Intervention services

1. Early supported discharge (ESD) team co‐ordination and delivery: in nine trials the ESD service comprised a multidisciplinary team which co‐ordinated discharge from hospital, post‐discharge care and provided rehabilitation and patient care at home or in a community setting (Adelaide 2000; Aveiro 2016; Belfast 2004; Copenhagen 2009; London 1997; Manchester 2001; Montreal 2000; Newcastle 1997; Stockholm 1998). The multidisciplinary team met on a regular basis to plan patient care.

2. ESD team co‐ordination: in four trials discharge home and the immediate post‐discharge care was planned and supervised by a co‐ordinated multidisciplinary team (Bergen 2014; Oslo 2000; Trondheim 2000; Trondheim 2004). However, care was subsequently handed over to existing community‐based agencies who provided continuing rehabilitation and support at home. These community‐based agencies did not always provide co‐ordinated multidisciplinary team care (i.e. input from a multidisciplinary team which met on a regular basis to plan patient care). However, in some trials the community teams were also multidisciplinary in nature and focused on working with stroke patients early after discharge (Bergen 2014; Oslo 2000). One recent trial randomised patients to one of two different forms of ESD service (Bergen 2014); based in a community day unit (Bergen 2014 ‐ Day unit) or in their homes with home‐visits from the community health team (Bergen 2014 ‐ Home care). Although both met the definition of an ESD service they have, where possible, been analysed separately to reflect this difference in design.

3. No ESD team: in four trials, patients had access to multidisciplinary team care in hospital, but this ended at hospital discharge (Adelaide 2016; Akershus 1998; ATTEND pilot 2015; Bangkok 2002). Their subsequent care was provided by a range of community stroke services which were; not planned or provided by a co‐ordinated team (Akershus 1998), were provided by trained healthcare volunteers (Bangkok 2002), or provided through supported training by a physiotherapist for patients and family (Adelaide 2016; ATTEND pilot 2015).

The boundary between groups 1 and 2 does not appear clear cut but indicates a spectrum of approaches where an ESD team plans and co‐ordinates discharge, provides early post‐discharge rehabilitation, and then hands over care to other community services.

ESD team structure, practices and procedures

Details of ESD team practices can best be obtained from the original trials. However, we previously developed a summary description of the services to indicate the type of service provided. From recorded staff contact time, we calculated standardised staffing levels (whole time equivalents (WTE) sufficient to manage a notional 100 new patients per year) (Adelaide 2000; Aveiro 2016; London 1997; Montreal 2000; Newcastle 1997; Stockholm 1998), or a typical team caseload (Belfast 2004; Trondheim 2000; Trondheim 2004). We assumed staff would have a 35‐hour working week with 20 hours direct contact time and 10 hours indirect contact time.

Typical ESD teams had approximately 3.1 WTE staff (range 2.6 to 4.6) as follows; medical 0.1, nursing (ranged from 0 to 1.2), physiotherapy 1.0, occupational therapy 1.0, speech and language therapy 0.3, assistant 0.4. Variable levels of social work (0 to 0.5 WTE) and secretarial support were also available (Table 1).

The ESD teams could either have a community (community in‐reach) or hospital base (hospital out‐reach) with experience in stroke rehabilitation/neurological rehabilitation (Adelaide 2000; Aveiro 2016; Belfast 2004; Bergen 2014; Copenhagen 2009; London 1997; Manchester 2001; Montreal 2000; Newcastle 1997; Oslo 2000; Stockholm 1998; Trondheim 2000; Trondheim 2004). All co‐ordinated their work through regular multidisciplinary team meetings. A typical approach would involve the early identification of the patient in hospital and a visit from the key worker (case manager) from the ESD team. Discharge was planned with the patient and carer, often involving a pre‐discharge home visit (attended by the patient) or environmental visit (not attended by the patient). Team input typically began on the day of discharge and could be provided as required. In practice this ranged from daily input to four to five days per week. Typically teams would agree recovery goals with the patient and negotiate the termination of services within three months (which would be tapered off as goals were achieved). Many teams used a patient‐held medical record and provided a formal discharge summary at the end of input.

Control services

These were categorised on whether organised stroke unit care was available to patients prior to discharge (Table 1). In 12 trials, all patients were recruited from a stroke unit or neurological rehabilitation unit staffed by a multidisciplinary team (Adelaide 2000; Adelaide 2016; Akershus 1998; ATTEND pilot 2015; Aveiro 2016; Bergen 2014; Copenhagen 2009; Oslo 2000; Stockholm 1998; Trondheim 2000; Trondheim 2004) or most patients (Belfast 2004). Five trials recruited a minority of patients from a multidisciplinary stroke unit setting (Bangkok 2002; London 1997; Manchester 2001; Montreal 2000; Newcastle 1997). Therefore, the control service was frequently provided in general wards. Discharge arrangements were variable in the control services with a minority undergoing a pre‐discharge home visit and variable follow‐up arrangements.

