Impact of major stroke service centralisation on mortality and care: analysis of admissions, interventions and outcomes in South Australia

Abstract

Background

See More

Major system reform is complex but can yield improved outcomes at multiple levels. We aimed to evaluate the impact of implementing a hub and spoke model of stroke care across metropolitan Adelaide (population 1.2 million), South Australia on mortality, morbidity, service and quality stroke indicators.

Methods

Analysis of 24 months of prospectively collected, patient-level data covering all metropolitan stroke admissions during the contiguous pre-, during- and post-implementation time periods, linked to mortality data from the National Death Index. The three metropolitan tertiary hospital-based stroke units undertook the implementation of a centralised ‘hub and spoke’ model: one central comprehensive stroke centre offering 24 h stroke reperfusion therapies, and two primary stroke centres providing 12 h thrombolysis. The main outcome measures were mortality (any cause) up to 180 days post-admission; reperfusion treatment proportions and timings; stroke care quality composite metric; length of stay.

Results

There were 3917 confirmed stroke admissions over the 24-month period (3325 (84.9%) ischaemic) and 650 deaths (19.6%) within 180 days. Compared to the baseline period, post-intervention mortality and discharge disability did not differ, although a possible temporary increase in ischaemic stroke mortality during implementation was seen. Rates of endovascular thrombectomy (EVT) (5.7% vs. 12.5%, adjusted Rate Ratio (aRR) = 1.94, 95%CI 1.21,3.10) and timeliness of EVT (median 126 min (IQR 83, 154) vs. 95 min (53, 132), p < 0.001) improved as did the composite stroke quality metric indicator (0.60, 95% CI 0.50, 0.70 vs. 0.64, 95% CI 0.50, 0.75; adjusted difference 0.041, 95% CI 0.015, 0.066). Length of stay decreased for ischaemic stroke (8.2 (SD 12.4) vs. 7.9 (SD 8.9) days, adjusted geometric mean ratio = 0.83, 95% CI 0.73. 0.94) but not for intracerebral haemorrhage.

Conclusion

The major implementation of a metropolitan centralised ‘hub and spoke’ model of acute stroke care was associated with overall significant improvements in process indicators but a possible temporary increase in ischaemic stroke mortality during implementation.

Background

Stroke remains the second highest cause of mortality internationally and the primary cause of long-term adult neurological-based disability [1]. Age-adjusted incidence of stroke is falling at a rate of 1-1.5% per year in high-income countries, [2] however this is offset by an ageing population. For example, United Kingdom authors calculate that due to the changes in the proportion of more aged subgroups alone, there will be an overall 13% increase in the number of first-ever stroke cases in the UK by 2045 [2]. Prevalence is also growing as stroke fatality reduces. This means that whilst there are increased survival rates, there is still residual disability, and so the overall burden of stroke is also increasing, especially in low-income countries [1].

This current and projected growth and accompanying economic pressures have led many health systems to investigate more efficient and effective models of stroke care delivery. The evidence base for stroke care is robust and dynamic, and successful system redesign examples have been reported internationally (for examples see International Guidelines and Statements from USA, [3] UK/Ireland, [4] or Australia and New Zealand [5]. Most consistently, this has been in the form of centralizing services to ensure stroke care is provided by health professionals with sufficient experience and currency of practice to provide a high degree of stroke-specific expertise [6, 7]. The major benefits of centralisation include standardized care in a dedicated stroke unit (staffed by stroke-expert multi-disciplinary teams [8] and access to emerging reperfusion through thrombolysis (intravenous tissue plasminogen activators or tPA) and endovascular therapy (EVT) [9,10,11]. These latter interventions are both time-limited and require expert-level skill and knowledge to firstly ascertain eligibility and to then deliver the interventions promptly and effectively.

One of the best documented centralisation processes of stroke care was performed around 2010 in the metropolitan London and Greater Manchester areas. Though implementing slightly different models, both areas reduced the number of sites offering full stroke intervention and care, to concentrate on central ‘hub/s’ fully equipped and staffed, whilst regional ‘spokes’ offered more limited services. These reforms resulted in efficiencies and improved effectiveness with reduced mortality, length of stay [12] and improvement in provision of evidence-based clinical interventions [13]. These improvements were re-evaluated to confirm the benefits were maintained 6–10 years later [14].

