Stroke Units in Their Natural Habitat
Systematic Review of Observational Studies
Background and Purpose— Within clinical trials, stroke patients allocated to receive organized inpatient (stroke unit) care are more likely to survive, return home, and regain independence than those allocated to conventional care. However, there are concerns that the benefits seen in clinical trials may not be replicated in routine practice. We carried out a systematic review of observational studies of stroke unit implementation.
Methods— We searched (up to January 2006) MEDLINE, EMBASE, CINAHL, Cochrane Library, British Nursing Index, Cochrane Stroke Group register, and recent conference abstracts for observational studies that compared the outcomes of stroke patients managed in a stroke unit versus non–stroke unit care. We excluded studies that did not describe either matching for baseline prognostic factors or adjustment for case-mix characteristics. The primary outcome was death within 1 year. We also recorded poor outcome (death, institutional care, or dependency). Data analysis used the generic inverse variance method in Revman 4.2. Where raw data were provided, effect sizes and variances were calculated accordingly. We used a random-effects model and explored for sources of heterogeneity.
Results— We identified 72 articles describing stroke unit outcomes; 25 were eligible for review; and 18 provided data on case fatality or poor outcome. Stroke unit care was associated with significantly reduced odds of death (odds ratio=0.79, 95% CI=0.73 to 0.86; P<0.00001) and of death or poor outcome (odds ratio=0.87, 95% CI=0.80 to 0.95; P=0.002) within 1 year of stroke. Results were complicated by significant heterogeneity (P<0.05), mainly in single-center studies.
Conclusions— Although these results are complicated by potential bias and heterogeneity, the observed benefit associated with stroke unit care in routine practice is comparable to that in clinical trials.
Organized inpatient (stroke unit) care has been shown to be effective in randomized trials, wherein patients who were managed within a stroke unit were more likely to survive, return home, and regain independence.1 However, commentators have pointed out that stroke unit care is a complex intervention,2 and there may be difficulties implementing stroke unit care in routine practice.3 First, stroke unit care has many facets that may be difficult to reproduce. Second, the conditions under which randomized trials are carried out are unusual and may not be easily replicated in routine practice. Given that the establishment of stroke units has become an important healthcare policy, it is important that we confirm their impact in routine practice.
A number of observational studies have addressed the question of whether the expected beneficial effects of stroke unit care can be observed in routine practice. These have produced apparently mixed results; some investigators have been unable to demonstrate successful implementation of the benefits of stroke unit care,3 whereas other studies have concluded that stroke unit care was either effective4 or effective but to a lesser degree than in the randomized trials.5
In view of these varying reports, we carried out a systematic review of observational studies that have measured the effects of stroke unit care in routine clinical settings. In doing so, we had to be aware of potential confounding factors within this type of observational study,6 in particular, the potential impact of patient case mix.
Our primary hypothesis was that in routine clinical settings, admission to a stroke unit would be associated with improved clinical outcomes even after accounting for potential confounding factors such as patient age and stroke severity. Our secondary hypothesis was that the size of this estimate would vary with the methodological quality of the study. We implemented our systematic review according to MOOSE guidelines for meta-analysis of observational studies.6
Type of Study
We included any observational study that allowed a comparison of care in a stroke unit (or units) with non–stroke unit care. This could include a cross-sectional survey, a controlled before-and-after study, a controlled (parallel-group) comparison, or a case-control design. We aimed to include studies that had either clearly demonstrated good matching of prognostic factors at baseline or had carried out an adequate case-mix adjustment during their analysis. The setting of the study (eg, multicenter or single center) and the type of patient matching and case-mix adjustment were recorded but not used as exclusion criteria.
Type of Intervention
We used a very broad definition of organized inpatient (stroke unit) care: namely, an area of a hospital dedicated to stroke patient care that was provided by a multidisciplinary team of stroke specialists. Where possible, we also recorded the type of stroke unit care according to the definitions previously described1 and the setting in which the stroke unit service was provided.