Settings of services

The trials identified come from nine countries (Australia, Canada, Denmark, India, Norway, Portugal, Sweden, Thailand, UK). Fourteen trials were established in city hospitals servicing largely urban areas while two (Aveiro 2016; Belfast 2004) covered a mixture of rural and urban areas. An additional trial recruited only patients from rural addresses who were admitted to a large urban hospital (Trondheim 2004).

Patient characteristics

Patients had a clinical diagnosis of stroke and the average patient age in the trials ranged from 60 to 80 years. There appeared to be a degree of selection of patients deemed suitable for the ESD services that was based on need (persisting disability), stability of their medical condition, and practicability (living within the local area). The average (mean or median) initial Barthel index (at the time of patient recruitment) in each study ranged from 10/20 to 19/20 with a lower IQR limit of 6 to 16/20 and an upper value of 14 to 19/20. Thus the typical patient population had an initial Barthel index of 14/20 with an IQR of 10 to 18.

We repeated this process to estimate the Barthel index (or equivalent score) at the time of discharge for those trials where the ADL score was recorded within one week prior to discharge (Adelaide 2000; Aveiro 2016; Belfast 2004; Bergen 2014; London 1997; Manchester 2001; Newcastle 1997; Trondheim 2000; Trondheim 2004). The average (mean or median) initial Barthel index (within one week prior to discharge) in each study ranged from 13/20 to 19/20 with a lower IQR limit of 10/20 to 16/20 and an upper value of 15/20 to 19/20. Thus the typical patient population prior to discharge had an initial Barthel index of 15/20 with an IQR of 11/20 to 17/20.

None of the trials recruited more than 70% of hospitalised stroke patients; a median of 33% (range 13% to 70%) of hospitalised stroke patients met the clinical criteria for the early discharge service (NB: in some trials, a further group of patients did not meet research criteria such as an ability to complete research assessments). We have summarised the inclusion and exclusion criteria of the individual trials in the Characteristics of included studies table.

Outcomes

Most trials included our main outcomes of death, residence (institutional care) and dependency (Barthel index, Rankin score or Functional Independence Measure), all recorded at the end of scheduled follow‐up, as well as our primary resource outcome length of initial hospital stay (Table 2). Missing data for the primary outcome are summarised in Table 3. Two trials subsequently reported further outcomes of death and dependency after scheduled follow‐up (at one year and five years) (Stockholm 1998; Trondheim 2000).

Secondary outcomes included a range of measures, which are summarised in the Characteristics of included studies table and the sampling analysis schedule provided in Table 4 and Table 5.

1.1: Death

Outcome data were available for 16 trials (2116 participants). We assumed participants with missing data (57 intervention participants and 53 controls) were alive. Overall, there was no significant difference in case‐fatality between the ESD team and conventional services (odds ratio (OR) 1.04, 95% confidence interval (CI) 0.77 to 1.40, P = 0.81, moderate‐grade evidence). There was no significant degree of statistical heterogeneity but a statistical interaction (P = 0.01) between subgroups suggesting a higher case fatality in the subgroup without a co‐ordinated ESD team (Analysis 1.1).

1.2: Death or requiring institutional care

Outcome data were available for 12 trials (1664 participants). We assumed participants with missing data (24 intervention participants and 19 controls) were alive and living at home. Overall, there was a significant reduction in the odds of patients dying or requiring long‐term institutional care (OR 0.75, 95% CI 0.59 to 0.96, P = 0.02, moderate‐grade evidence) with no significant heterogeneity. This equates to an extra five (one to eight) patients living at home for every 100 treated (Analysis 1.2).

1.3: Death or dependency

Outcome data were available for 16 trials (2359 participants). We assumed participants with missing data (99 intervention participants and 84 controls) were alive and independent. Overall, there was a significant reduction in the odds of the combined adverse outcome of death or dependency (OR 0.80, 95% CI 0.67 to 0.95, P = 0.01, moderate‐grade evidence) with no significant heterogeneity. This equates to an extra five (one to nine) patients regaining independence for every 100 receiving ESD services. There was no substantial degree of statistical heterogeneity (Analysis 1.3).