In 2014, a major health reform process was initiated in South Australia (SA) – a state in Australia of approximately 1.6 million people. Scoping evidence had demonstrated that stroke care in metropolitan Adelaide (population 1.2 m) was variable between sites and between business hours and out-of-hours. Based on the London HASU model [12], centralisation of acute stroke reperfusion services was proposed. Whereas previously there were two 24-hour EVT centres, this specialised treatment was centralised to one central hub (the Royal Adelaide Hospital [RAH]). This same central hub also assumed responsibility for all thrombolytic therapy out of hours, with dedicated overnight medical and nurse staffing. A 7-day, 12-hour, daytime thrombolytic (non-EVT) ‘Code Stroke’ service was established at both the 14 km-distant southern spoke (Flinders Medical Centre, which previously provided a full 24-hour hyperacute service) and the 28 km-distant northern spoke (Lyell McEwin Hospital, which previously provided a service 0800–1600 Monday to Friday).

This system re-design commenced implementation in March 2017 and was completed in September 2017, at which time the RAH moved from an older hospital to a new bespoke site. In the lead-up to the move, patient flows were partially diverted from the ‘old RAH’ to permit planning and relocation. The service model also implemented standardized clinical protocols across all sites (single service, multiple site model) with formal linkages, data collection and governance.

An evaluation process was also recommended, tasked with determining the effects of redesign on process and outcome. Risks were identified around increased (unacceptable) ambulance travel times when bypassing the suburban spokes (missing valuable time to intervention and resulting in reduced outcomes); the distance from family/community supports if not in their ‘local’ hospital; poor communication between units and with ambulance services; overloading of the central hub and issues with repatriation back to the ‘spoke’ unit resulting in increased length of stay/s.

We investigated the effects of implementing these changes to stroke service delivery across metropolitan Adelaide on mortality and key service indicators. The most fundamental aim was to determine that no harm or reduction in favourable outcomes occurred as a result of the changes (particularly with respect to the theorised risks) and further aims were to determine any benefits, as part of the full implementation process, particularly with respect to standards or benchmarked indicators and in equity of access to hyperacute services.

Our primary hypothesis was that mortality rates would not be worse in the post-implementation period compared to the pre-implementation period. Secondary hypotheses included that rates (and timing) of access to hyperacute interventions would increase (tPA and EVT), length of stay (in tertiary hospital/s) would reduce, and that other national stroke quality indicators would demonstrate improvements. The quality indicators of interest were the extant national Acute Stroke Clinical Care Standard indicators [15] and included access to stroke unit care, early rehabilitation, minimizing risk of another stroke, and transition from hospital care.

Methods

Ethical approval to collect patient-level data from all stroke admissions was obtained from SA Health, each of the participating Local Health Networks (LHNs: Central, Northern and Southern) and from the University of South Australia (HREC/16/RAH/509). Approval was obtained for an opt-out process, whereby each person admitted to one of the hospitals was given an information sheet advising that their data would be collected and used for the purposes of this project. However, if they did not wish for this to happen, they were given options to opt-out including discussing verbally with a staff member, writing to an address (mailbox or email) or using a free telephone number. This process met the requirements of the National Statement on Ethical Conduct in Human Research 2007 (Sect. 2.3.6) [16].

Data were extracted from medical records of all non-opting out participants following admission, in the pre-defined time period commencing December 2016 and finishing December 2018. This allowed for 4 months of pre-transition data (until March 2017), 6 months of data over the transition period (which also included the move of the old Royal Adelaide Hospital to a new site 2 km away) until September 2017 and then a further 14 months of data for the post-transition period. Ward and research staff worked together to identify all potential stroke diagnoses based on clinical symptoms and signs. Dual ascertainment occurred – both prospectively and retrospectively using International Classification of Diseases (ICD-10) stroke codes (I61, 62.9, 63 and I64 (first three codes)). All cases were validated as confirmed stroke by senior clinicians.

Outcomes

Our primary outcome was mortality within 6 months of admission. Secondary outcomes were time from arrival to commencement of reperfusion therapy (intravenous tPA or arterial puncture for EVT); length of inpatient stay, independent mobility at discharge (modified Rankin Score 0–3 versus 4–6) and a composite quality indicator (based on the extant Australian Acute Stroke Clinical Care Standards) [15]. The quality composite indicator comprised the sum of the following binary quality indicators, adapted for local use:

• ROSIER Stroke screening in ambulance and/or Emergency Department.

• Thrombolysed for ischaemic stroke.