Type of Comparison
We aimed to identify studies in which the comparator group was the absence of stroke unit care. We therefore excluded studies that compared 2 different types of stroke unit care. The type of comparator service was recorded but not used as an exclusion criterion.
Type of Participants
We included studies that had recruited patients with a clinical diagnosis of stroke. In instances where the stroke unit service had recruited a mixed patient group, at least 80% had to have a diagnosis of stroke.
Type of Outcomes
The primary outcome was death, but we also recorded poor outcome (death, failure to be discharged home, or failure to regain independence in daily activities). Our primary aim was to identify outcomes at the end of the scheduled follow-up (within 1 year), but we also anticipated some longer-term follow-up studies (>1 year).
We used a range of approaches to identify appropriate studies. First, we undertook searches of computerized databases by using search strategies developed by an information specialist (see Appendix) using the key words “stroke unit,” “outcome,” “stroke outcome,” and “stroke service.” The databases searched included MEDLINE, EMBASE, CINHAL, the Cochrane Library, and the British Nursing Index. We also checked the reference lists of included articles. No language restriction was used. In addition, we searched the Cochrane Collaboration Stroke Group Trials Register. Finally, we reviewed recent conference abstracts, including the meetings of the American Stroke Association, European Stroke Congress, and British Geriatrics Society from 2000 to 2005. The searching was completed in January 2006. Two independent reviewers screened titles and abstracts to decide on study eligibility and extract data from the included studies. Any disagreements were resolved through discussion.
Data analysis used case-mix–adjusted outcome data (usually expressed as odds ratios and 95% CIs). In instances where raw data were provided, the effect size and variance were calculated accordingly. The statistical analysis used the generic inverse variance approach to combine log odds ratios and standard errors. We used a random-effects model and explored for sources of inconsistency (I2) and heterogeneity. Revman 4.2 software was used in this analysis.
All available data were included in the primary analysis. Subsequent subgroup analyses were planned according to the type of study (multicenter versus single center), degree of case-mix adjustment, type of article (fully published versus abstract only), type of stroke unit, different levels of stroke severity, and different stroke types (infarct or hemorrhage).
The review profile is shown in Figure 1. A total of 25 reports were included, of which 7 did not present any useable outcome data (Table 1⇓). Of the 25, 18 were from northern Europe, 3 from Mediterranean countries, and 1 each from North America, Central Europe, and Asia. The 18 reports with useable data included 12 reports (10 224 participants) of 10 single-center studies (before-and-after or controlled cohort design) and 6 reports (32 012 participants) of 6 multicenter cross-sectional studies. The 7 reports without useable data all reported trends or significant improvements in outcome associated with stroke unit care (Table 1⇓).
The main analysis by death recorded within 1 year of follow-up is shown in Figure 2. Stroke unit care was associated with an odds ratio for death of 0.79 (95% CI=0.73 to 0.86, P<0.00001) with substantial inconsistency (I2=45.5%) and significant heterogeneity (P=0.02). The main source of heterogeneity appeared to be within single-center studies. Focusing on multicenter studies alone produced a similar result (odds of death=0.82, 95% CI=0.77 to 0.87, P<0.00001) without significant inconsistency (I2=0%) or heterogeneity (P=0.46). Death recorded after >1 year of follow-up was available for 4 studies9,10,14,22 and showed an odds ratio of 0.82 (95% CI=0.69 to 0.97, P=0.02) with substantial inconsistency (I2=63.7%) and significant heterogeneity (P=0.04).
The main analysis for the number of patients failing to return home or failing to regain independence is shown in Figure 3. Those managed in a stroke unit showed a reduced odds of death or poor outcome (odds ratio=0.87, 95% CI=0.80 to 0.95, P=0.002). Once again, there was inconsistency (I2=55%) and significant heterogeneity (P=0.005), which was more evident in single-center studies.
Of the 6 planned subgroup analyses, 2 (effect of different types of stroke unit; infarct versus hemorrhage) were not possible because of a lack of information. The remainder are shown in Table 2. The only statistically significant subgroup effect was with publication type, for which full journal publications had a smaller effect size than did those published in abstract form only. Within all subgroups, there was a statistically significant reduction in the odds of death associated with stroke unit care. The improved survival with stroke unit care was still present when the analysis was restricted to full publications of multicenter studies,5,19–23 for which the odds ratio was 0.82 (95%=CI 0.77 to 0.87, P<0.00001).