1.4: Activities of daily living (ADL)

These data were available (in a variety of formats) for 12 trials (1449 participants). Overall, there was no apparent difference in the ADL scores of survivors for whom data were available with no significant heterogeneity (Analysis 1.4).

1.5: Extended activities of daily living

These data were available (in a variety of formats) for 11 trials (1262 participants). Overall, there was an apparent increase in extended ADL scores among survivors receiving ESD services (standardised mean difference (SMD) 0.14, 95% CI 0.03 to 0.25, P = 0.01, low‐grade evidence). These results were largely dependent on data from the two subgroups of trials evaluating an ESD team (Analysis 1.5).

1.6: Subjective health status

These data were available (in a variety of formats) from 11 trials (1202 participants). Overall, there was no apparent difference in the subjective health status scores of both groups. There was no significant degree of heterogeneity (Analysis 1.6).

1.7: Mood status

These data were available (in a variety of formats) from nine trials (915 participants). Overall, there was no apparent difference in mood scores. There was no significant heterogeneity. Additional dichotomous data from one trial (London 1997) indicated that those people in the ESD service group were more likely to express anxiety (P = 0.02) and non‐significant trends towards higher levels of depression (Analysis 1.7).

1.8: Patient satisfaction

These data were available (in a variety of formats) from five trials (513 participants). Overall, there was a pattern of ESD service patients being more likely to report satisfaction with outpatient services or services in general (OR 1.60, 95% CI 1.08 to 2.38, P = 0.02, low‐grade evidence). There was no significant heterogeneity (Analysis 1.8).

3.1: Subjective health status

These data were available (in a variety of formats) from nine trials (813 carers). Overall, there was no apparent difference in scores and no significant heterogeneity (Analysis 3.1).

3.2: Mood status

These data were available from only three trials with 122 carers. Overall, there was no apparent reduction in the mood score of carers receiving ESD services, but significant heterogeneity was apparent between trials (Analysis 3.2).

3.3: Carer satisfaction

These data were available (in a variety of formats) from four trials (279 carers). Overall, there was no convincing difference in the odds of carers who received ESD services expressing satisfaction with services (OR 1.56, 95% CI 0.87 to 2.81) (Analysis 3.3).

4.1: Length of initial hospital stay

We were able to analyse data on length of initial hospital stay (using the longest available of acute care and rehabilitation for the index admission) for 16 trials (2161 patients) (Analysis 4.1). Across all trials, there was a reduction in the length of hospital stay (MD ‐ 5.5 days, 95% CI 2.9 to 8.2, P < 0.0001, moderate‐grade evidence), which is approximately equivalent to five days. There was, however, considerable heterogeneity, which reduces confidence in the estimates.

Data were incomplete for total length of stay including hospital readmissions. An analysis of the pattern of discharges based on six trials that could provide data (Adelaide 2000, Belfast 2004, London 1997, Manchester 2001, Oslo 2000, Stockholm 1998) is shown in Table 6.

4.2: Hospital readmissions

Seven trials (784 participants) provided data on the number of participants readmitted to hospital after the index admission. Readmission rates during scheduled follow‐up (27% versus 25%) were very similar (OR 1.09, 95% CI 0.79 to 1.51, P = 0.59, low‐grade evidence) between the ESD service and conventional care groups (Analysis 4.2).

Costs

Costing data are currently available from seven trials (Table 7), which estimated total costs up to three months (Montreal 2000), six months (Adelaide 2000; Newcastle 1997) or one year (London 1997; Stockholm 1998; Trondheim 2000) after randomisation. Estimated costs ranged from 23% less to 15% greater for the ESD group in comparison to controls. These estimates were reported to be stable in sensitivity analyses.

Analyses by methodological characteristics

Analysis of the primary outcome restricted to the 12 trials that reported concealed randomisation and blinded follow‐up showed a convincing reduction in death or dependency (OR 0.75, 95% CI 0.62 to 0.92, P = 0.005) with no heterogeneity (Adelaide 2000; ATTEND pilot 2015; Belfast 2004; Bergen 2014; Copenhagen 2009; London 1997; Manchester 2001; Montreal 2000; Oslo 2000; Stockholm 1998; Trondheim 2000; Trondheim 2004). Analysis restricted to the 10 trials that reported concealed randomisation and blinded follow‐up plus a very high rate of patient follow‐up for the primary outcome (1277/1318; 3.1% participants missing) showed similar results (OR 0.75, 95% CI 0.60 to 0.93, P = 0.01) with no heterogeneity (Adelaide 2000; ATTEND pilot 2015; Belfast 2004; London 1997; Manchester 2001; Montreal 2000; Oslo 2000; Stockholm 1998; Trondheim 2000; Trondheim 2004).

