acute stroke admitted to emergency department within 24 hours of symptoms Excluded TIA or recurrent stroke
Interventions	Small (6‐bed) ward within Internal Medicine department Used the American Heart Association protocol, management of physiological abnormalities, multi‐disciplinary team approach Compared with conventional care in general medical ward
Outcomes	Death, cause of death, length of stay Recorded up to 6.5 years (12‐month data used in primary analysis)
Notes	Unpublished at present
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"Randomised ... using numbered opaque sealed envelopes"
Allocation concealment (selection bias)	Low risk	"Opaque sealed envelopes"
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	Difficult to conceal
Blinding of outcome assessment (detection bias) All outcomes	High risk	Unblinded outcome assessment
Incomplete outcome data (attrition bias) All outcomes	Low risk	No missing outcome data
Selective reporting (reporting bias)	Low risk	All pre‐specified outcomes reported

*Study characteristics*
Methods	RCT
Participants	People with acute stroke within 7 days of stroke
Interventions	Comprehensive stroke ward (n = 215) within general medical service vs conventional care in general medical ward (n = 202)
Outcomes	Death, Barthel Index, place of residence, length of hospital stay recorded at discharge
Notes	Not yet published
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Computer‐generated randomisation in closed envelopes
Allocation concealment (selection bias)	Low risk	Adequate allocation concealment
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	Difficult to conceal
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	Unclear
Incomplete outcome data (attrition bias) All outcomes	Unclear risk	Unclear
Selective reporting (reporting bias)	Unclear risk	Unclear

*Study characteristics*
Methods	RCT
Participants	People (56 men) with acute ischaemic stroke; timing of randomisation unclear Mean age intervention group: 61.4 years (SD 9.05); mean age control group: 60.9 years (SD 8.2)
Interventions	Stroke unit plus integrated traditional Chinese medicine (n = 58) vs 'Western medicine' stroke unit (n = 42)
Outcomes	Death, NIHSS at 30 days, Barthel Index Length of follow‐up unclear
Notes	Limited translated data available
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Random number generator
Allocation concealment (selection bias)	Unclear risk	Unclear
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	Unclear
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	Unclear
Incomplete outcome data (attrition bias) All outcomes	Unclear risk	Unclear
Selective reporting (reporting bias)	Unclear risk	Unclear

*Study characteristics*
Methods	RCT
Participants	Participants (137 men) with acute ischaemic stroke, randomised on admission Average age 61.9 years in intervention group vs 63.4 years in control group
Interventions	Stroke unit with integrated traditional Chinese medicine (n = 100) vs general medical ward (n = 100)
Outcomes	Death, dependency (Barthel Index, OHS), discharge NIHSS Cost‐effectiveness analysis
Notes	Limited translated data available Overall numbers in intervention and control groups differed between original publication and data in published meta‐analysis
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Random number generator
Allocation concealment (selection bias)	Unclear risk	Unclear
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	Unclear
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	Unclear
Incomplete outcome data (attrition bias) All outcomes	Low risk	No missing outcome data
Selective reporting (reporting bias)	Unclear risk	Unclear

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
2.1 Poor outcome by the end of scheduled follow‐up	14	3321	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.78 [0.68, 0.91]
2.1.1 Comprehensive stroke ward vs general medical ward	10	2788	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.78 [0.66, 0.91]
2.1.2 Rehabilitation stroke ward vs general medical ward	4	533	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.83 [0.57, 1.23]
2.2 Death by the end of scheduled follow‐up	15	3523	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.75 [0.63, 0.90]
2.2.1 Comprehensive stroke ward vs general medical ward	11	2988	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.77 [0.63, 0.93]
2.2.2 Rehabilitation stroke ward vs general medical ward	4	535	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.69 [0.46, 1.05]
2.3 Death or institutional care by the end of scheduled follow‐up	13	2924	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.74 [0.63, 0.87]
2.3.1 Comprehensive stroke ward vs general medical ward	9	2391	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.74 [0.62, 0.88]
2.3.2 Rehabilitation stroke ward vs general medical ward	4	533	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.76 [0.52, 1.09]
2.4 Death or dependency by the end of scheduled follow‐up	12	2839	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.75 [0.64, 0.88]
2.4.1 Comprehensive stroke ward vs general medical ward	8	2306	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.73 [0.62, 0.87]
2.4.2 Rehabilitation stroke ward vs general medical ward	4	533	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.83 [0.57, 1.23]
2.5 Length of stay (days) in a hospital or institution	10	2547	Mean Difference (IV, Random, 95% CI)	‐2.19 [‐5.19, 0.82]
2.5.1 Comprehensive stroke ward vs general medical ward	9	2373	Mean Difference (IV, Random, 95% CI)	‐2.79 [‐5.68, 0.10]
2.5.2 Rehabilitation stroke ward vs general medical ward	1	174	Mean Difference (IV, Random, 95% CI)	16.34 [2.82, 29.86]