The potential therapeutic benefit of organized (stroke unit) care has been recognized for several years.1 Within clinical trials, patients who were managed within a stroke unit were more likely to survive, return home, and regain independence in everyday activities.1 However, implementing a complex intervention such as stroke unit care is not a straightforward process, and there has always been a concern that without the particular circumstances of a clinical trial, stroke unit care may be less effective. For instance, under the conditions of a clinical trial, one might expect to have an expert and motivated multidisciplinary team staffing the stroke unit, which may not be easily replicated in routine practice. Alternatively, the establishment of a new stroke unit for the purposes of a clinical trial may strip expertise from conventional-care settings.
This current review confirms that in observational studies comparing outcomes of stroke patients managed in the stroke unit as opposed to a non–stroke unit setting, stroke unit care was associated with reduced odds of death and reduced odds of poor outcome (death, institutional care, or dependency). The results were complicated by significant heterogeneity but remained statistically significant, even when restricted to the most conservative estimates from full-publication reports of multicenter studies. Furthermore, the effect size for reducing the odds of death (odds ratio=0.79) is remarkably close to that of the clinical trials.1 Overall, despite the potential problems in the available data, these results appear to corroborate those of the clinical trials.
The main strength of our analysis is that we focused on observational studies in routine clinical practice. The results are likely to reflect routine clinical care, particularly in those multicenter studies that covered a large number of hospital sites. Our systematic review was carried out according to recognized guidelines,6 and we went to considerable length to identify reports, regardless of their publication characteristics. We had a preplanned approached in exploring for sources of heterogeneity and used recognized statistical methods to handle such data.
The main weaknesses in our study concern both the imprecision around the definition of stroke unit care and the very real possibility of publication bias with observational studies. Publication bias is likely to be an important factor, particularly in single-center studies, and this possibility is supported by the observed subgroup effect around publication type (Table 2), such that reports published in abstract form only had a significantly different effect size from those reported as full publications. However, even when we focused only on full-publication reports of those studies least likely to be subject to publication bias (multicenter studies), we would still conclude that admission to a stroke unit is associated with improved clinical outcomes.
A number of issues remain unresolved. Previous multicenter studies5 have suggested that the benefits of routine stroke unit care may not apply to those with more severe stroke (reduced consciousness on admission). In this review, we had limited information to explore such interactions, although the information available suggested that the benefit of stroke unit care in reducing case fatality was similar between those with lesser or greater stroke severity. Similarly, we did not have sufficient information to explore whether stroke units benefited those with different stroke types. It is notable that within the clinical trials, stroke unit care appeared to benefit a broad range of stroke patients.1
In summary, the observed benefit associated with stroke unit care in observational studies of routine practice is comparable to that observed in clinical trials. These observations cannot be explained by the potential methodological limitations of the studies. Future emphasis should be placed on defining more exactly the important characteristics of stroke unit care and ensuring that they are implemented widely.
We are very grateful to A. Wright and L. Garrity for support with literature searching. The title of this article was inspired by the original RIKS stroke publication.4
- Received December 20, 2006.
- Accepted January 9, 2007.
Stroke Unit Trialists’ Collaboration. Organised inpatient (stroke unit) care for stroke. Cochrane Database of Systematic Reviews 2001, issue 3. article No. CD000197. DOI: 10.1002/14651858.CD000197.
Campbell M, Fitzpatrick R, Haines A, Kinmonth AL, Sandercock P, Spiegelhalter D, Tyrer P. Framework for design and evaluation of complex interventions to improve health. BMJ. 2000; 321: 694–696.
Gompertz P, Pound P, Briffa J, Ebrahim S. How useful are non-random comparisons of outcomes and quality of care in purchasing hospital services? Age Ageing. 1995; 24: 137–141.