For the primary outcome of death or dependency, data were missing for 99 (8.0%) intervention participants and 84 controls (7.5%). Our primary analysis assumed they were alive and independent (OR 0.80, 95% CI 0.67 to 0.95) (Analysis 1.3). The result would be similar if all missing participants were assumed to be dead or dependent (OR 0.82, 0.70 to 0.97). The confidence intervals around the apparent effect of ESD services would only cross unity if there was a substantial imbalance in missing data outcomes favouring control services.

Analyses by participant age and gender

Subgroup data for the primary outcome (death or dependency) were available for at least nine trials. Smaller amounts of data were available for death, death or institutionalisation, and length of stay. There was no significant association of participant age or gender with the apparent effect of the ESD service (Analysis 5.1; Analysis 5.2; Analysis 6.1; Analysis 6.2).

Analyses by initial stroke severity

Data were available for 11 trials (1545 participants). Subgroup analysis by initial stroke severity revealed a differential effect in the odds of death or dependency between participants with moderate initial stroke severity (initial Barthel Index of > 9/20) and those in the severe subgroup (initial Barthel Index < 10/20). In the moderate subgroup there was a reduction (OR 0.77, 95% CI 0.61 to 0.98) as opposed to a non‐significant increase in the severe subgroup (OR 1.40, 95% CI 0.83 to 2.36); test for subgroup interaction P = 0.04. Similar patterns of results were seen for the outcome death or institutional care. The reduction in length of hospital stay was much greater (P < 0.0001) for the severe stroke subgroup (MD 28 days, 95% CI 17 to 40) than the moderate group (MD 3 days, 95% CI 1 to 7). Similar results were obtained if the Barthel index at randomisation was used from the two trials that randomised patients up to several weeks after stroke (Adelaide 2000; London 1997: Analysis 7.1; Analysis 7.2). Similar results were obtained if the three most recent trials for which we do not have individual patient data were included but analysed according to the mean Barthel index at randomisation (Adelaide 2016; Aveiro 2016; Bergen 2014).

These results suggest that the greatest benefit in clinical outcomes was with the mild and moderate groups but the greatest reduction in hospital bed days was with the severe subgroup.

Analyses by carer availability

Eleven trials (1341 participants) could provide subgroup data on the availability of a carer. There was no apparent interaction of ESD service effect with the presence of a carer (Analysis 8.1; Analysis 8.2).

Analyses by control service organisation

Subgroup analyses were carried out according to the background (control) service available; stroke unit or other ward. There were no apparent interactions with control service characteristics (Analysis 9.1; Analysis 9.2).

Analyses by ESD service organisation

There was no significant interaction with the background service (stroke unit or other ward) or the base for the ESD team (community in‐reach or hospital out‐reach). The reduction in length of hospital stay was slightly greater in the hospital out‐reach group (MD 5 days, 95% CI 1 to 9) than the community in‐reach group (MD 4 days, 95% CI 1 to 7) but this was not statistically significant (P = 0.74) (Analysis 10.1; Analysis 10.2; Analysis 11.3; Analysis 11.4; Analysis 9.1; Analysis 9.2).

The ESD services studied were classified according to the organisation of the multidisciplinary team (see Description of studies). Using this classification, there was a potential subgroup interaction (P = 0.06) by ESD characteristics. The trials with a co‐ordinated multidisciplinary ESD team (Analysis 11.3) showed an odds of death or dependency of OR 0.73 (95% CI 0.60 to 0.89) compared with OR 1.11 (95% CI 0.75 to 1.62) in those without an ESD team.

The staffing levels of each service did not differ sufficiently to allow meaningful subgroup analyses based on staff mix, service intensity, and supportive versus rehabilitative interventions.

Analysis of 'core' ESD services

Some commentators have criticised the original inclusion of trials that did not incorporate a robust multidisciplinary rehabilitation programme in the community (Akershus 1998; Adelaide 2016; ATTEND pilot 2015; Bangkok 2002). The remaining 13 trials are much more typical of what has become accepted as a 'core' ESD service (Fisher 2011). If the analyses are restricted to those 13 trials the results are more convincing for ESD services: death (OR 0.78, 95% CI 0.54 to 1.11; P = 0.17; Analysis 11.1), death or institutional care (OR 0.65, 95% CI 0.49 to 0.87; P = 0.003; Analysis 11.2), death or dependency (OR 0.73, 95% CI 0.60 to 0.89; P = 0.002; Analysis 11.3) and reduction in length of stay (MD 6 days; 95% CI 3 to 9; P < 0.0001; Analysis 11.4).