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
3.1 Poor outcome by the end of scheduled follow‐up	2	438	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.80 [0.52, 1.22]
3.1.1 Mobile stroke team vs general medical ward	2	438	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.80 [0.52, 1.22]
3.2 Death by the end of scheduled follow‐up	2	438	Peto Odds Ratio (Peto, Fixed, 95% CI)	1.08 [0.71, 1.65]
3.2.1 Mobile stroke team vs general medical ward	2	438	Peto Odds Ratio (Peto, Fixed, 95% CI)	1.08 [0.71, 1.65]
3.3 Death or institutional care by the end of scheduled follow‐up	2	438	Peto Odds Ratio (Peto, Fixed, 95% CI)	1.27 [0.84, 1.93]
3.3.1 Mobile stroke team vs general medical ward	2	438	Peto Odds Ratio (Peto, Fixed, 95% CI)	1.27 [0.84, 1.93]
3.4 Death or dependency by the end of scheduled follow‐up	2	438	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.80 [0.52, 1.22]
3.4.1 Mobile stroke team vs general medical ward	2	438	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.80 [0.52, 1.22]
3.5 Length of stay (days) in a hospital or institution	0	0	Std. Mean Difference (IV, Random, 95% CI)	Not estimable
3.5.1 Mobile stroke team vs general medical ward	0	0	Std. Mean Difference (IV, Random, 95% CI)	Not estimable

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
4.1 Poor outcome by the end of scheduled follow‐up	6	630	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.65 [0.47, 0.90]
4.1.1 Mixed rehabilitation ward vs general medical ward	6	630	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.65 [0.47, 0.90]
4.2 Death by the end of scheduled follow‐up	6	630	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.91 [0.58, 1.42]
4.2.1 Mixed rehabilitation ward vs general medical ward	6	630	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.91 [0.58, 1.42]
4.3 Death or institutional care by the end of scheduled follow‐up	5	578	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.71 [0.51, 0.99]
4.3.1 Mixed rehabilitation ward vs general medical ward	5	578	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.71 [0.51, 0.99]
4.4 Death or dependency by the end of scheduled follow‐up	6	630	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.65 [0.47, 0.90]
4.4.1 Mixed rehabilitation ward vs general medical ward	6	630	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.65 [0.47, 0.90]
4.5 Length of stay (days) in a hospital or institution	3	387	Mean Difference (IV, Random, 95% CI)	3.85 [‐13.49, 21.18]
4.5.1 Mixed rehabilitation ward vs general ward	3	387	Mean Difference (IV, Random, 95% CI)	3.85 [‐13.49, 21.18]