• Thrombolysed for ischaemic stroke within 60 min of arrival.

• Thrombectomy in ischaemic stroke.

• Admission to stroke unit.

• Physiotherapist assessment within 48 h.

• Discharged on appropriate medicines.

  • ICH - antihypertensives (unless contraindication).

  • ischaemic stroke –antihypertensive, statins, anticoagulants or antithrombotics (unless contraindication).

  • All cardioembolic ischaemic stroke due to AF or atrial flutter – anticoagulation (unless contraindicated).

• Risk factor advice on discharge.

• Documented care plan.

Data items were selected from the standardised National Stroke Data Dictionary and the Australian Stroke Data Tool (AusDAT) (http://australianstrokecoalition.com.au/ausdat) and were entered into paper-based forms or directly into online spreadsheets with sense-checks. Items included identifying information to allow data linkage and patient ascertainment and stroke characteristics, as well as the above quality indicators.

Further data were sought including mortality data from the National Death Index (NDI) which were linked to records, using a unique health identifier number, for the 180 days post-admission.

Statistical analysis

Our statistical analysis plan was lodged in Open Science Framework [17]. All analyses were performed in R version 4.3.2 [18]. Patient characteristics and clinical quality indicators were summarised by time period using descriptive statistics. Our primary outcome - analysis of mortality within 6 months of admission - used survival analysis with the R package ‘survival’. Patients were followed from admission until death, next stroke, or 6 months, whichever came first. Only the first stroke per patient during the study period was included.

The difference in mortality between time periods was displayed using Kaplan-Meier plots for each stroke type separately, stratified by time period, and compared using the log-rank test. Cox proportional hazards models were used to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) for the time periods relative to the reference (pre-transition) period for each stroke type separately. A modification was made to the original statistical analysis plan to analyse stroke type separately to deal with non-proportionality in the Cox proportional hazards model. Covariates included in the model were age, sex, whether a patient could walk on admission, pre-stroke disability, prior stroke, vessel occlusion for ischaemic stroke, smoking status, number of co-morbidities, external transfers (from non-stroke hospitals), season and clustered by hospital.

Absolute differences in mortality rates between the time periods were estimated by predicting mortality with the Cox models in the stroke population standardised across the three time periods [19, 20]. Bootstrapping was used to generate 95% confidence intervals for mortality risk differences. Post hoc power calculations for the primary outcome were performed using the MedCalc calculator for survival analysis.

Analysis of secondary outcomes included all strokes in the study period, including second or subsequent strokes. Time to thrombolysis or reperfusion for thrombectomy between time periods were compared using the Kruskal Wallis test. Patients with missing or implausible times were excluded from the analysis. Differences in rates between time periods were analysed using Poisson regression, using robust standard errors by hospital clusters and reported as rate ratios (RRs) and 95% CIs. Due to the evolving evidence for expanded window thrombectomy (from 6 to 24 h) during the study period, [21, 22] we analysed thrombectomy rates overall, as well as thrombectomy excluding patients presenting 6–24 h since last known well time (the eligibility upper limit in 2016).

The composite score ranged between 0 and 1, with 1 representing the maximum quality. A change greater than 0 represents an increase in quality. The change between the median composite quality indicator between periods was analysed using quantile regression with the R package ‘lqmm’. Covariates were the same as those in the Cox proportional hazards models, plus stroke type.

Linear regression was used to analyse log-transformed length of stay (LOS), with the same covariates as the Cox proportional hazards models, fitted for each stroke type separately. After back-transformation, geometric mean ratios and 95% confidence intervals were reported for the time intervals compared with the reference period.

Time to in-hospital death, time to discharge without a disability (modified Rankin Score 0–2), disability (modified Rankin Score 3–5) or death and discharge to a nursing home or death were analysed using Cox proportional hazards models with the same covariates as the primary outcome analysis. Outcomes were censored 30 days after hospital arrival. Hazard ratios and 95% confidence intervals were calculated by stroke type for the time period comparisons.

Missing data in the outcomes or covariates, were accounted for using inverse probability weighting [23].

Sensitivity Analyses were performed for the primary outcome using the National Institutes of Health Stroke Scale (NIHSS) at admission as a covariate instead of whether a patient can walk. Missing NIHSS was imputed using inverse probability weighting [23]. Sensitivity analyses were also performed for the primary outcome excluding patients potentially meeting expanded post-2017 EVT criteria (presenting 5–24 h since last known well time with either an internal carotid artery and/or M1 and/or M2 Middle cerebral artery occlusion with a NIHSS of > 5).