Stegmayr B, Asplund K, Hulter-Asberg K, Norrving B, Peltonen M, Terent A, Wester PO. Stroke units in their natural habitat: can results of randomized trials be reproduced in routine clinical practice? Stroke. 1999; 30: 709–714.
Stroup DF, Berlin JA, Morton SC, Olkin I, Williamson GD, Rennie D, Moher D, Becker BJ, Sipe TA, Thacker SB, for the meta-analysis of observational studies in epidemiology (MOOSE) group. Meta-analysis of Observational Studies in Epidemiology: a proposal for reporting. JAMA. 2000; 283: 2008–2012.
Davenport RJ, Dennis MS, Warlow CP. Effect of correcting outcome data for case mix: an example from stroke medicine. BMJ. 1996; 312: 1503–1505.
Jørgensen HS, Kammersgaard LP, Nakayama H, Raaschou HO, Larsen K, Hübbe P, Olsen TS. The effect of a stroke unit: reductions in mortality, discharge rate to nursing home, length of hospital stay, and cost: a community-based study. Stroke. 1995; 26: 1178–1182.
Jørgensen HS, Kammersgaard LP, Nakayama H, Raaschou HO, Larsen K, Hübbe P, Olsen TS. Who benefits from treatment and rehabilitation in a stroke unit? : a community-based study. Stroke. 2000; 31: 434–439.
Jørgensen HS, Kammersgaard LP, Nakayama H, Raaschou HO, Larsen K, Hübbe P, Olsen TS. Treatment and rehabilitation on a stroke unit improves 5-year survival: a community-based study. Stroke. 1999; 30: 930–933.
Koton S, Schwammenthal Y, Merzeliak O, Philips T, Tsabari R, Orion D, Chapman J, Tanne D. Management and outcome of acute stroke: differences between an organised short-term acute stroke unit and general medical wards. Cerebrovasc Dis. 2005; 19 (suppl 2): 58. Abstract.
Olsen TS, Rasmussen BH, Gerner U, Kammersgaard LP. Focus on stroke units, diabetes and ischemic heart disease may improve stroke survival. Cerebrovasc Dis. 2004; 17 (suppl 5): 116. Abstract.
Phillips SJ, Eskes GA, Gubitz GJ. Description and evaluation of an acute stroke unit. Can Med Assoc J. 2002; 167: 655–660.
Ronning OM, Guldvog B, Stavem K. The benefit of an acute stroke unit in patients with intracranial haemorrhage: a controlled trial. J Neurol Neurosurg Psychiatry. 2001; 70: 631–634.
Stavem K, Rønning OM. Survival of unselected stroke patients in a stroke unit compared with conventional care Q J Med. 2002; 95: 143–152.
Bhalla AJ, Grieve R, Tilling K, Rudd AG, Wolfe CDA. Older stroke patients in Europe: stroke care and determinants of outcome. Age Ageing. 2004; 33: 618–624.
Cadhilac D. Evaluation of the New South Wales Greater Metropolitan Transition Taskforce. Stroke Unit Initiative, 2003/2004 Final Report: National Stroke Research Institute report. Melbourne, Australia: 2004.
Daffertshofer M, Mielke O, Pullwitt A, Felsenstein M, Hennerici M. Transient ischemic attacks are more than ministrokes. Stroke. 2004; 35: 2453–2458.
Glader EL, Stegmayr B, Johansson L, Hulter-Åsberg K, Wester PO. Differences in long-term outcome between patients treated in stroke units and in general wards: a 2-year follow-up of stroke patients in Sweden. Stroke. 2001; 32: 2124–2130.
Rudd AG, Hoffman A, Irwin P, Lowe D, Pearson MG. Stroke unit care and outcome: results from the 2001 National Sentinel Audit of Stroke (England, Wales, and Northern Ireland). Stroke. 2005; 36: 103–106.
Herzig R, Vlachova I, Kettnerova Z, Krupka B, Mares J, Kral M, Urbanek K. Stroke unit versus conventional neurological ward: do stroke patients really benefit? Cerebrovasc Dis. 2004; 17 (suppl 5): 103. Abstract.