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
5.1 Poor outcome by the end of scheduled follow‐up	7		Odds Ratio (M‐H, Random, 95% CI)	Subtotals only
5.1.1 Acute ward vs mixed rehabilitation ward	1	211	Odds Ratio (M‐H, Random, 95% CI)	1.24 [0.72, 2.14]
5.1.2 Acute stroke ward vs comprehensive stroke ward	2	101	Odds Ratio (M‐H, Random, 95% CI)	0.81 [0.18, 3.69]
5.1.3 Comprehensive stroke ward vs mobile stroke team	1	304	Odds Ratio (M‐H, Random, 95% CI)	0.73 [0.46, 1.14]
5.1.4 Rehabilitation stroke ward vs mixed rehabilitation ward	3	331	Odds Ratio (M‐H, Random, 95% CI)	0.81 [0.45, 1.45]
5.2 Death by the end of scheduled follow‐up	9		Odds Ratio (M‐H, Random, 95% CI)	Subtotals only
5.2.1 Acute stroke ward vs mixed rehabilitation ward	1	211	Odds Ratio (M‐H, Random, 95% CI)	1.41 [0.76, 2.58]
5.2.2 Acute stroke ward vs comprehensive stroke ward	2	101	Odds Ratio (M‐H, Random, 95% CI)	0.30 [0.03, 2.90]
5.2.3 Comprehensive stroke ward vs mobile stroke team	1	304	Odds Ratio (M‐H, Random, 95% CI)	0.32 [0.16, 0.64]
5.2.4 Rehabilitation stroke ward vs mixed rehabilitation ward	3	331	Odds Ratio (M‐H, Random, 95% CI)	0.50 [0.28, 0.90]
5.2.5 Stroke ward (plus TCM) vs stroke ward (without TCM)	2	366	Odds Ratio (M‐H, Random, 95% CI)	0.54 [0.13, 2.30]
5.3 Death or institutional care by the end of scheduled follow‐up	7		Odds Ratio (M‐H, Random, 95% CI)	Subtotals only
5.3.1 Acute ward vs mixed rehabilitation ward	1	211	Odds Ratio (M‐H, Random, 95% CI)	1.32 [0.76, 2.30]
5.3.2 Acute stroke ward vs comprehensive stroke ward	2	101	Odds Ratio (M‐H, Random, 95% CI)	1.28 [0.14, 11.31]
5.3.3 Comprehensive stroke ward vs mobile stroke team	1	304	Odds Ratio (M‐H, Random, 95% CI)	0.38 [0.21, 0.68]
5.3.4 Rehabilitation stroke ward vs mixed rehabilitation ward	3	331	Odds Ratio (M‐H, Random, 95% CI)	0.70 [0.45, 1.09]
5.4 Death or dependency by the end of scheduled follow‐up	7		Odds Ratio (M‐H, Random, 95% CI)	Subtotals only
5.4.1 Acute ward vs mixed rehabilitation ward	1	211	Odds Ratio (M‐H, Random, 95% CI)	1.24 [0.72, 2.14]
5.4.2 Acute stroke ward vs comprehensive stroke ward	2	101	Odds Ratio (M‐H, Random, 95% CI)	0.81 [0.18, 3.69]
5.4.3 Comprehensive stroke ward vs mobile stroke team	1	304	Odds Ratio (M‐H, Random, 95% CI)	0.73 [0.46, 1.14]
5.4.4 Rehabilitation stroke ward vs mixed rehabilitation ward	3	331	Odds Ratio (M‐H, Random, 95% CI)	0.81 [0.45, 1.45]
5.5 Length of stay (days) in a hospital or institution	7		Mean Difference (IV, Random, 95% CI)	Subtotals only
5.5.1 Acute stroke ward vs mixed rehabilitation ward	1	211	Mean Difference (IV, Random, 95% CI)	‐2.00 [‐11.19, 7.19]
5.5.2 Acute (±rehabilitation) stroke ward vs comprehensive stroke ward	2	101	Mean Difference (IV, Random, 95% CI)	‐2.89 [‐20.10, 14.33]
5.5.3 Comprehensive stroke ward vs mobile stroke team	1	304	Mean Difference (IV, Random, 95% CI)	2.50 [‐5.42, 10.42]
5.5.4 Rehabilitation stroke ward vs mixed rehabilitation ward	3	331	Mean Difference (IV, Random, 95% CI)	15.80 [‐45.71, 77.31]

*Study characteristics*
Methods	RCT Divided randomly using SPSS software package
Participants	People with stroke admitted to hospital with first or recurrent stroke Subarachnoid haemorrhage and tumour were excluded
Interventions	New comprehensive stroke unit, early multi‐disciplinary rehabilitation Control participants were admitted to general medical or general neurology ward
Outcomes	Death, NIHSS, Barthel Index, Oxford Handicap Scale, patient satisfaction at time of discharge
Notes	Some unpublished data included No institutional care available
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"Divided randomly into two groups using SPSS software package"
Allocation concealment (selection bias)	Unclear risk	Method of allocation concealment not described
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	Difficult to conceal
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	Unclear
Incomplete outcome data (attrition bias) All outcomes	Low risk	Similar numbers in treatment (n = 20) and control (n = 21) groups with missing data
Selective reporting (reporting bias)	Low risk	Data on all pre‐specified outcomes reported