Results

Population description

Overall, 3796 people with stroke were included over the two-year time period: 3218 (85%) were ischaemic stroke. Patient characteristics in each time period are shown in Table 1. Median age and proportion of each sex were similar across the three periods. The proportion of external transfers (from non-stroke hospitals) increased from 4% in the pre-transition period to 6% in the post-transition period. Some of these may have been rural hospitals but also included metropolitan hospitals without stroke services (where patients sustained stroke during a hospitalization or were ‘walk-in’ stroke patients (ambulance protocols otherwise mandated bypass of these hospitals). NIHSS increased by one point and the proportion who could walk on arrival was lower in the post- compared with the pre-transition period.

Table 1 Patient characteristics at hospital arrival by time period, noting the transition period also included the moving from the old RAH to the new RAH premises

Mortality

The six-month mortality rate for the 2-year period was 789 deaths from 3,670 first strokes (21.5%). Figure 1 shows the Kaplan Meier plots for mortality for the pre-, transition and post-periods for ischæmic stroke and intracranial haemorrhage (ICH). Mortality rates were higher in ICH than ischaemic stroke. The log-rank test did not show any differences between time periods for either ischaemic stroke (p = 0.36) or ICH (p = 0.63).

Fig. 1

Kaplan-Meier plots for mortality across the three time periods and for both ischaemic and intra-cranial haemorrhagic (ICH) stroke

After adjustment for risk factors in the Cox models, ischaemic stroke mortality during the transition period was significantly higher than the pre-transition period, with HR = 1.64 (95% CI 1.10,2.44). In the post-transition phase, mortality rates were similar to the pre-transition period (HR 1.01, 95% CI 0.72–1.42). For ICH, there was no difference in mortality rates between any of the time periods [transition vs. pre: HR 0.77, 95% CI (0.37, 1.57); post vs. pre: HR 1.01, 95% CI (0.55, 1.85)].

Absolute differences in mortality by time period were estimated for the transition and post-periods compared with the pre-transition period (Table 2). Mortality was higher for ischaemic stroke (risk difference (RD) = 6.5%, 95% CI (1.5%, 11.2%) during the transition period but returned to pre-transition risk levels in the post-transition period (RD = 0.1%, 95% CI (−14.6, 13.3)). There were no significant changes for intracranial hemorrhage. Overall mortality was also higher in the transition period (RD = 4.5%, 95% CI (0.9, 8.0)) but returned to baseline levels in the post-transition period (RD = 0.1%, 95% CI (−3.2, 2.7)).

Power to detect mortality differences in the Transition vs. Pre periods for ischaemic stroke and overall was 91% and 80%, respectively. See Supplementary file, Table S1. All other comparisons had < 1% power; however, mortality differences in the Post vs. Pre periods for ischaemic stroke and overall were very small, approximately 0.1%. Power in ICH was restricted by small sample sizes.

Table 2 Absolute risk differences for 6-month mortality by time period and stroke type

Rates and timing of thrombolysis and thrombectomy

Thrombolytic rates were similar in the transition and post- periods compared with the pre-transition period (15.4% pre, 13.7% transition and 15.6% post (adjusted rate ratio (aRR) for transition vs. pre 0.93 (95% CI 0.64, 1.34) and post vs. pre 0.95 (95% CI 0.70, 1.29).

EVT rates increased in the post-transition periods compared with pre (5.7% pre, 7.1% transition and 12.5% post; sRR for post vs. pre [aRR 1.94, 95% CI 1.21, 3.10], versus transition vs. pre [aRR = 1.21 (95% CI 0.61, 1.28)]). When restricted to thrombectomy within 6 h of last known well time, rates were 5.7%, 6.6% and 11.2%, respectively (post vs. pre aRR 1.64, (95% CI 1.01, 2.68)).

Time from arrival to thrombolytic did not change significantly over the study period (p = 0.27) (Table 3). Arrival to thrombectomy improved from 126 min (IQR 83, 154) during the pre-transition phase to 95 min post-transition (IQR 53, 132) (p < 0.001).

Table 3 Descriptive statistics and estimated differences for thrombolysis and thrombectomy rates in ischaemic stroke, composite quality indicator and length of stay across the three time periods

Composite quality indicator

The composite score for the quality indicators ranged between 0 and 1, with 1 representing the maximum quality. Pre-transition, the score was 0.60 (95% CI 0.50, 0.70), improving to 0.64 (95% 0.50,0.75) in the post-transition period (Table 3).