*Study characteristics*
Methods	RCT Subgroup of Dover 1984 (stroke unit vs general medical ward)
Participants	People with stroke up to 9 weeks after stroke onset (majority within 3 weeks) Fit for transfer to rehabilitation ward
Interventions	Stroke rehabilitation ward (dedicated stroke unit) (n = 98) vs general medical ward (n = 89) Organised care provided for months if required
Outcomes	Death, Rankin score, place of residence, length of stay in hospital up to 1 year after stroke
Notes	Stroke severity subgroup data inferred from distribution in the whole group
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"Random allocation ... by the secretary opening the next in a stack of serially numbered sealed envelopes"
Allocation concealment (selection bias)	Low risk	Adequate allocation concealment
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	Difficult to conceal
Blinding of outcome assessment (detection bias) All outcomes	High risk	Unblinded outcome assessment
Incomplete outcome data (attrition bias) All outcomes	Low risk	Missing outcome data explained; broadly similar numbers between intervention and control groups
Selective reporting (reporting bias)	Unclear risk	Rankin score pre‐specified but not reported Disability reported in an alternate way

*Study characteristics*
Methods	RCT Blinded assessment of outcomes
Participants	People with acute ischaemic stroke admitted within 24 hours (conscious, hemiparetic, no prior dependency)
Interventions	Acute stroke unit with continuous physiological monitoring and intervention for 48 hours All other care as per conventional stroke unit Transfer to conventional stroke unit after 48 hours Conventional stroke unit: comprehensive stroke ward with intermittent physiological monitoring Both units had a multi‐disciplinary team meeting once per week Both units had discharge for rehabilitation at about 2 weeks
Outcomes	Death or poor outcome (institutional care or Rankin score > 3 or Barthel Index < 12) recorded at 3 months Complications and interventions, length of stay
Notes
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"Randomised ... using an envelope system on a one to one basis ..."
Allocation concealment (selection bias)	Low risk	Opaque sealed envelopes
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	Difficult to conceal but similar care routines
Blinding of outcome assessment (detection bias) All outcomes	Low risk	Blinded outcome assessment
Incomplete outcome data (attrition bias) All outcomes	Low risk	No missing outcome data
Selective reporting (reporting bias)	Low risk	All pre‐specified outcome data reported

*Study characteristics*
Methods	RCT with 3:2 allocation to intervention:control
Participants	People with stroke up to 1 year after stroke onset Appropriate for rehabilitation service
Interventions	Rehabilitation service (mixed rehabilitation unit) (n = 56) vs general medical ward (which had some specialist nursing input) (n = 35) Organised care provided for months if required
Outcomes	Functional status and place of residence at end of follow‐up
Notes	Intervention and control services not clearly defined No deaths reported
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Fisher's table of random numbers
Allocation concealment (selection bias)	Unclear risk	Method of allocation concealment not described
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	Unclear
Blinding of outcome assessment (detection bias) All outcomes	High risk	Unblinded outcome assessment
Incomplete outcome data (attrition bias) All outcomes	Low risk	No missing outcome data reported
Selective reporting (reporting bias)	Unclear risk	Outcomes were not clearly pre‐specified

*Study characteristics*
Methods	RCT by means of randomised numbers in the emergency room Blinded follow‐up
Participants	Clinical stroke diagnosis (confirmed on CT scan) within 7 days of onset
Interventions	Comprehensive stroke unit within neurology department (n = 35) vs conventional care in general medical ward
Outcomes	Death, Rankin score, length of stay up to 6 months
Notes	‐
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"List of randomized numbers available in the emergency room"
Allocation concealment (selection bias)	Unclear risk	Method of allocation concealment not described
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	Difficult to conceal
Blinding of outcome assessment (detection bias) All outcomes	Low risk	Blinded outcome assessment
Incomplete outcome data (attrition bias) All outcomes	Low risk	No missing outcome data
Selective reporting (reporting bias)	Low risk	All pre‐specified outcomes reported

*Study characteristics*
Methods	RCT Blinded assessment of outcome
Participants	People with stroke within 7 days of stroke onset Able to tolerate intensive rehabilitation
Interventions	Intensive rehabilitation in neurological rehabilitation unit (mixed rehabilitation ward) (n = 50) vs general ward (n = 45) Organised care provided for months if required
Outcomes	Death, Lehman (disability) score, place of residence, total time in hospital up to 1 year after stroke
Notes	Majority of people screened failed to meet inclusion criteria for the trial
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"Randomised using sealed envelopes"
Allocation concealment (selection bias)	Low risk	"Sealed envelopes"
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	Difficult to conceal
Blinding of outcome assessment (detection bias) All outcomes	Low risk	Blinded outcome assessment
Incomplete outcome data (attrition bias) All outcomes	Low risk	No missing outcome data
Selective reporting (reporting bias)	Unclear risk	Outcomes were not clearly pre‐specified

PERMALINK

Organised inpatient (stroke unit) care for stroke: network meta‐analysis

Peter Langhorne

Samantha Ramachandra

Abstract

Background

Objectives

Search methods

Selection criteria

Data collection and analysis

Main results

Authors' conclusions

Plain language summary

Summary of findings

Background

Description of the condition

Description of the intervention

Why it is important to do this review

Objectives

Methods

Criteria for considering studies for this review

Types of studies

Types of participants

Types of interventions

Types of outcome measures

Primary outcomes

Secondary outcomes

Search methods for identification of studies

Electronic searches

Searching other resources

Data collection and analysis

Selection of studies

Data extraction and management

Assessment of risk of bias in included studies

Measures of treatment effect

Unit of analysis issues

Dealing with missing data

Assessment of heterogeneity

Assessment of reporting biases

Data synthesis

Pairwise comparisons

Network meta‐analysis

GRADE and 'Summary of findings'

Subgroup analysis and investigation of heterogeneity

Sensitivity analysis

Results

Description of studies

Results of the search

1.

Included studies

Service characteristics within organised (stroke unit) care and conventional care settings

1. Typical characteristics of different models of organised stroke care.

2. Service comparisons in standard analyses.

Excluded studies

Risk of bias in included studies

2.

3.

Allocation

Blinding

Incomplete outcome data

Selective reporting

Other potential sources of bias

Effects of interventions

Summary of findings 1. Organised inpatient (stroke unit) care versus alternative service.

Summary of findings 2. Stroke ward versus general medical ward.

Summary of findings 3. Mobile stroke team versus general medical ward.

Summary of findings 4. Mixed rehabilitation ward versus general medical ward.

Pairwise comparisons

Section 1. Organised inpatient (stroke unit) care versus alternative service (Comparison 1)

Outcome 1.1. Poor outcome by the end of scheduled follow‐up

1.1. Analysis.

Outcome 1.2. Death by the end of scheduled follow‐up

1.2. Analysis.

Outcome 1.3. Death or institutional care by the end of scheduled follow‐up

1.3. Analysis.

Outcome 1.4. Death or dependency by the end of scheduled follow‐up

1.4. Analysis.

Outcomes 1.5 and 1.6. Length of stay (days) in a hospital or institution or both

1.5. Analysis.

1.6. Analysis.

3. Service comparisons in network meta‐analyses (NMAs)^a.

4. Inconsistency table for the network meta‐analysis: poor outcome at the end of scheduled follow‐up^a.

5. Inconsistency table for the network meta‐analysis: death at the end of scheduled follow‐up^a.

6. Inconsistency table for the network meta‐analysis: death or institutional care at the end of scheduled follow‐up^a.

7. Inconsistency table for the network meta‐analysis: death or dependency at the end of scheduled follow‐up^a.

*Study characteristics*
Methods	RCT Telephone randomisation and blinded follow‐up
Participants	People with acute stroke within 5 days of symptoms No recent myocardial infarction or fracture
Interventions	Mobile stroke team (stroke physician, therapist) in 2 acute hospitals provided early assessment and advice to staff, co‐ordinated early therapy input, encouraged guideline adherence Controls received usual medical ward‐based care
Outcomes	Death, institutional care, dependency, simple questions, Nottingham extended ADL score, Frenchay Aphasia Screening Test, EuroQol, Hospital Anxiety and Depression Scale Recorded up to 12 months
Notes	5 intervention and 4 control missing from final follow‐up 23 people underwent secondary randomisation in trial of early supported discharge team
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"Offsite office using a computer generated schedule"
Allocation concealment (selection bias)	Low risk	"Allocated using a simple computer generated procedure ... initially and then in the later stages a minimisation procedure"
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	Difficult to conceal
Blinding of outcome assessment (detection bias) All outcomes	Low risk	Blinded outcome assessment
Incomplete outcome data (attrition bias) All outcomes	Low risk	Proportionately small; similar numbers missing from intervention and control groups at 12 months
Selective reporting (reporting bias)	Low risk	All pre‐specified outcomes reported