A significant increase in the composite score (increase in quality) was found comparing the pre- to post-transition period after adjusting for risk factors (adjusted difference in median score 0.041, 95% CI 0.015, 0.066). Individual quality indicators for each of the three time periods are reported in Supplementary Table S2.

Length of stay (LOS)

Length of stay for ischaemic stroke was significantly lower in the transition and post-periods compared with pre-implementation [pre 8.2 days (SD 12.4), transition 6.7 days (SD 34.5), post 7.9 days (SD 8.9); adjusted geometric mean ratio: transition vs. pre 0.78 (95% CI 0.67, 0.90); post vs. pre 0.83 (95% CI 0.74, 0.94)] (Table 3), corresponding to a 22% and 17% reduction in LOS in the transition and post-transition periods. After adjustment for risk factors, there was no difference between time periods for ICH.

For both stroke types in the pre-transition period, 451 patients had a total LOS of 3823 days (median 5.1 days (IQR 2.9–9.1); mean 8.2 days (SD 12.4)). In a post-transition period which matched the months of the pre period (1/12/2017-3/5/2018) there were 489 patients with a total LOS of 3559 days (median LOS 4.6 (IQR 2.3–7.9) days; mean 7.3 (SD 9.4) days). This equates to approximately 11 extra stroke patients per month in the post-transition period while LOS reduced by more than half a day per patient. Furthermore, the variability (SD) and the 75th percentile in LOS were lower in the post-transition period, meaning the number of long stay patients was fewer.

Discharge outcomes

Disability at discharge, discharge to a nursing home or death and in-hospital death in ischaemic stroke and disability at discharge in ICH were all higher in the transition compared to the pre period for ischaemic stroke (See Supplementary Tables S3 and S4). In the post-implementation period, however, all discharge outcomes were not significantly different to the pre period for both stroke types.

Sensitivity analysis

Sensitivity analyses excluding patients with intracranial/M1/M2 vessel occlusion in ischaemic stroke showed similar absolute risk differences in six-month mortality between time periods (see Supplementary Table S5). A sensitivity analysis stratifying by NIHSS quartiles rather than ability to walk was also performed after imputation of missing NIHSS values with inverse probability weighting [23]. In this analysis, there was no significant difference between absolute risk differences at any time period (see Supplementary Table S6). However, this result should be treated with caution since it was possible that NIHSS was missing not at random, since patients who died were more likely to have missing NIHSS, and this was not accounted for in the imputation.

Discussion

Major health service reform is costly from a staffing and resourcing perspective. Therefore, it is imperative that data are gathered and evaluated to report the benefits (or harms) of such significant changes. Much of what was evaluated in the 2-year period of systematic change in the Adelaide experience is comparable to the effects and outcomes reported in previous studies of similar undertakings in the UK and elsewhere. However, there is surprisingly little reported on major service reforms, and it unclear whether this is because reforms happen but are not evaluated, or major stroke service reforms are rarely undertaken.

Our service reforms had some similarities to those reported in the London and Manchester experiences in the UK some years prior [24]. We also used processes of consultation, data analysis, model development and specification/selection of services (hub and spoke centralisation). The leadership and governance in Adelaide were also a combination of government Health Department alongside the statewide Stroke Network (a multidisciplinary committee including clinical and consumer representation). However, in contrast to the UK experience, our service reforms did not extend into rehabilitation, other than repatriation to the spoke hospitals influenced referral to inpatient rehabilitation; we had limited a priori evaluation plans (other than the data in this report, for example no economic evaluation) and our scale of change (number of hospitals and population involved) was significantly less [12].

Specifically, we found no harm after the reforms regarding mortality. Risk evaluations during planning had identified triage and travel times for repatriation to the central location were a potential factor for harm. Fortunately, the mitigation modelling proved correct - these times did not increase and did not impact negatively. The increase in mortality during the transition period was potentially related to the geographical move to the new RAH, or alternatively a chance finding, given the asymmetric intervention periods. The London experience was a significant reduction in mortality (absolute and relative − 1.1% and 5% respectively) whereas the reduction in mortality rates in Greater Manchester were the same as those seen in the rest of England over the same period [12].