*Study characteristics*
Methods	RCT  Blinded assessment of outcome
Participants	Unselected people with stroke within 7 days of stroke onset
Interventions	Mobile stroke team (dedicated stroke unit) (n = 65) vs conventional care on general medical ward (n = 65) Study ended at 6 weeks post stroke
Outcomes	Death, Barthel Index, place of residence, length of initial hospital stay up to 6 weeks after stroke
Notes	Short follow‐up period 1 intervention and 3 control patients lost to follow‐up
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Patients were stratified ... "block randomization within each stratum"
Allocation concealment (selection bias)	Low risk	"Two series of numbered sealed envelopes"
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	Difficult to conceal
Blinding of outcome assessment (detection bias) All outcomes	Low risk	Blinded outcome assessment
Incomplete outcome data (attrition bias) All outcomes	Low risk	3 participants (1 intervention; 2 control) removed from study due to non‐stroke diagnosis following randomisation 1 additional participant not admitted from the emergency room
Selective reporting (reporting bias)	Unclear risk	Not all pre‐specified outcomes reported

*Study characteristics*
Methods	RCT: prospective, single‐blind, randomised controlled trial across 4 hospitals
Participants	People with stroke (ischaemic and haemorrhagic) on day 1 of admission if vacant beds existed in both acute and rehabilitation hospitals and person had an acute stroke within previous 24 to 48 hours with sufficient neurological impairment and disability to require ongoing rehabilitation
Interventions	The 'intervention' was the 'early' commencement of a rehabilitation process, with a significant portion of acute care to be spent in a rehabilitation setting Traditional stroke care (TSC): participants were admitted into an acute stroke unit (ASU) and were transferred to a rehabilitation unit at the end of their acute stroke phase (after completion of investigations and acute treatment, and when medically stable as per usual practice). Therefore participants were cared for in 2 different stages (acute and rehabilitation) by different nursing and allied health teams Comprehensive stroke care (CSC): participants were pressed to be transferred to a rehabilitation bed, aiming within 24 to 48 hours after arrival at ASU (or the next working day if weekend). This occurred when participants were still in acute stroke phase and might require attention to acute medical problems should they arise. Hence it was not equivalent to early transfer to a rehabilitation unit. Participants in the CSC arm were cared for by the same nursing and allied health team for a larger portion of their hospital stay All standard and best possible care was given to participants in both arms and the same treatment interventions were available to participants admitted to either arm (with the exception of rehabilitation process allowed to happen earlier after faster arrival in rehabilitation setting in the CSC arm)
Outcomes	Death FIM at discharge and at 3 months Modified Rankin score at discharge FIM efficiency (11) (change in FIM score ÷ total LOS) between participants who received CSC and those who received TSC. The FIM efficiency is an indicator of the rate of functional improvement per day of hospital stay Total hospital length of stay (acute and rehabilitation units combined) All FIM assessments were performed by the same research officer who had passed training sessions in performing the assessment tool (i.e. 1 research officer for each pair of acute rehabilitation units/hospitals) Length of stay for all participants was decided by the team caring for the participants, and information was extracted from medical records and counter‐checked with the team by research officers upon discharge of participants
Notes	Numbers independent at 3 months calculated as number FIM > 110 calculated from mean and SD
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"Randomization was generated centrally by a biostatistician using a computer software program Stata11 (StataCorp LP, College Station, TX, USA)"
Allocation concealment (selection bias)	Low risk	"The generated results were concealed and stored locally. Clinicians at the rehabilitation services were not informed as to whether a patient was randomized into the study"
Blinding of participants and personnel (performance bias) All outcomes	Low risk	"Clinicians at the rehabilitation services were not informed as to whether a patient was randomized into the study" "All standard and best possible care was given to participants in both arms, and the same treatment interventions were available to patients admitted to either arm"
Blinding of outcome assessment (detection bias) All outcomes	Low risk	"The research officers who conducted baseline measures and subsequent FIM assessments at discharge and 90 days telephone follow‐up interview were also blind to group allocation" "The rehabilitation teams were blind to the group allocation of the patients"
Incomplete outcome data (attrition bias) All outcomes	Low risk	A total of 4/47 (9%) participants withdrew
Selective reporting (reporting bias)	Unclear risk	Unclear

*Study characteristics*
Methods	RCT with 5:4 allocation of intervention:control Blinded assessment of outcome
Participants	Patients with stroke at 2 weeks after stroke onset Able to participate actively in rehabilitation
Interventions	Stroke rehabilitation ward in department of geriatric medicine (n = 176) vs conventional care in geriatric medical (mixed rehabilitation) ward (n = 63) or general medical ward (n = 76) Organised care provided for months if required
Outcomes	Death, Barthel Index, place of residence, Nottingham Health Profile, length of hospital stay up to 1 year after stroke
Notes	Some cross‐over from general medical ward to geriatric medicine department 3 intervention and 4 control participants lost to follow‐up
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"Stratified based on admission ward ... then randomly allocated"
Allocation concealment (selection bias)	Unclear risk	Method of allocation concealment unclear
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	Difficult to conceal
Blinding of outcome assessment (detection bias) All outcomes	Low risk	Blinded outcome assessment
Incomplete outcome data (attrition bias) All outcomes	Unclear risk	Small numbers (3 intervention; 4 control) lost to follow‐up Some secondary outcome assessments not completed or partially completed; this varied between groups
Selective reporting (reporting bias)	Low risk	All pre‐specified outcomes reported

*Study characteristics*
Methods	RCT Subgroup of Nottingham (stroke unit vs general medical ward)
Participants	People with stroke at 2 weeks after stroke onset Able to participate actively in rehabilitation
Interventions	Stroke rehabilitation ward in department of geriatric medicine (n = 78) vs conventional care in geriatric medical (mixed rehabilitation) ward (n = 63) Organised care provided for months if required
Outcomes	Death, Barthel Index, place of residence, Nottingham Health Profile, length of hospital stay up to 1 year after stroke
Notes	Some cross‐over from general medical ward to geriatric medicine department
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"Stratified based on admission ward ... then randomly allocated"
Allocation concealment (selection bias)	Unclear risk	Method of allocation concealment unclear
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	Difficult to conceal
Blinding of outcome assessment (detection bias) All outcomes	Low risk	Blinded outcome assessment
Incomplete outcome data (attrition bias) All outcomes	Low risk	Some secondary outcome assessments not completed or partially completed; this varied between groups
Selective reporting (reporting bias)	Low risk	All pre‐specified outcomes reported

*Study characteristics*
Methods	RCT Subgroup of Orpington 1993 (stroke unit vs general medical ward)
Participants	People who survived stroke for 2 weeks Suitable for transfer to rehabilitation ward
Interventions	Stroke rehabilitation ward (n = 53) vs conventional care in general medical (n = 48) ward Organised care provided for months if required
Outcomes	Death, Barthel Index, place of residence, length of initial hospital stay at end of follow‐up
Notes	Stroke severity subgroup data inferred from distribution in whole group
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"Randomised with the use of Geigy table of random numbers"
Allocation concealment (selection bias)	Low risk	"Randomisation was computerized"
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	Difficult to conceal
Blinding of outcome assessment (detection bias) All outcomes	High risk	Unblinded outcome assessment
Incomplete outcome data (attrition bias) All outcomes	Unclear risk	2 intervention and 5 control participants lost to follow‐up
Selective reporting (reporting bias)	Low risk	All pre‐specified outcomes reported

*Study characteristics*
Methods	RCT Blinded outcome assessment
Participants	People with acute stroke (meeting WHO definition of stroke) from a community stroke register Intermediate stroke severity
Interventions	3‐arm comparison of: comprehensive stroke ward (co‐ordinated multi‐disciplinary team care) (n = 152); general ward with input from hospital mobile stroke team (comprising medical, physiotherapy, occupational therapy, speech therapy but not nursing or medical specialists) (n = 152); and domiciliary multi‐disciplinary stroke team (not relevant to this review)
Outcomes	Death, dependency (Barthel Index), place of residence, length of stay, resource use up to 12 months 3 control participants lost to follow‐up
Notes	‐
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"Unstratified ... using the block randomization technique .... computer generated random numbers"
Allocation concealment (selection bias)	Low risk	"Allocation schedule prepared using computer generated random numbers"
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	Difficult to conceal
Blinding of outcome assessment (detection bias) All outcomes	Low risk	Blinded outcome assessment
Incomplete outcome data (attrition bias) All outcomes	Unclear risk	3 control participants lost to follow‐up at 12 months
Selective reporting (reporting bias)	Low risk	All pre‐specified outcomes reported

*Study characteristics*
Methods	RCT by means of sealed envelopes (stratified by age and side of lesion)
Participants	People with acute stroke patients (within 8 days of symptoms) meeting WHO diagnostic criteria
Interventions	Comprehensive stroke ward (n = 31) vs conventional care in general medical ward (n = 34)
Outcomes	Death, dependency (Rankin score), place of residence, length of hospital stay at 6 months after randomisation
Notes	Staffing levels were higher in the stroke unit group
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Translation ‐ randomised by the envelope method (drawing lots), stratified by age and side of lesion
Allocation concealment (selection bias)	Low risk	Sealed envelopes
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	Difficult to conceal
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	Unclear
Incomplete outcome data (attrition bias) All outcomes	Low risk	No obvious missing outcome data
Selective reporting (reporting bias)	Unclear risk	Unclear

Study	Reason for exclusion
Abissi 1995	Trial tested a care plan protocol only No other aspect of organisation was under evaluation
Akhtar 2015	Not an RCT
Al‐Qahtany 2014	Not an RCT
Asplund 2000	Trial of a geriatric assessment unit
Cavallini 2003	Quasi‐randomised treatment allocation
Davis 2000	Intervention and control arms of trial were treated within the same stroke unit
Di Lauro 2003	Intervention and control arms of trial were treated within the same stroke unit
Diagana 2008	Quasi‐randomised treatment allocation
Durastanti 2005	Quasi‐randomised treatment allocation
Felix 2016	Trial of stroke education
Fu 2006	Not an RCT
HAMLET 2009	Does not report outcomes for different medical treatment arms
Hamrin 1982	Quasi‐randomised treatment allocation
He 2014	Not a fully randomised cluster‐RCT
Inoue 2013	Not an RCT
Janssen 2014	Trial of enriched environment
Koton 2005	Treatment allocated by selection criteria
Langhorne 2001	Study tested a care plan protocol only No other aspect of organisation was under evaluation
Middleton 2006	Care pathway study only
Middleton 2018	Care pathway study
Moloney 1999	Care pathway study only
Pappa 2009	Non‐randomised
Patel 2000	Quasi‐randomised treatment allocation
Pearson 1988	No available outcome data
Rai 2016	Not a full RCT
Raiborirug 2017	Not an RCT
Ricauda 2004	Trial comparing home care team vs general medical ward
Ronning 1998	Quasi‐randomised treatment allocation
Ronning 1998a	A portion of the data was collected retrospectively All prospective data are included in the Akershus study (Ronning 1998)
Ronning 1998b	Comparison of stroke rehabilitation ward vs discharge to community‐based stroke rehabilitation
Shiraishi 2004	Non‐randomised treatment allocation
Silva 2004	Treatment allocated by the study neurologist
Stone 1998	No available outcome data
Strand 1985	Quasi‐randomised treatment allocation
Von Arbin 1980	Quasi‐randomised treatment allocation
Walter 2005	Non‐randomised treatment allocation
Wang 2004	No available outcome data
Yagura 2005	Quasi‐randomised treatment allocation

Methods	RCT
Participants	People with acute stroke
Interventions	"Standardised tertiary rehabilitation" (n = 51) vs usual inpatient care (n = 51)
Outcomes	Functional outcome (unknown scale) and quality of life (WHOQOL‐BREF) at 1, 3, and 6 months Cost analysis
Notes	Currently no useable data