The finding that the mortality rate for ICH went down in the transition period (whilst the IS rate went up) was unexpected. In this period, 42% of patients with ICH strokes could walk on arrival, much higher than 31% in the Pre period (see Supplementary Table S7). In IS patients, ability to walk on arrival was only slightly higher (63%) in the Transition period than the Pre period (59%). Median NIHSS was lower in the transition than the pre period for ICH, but higher for IS. These differences may explain the decrease in mortality rates in Table 2 in the Transition period for ICH but not IS. It should also be noted the ICH mortality rates did have wide confidence intervals (only 48 cases in the pre-period) and failed to reach significance compared across periods.

Rates and time to thrombolysis did not change significantly over the evaluation periods. This is interpreted as positive as major reforms for pre-hospital notification had already been implemented in the metropolitan area and therefore the important triage to primary centers (such as had been reported in the Chicago reforms [6] had already been achieved. As mentioned above, it was important to confirm that bypassing the spokes during the overnight periods did not negatively impact tPA rates or timing. Median times to tPA of around 70 min and rates of around 15% compare favourably to reported results elsewhere [9].

Our rates and speed of thrombectomy increased significantly from 5.7 to over 12% of all ischaemic stroke. Recent estimates suggest that rates of between 3% and 22% are potentially eligible depending on the specific selection criteria used [25]. As mentioned earlier some of the increase in rates in this study were expected to be related to the change in criteria (expanding the window of eligibility to 24 h) however, our analyses excluded delayed window patients still found a significant increase (at around 11%). The reduction in median time to EVT (126 min to 94) was statistically significant as well as clinically meaningful – with every hour of delay impacting outcomes (see HERMES study) [25]. Using the formula from a re-analysis of EVT data (where every minute saved in onset-to-treatment time grants 4.2 days of extra healthy life) [26] this decrease of 22 min represents an extra 92 days (13 weeks) of healthy (disability free) life for stroke survivors.

Quality indicators improved as a composite. This appears to be driven by larger percentage increases in thrombectomy, as well as the indicators for early instigation of physiotherapy assessment and rehabilitation. This latter is most likely a reflection of allied health being employed on weekends to commence these processes – avoiding the previous potential 2-day delay for strokes admitted later in the week.

Length of stay decreased for ischaemic stroke by more than half a day per patient and the variability decreased (SD and the 75th percentile in LOS was lower in the post-transition period) meaning there were fewer long-stay outliers. The London and Greater Manchester reforms also produced a reduction of length of stay and only in ischaemic stroke [12] but at a greater magnitude of 1.4 to 2 days less. However, the mean lengths in the two countries differed, with the UK stays initially 21days dropping to 17–18 whereas the mean days for Adelaide were 8 dropping to around 7.

Our service improvements in this area equate to 11 extra stroke patients per month in the post-transition period, which is an important metric for hospital services to ensure both equity of access as well as efficient movement through the system. The discharge disability levels did not change, so the earlier discharge was not as a result of people being transferred to rehabilitation, home or aged care at a more vulnerable or dependent stage.

Our clinical recommendations from these findings are that it is vital to continue to collect data that includes mortality as well as severity (age and NIHSS) so that age- and severity-adjusted surveillance can be evaluated. Data collection was initially a major impost on staff and therefore sustainable systems need to be embedded as simple, routine data collection with logic checks. Easily accessible and interpretable data like these will enable future quality improvement activities to occur in areas of need.

Whilst we are confident that the reforms improved equity and access, we need to continue to monitor equity of access to timely best practice for the whole of state population including those in rural and remote areas, and those who are culturally and/or linguistically diverse. It is worth noting that separate to these metropolitan reforms, subsequent and complementary processes for rural and remote telestroke services were established in key South Australia regions and resulted in improved care and lower stroke mortality for those patients [27].

There are some limitations that need to be acknowledged. First, we had asymmetric time periods as our research funding only covered prospective data collection for 3 months in the pre-transition phase. We chose to prospectively collect data, which limited the time for baseline collection, and hence limited the precision of pre-intervention measurements. Second, unlike the UK studies we did not have national data to compare parallel or secular trends that may have influenced improvements. Lastly, our sample size was low compared with the London and Manchester HASU study [12, 13].

Conclusions

This real-world implementation study, with comprehensive baseline and prospective stroke data, evaluated service change that is potentially applicable to many medium to large cities.

Centralising stroke care in Adelaide, Australia, was associated with improved intervention and care metrics and length of stay. Stroke outcomes and mortality at discharge did not differ, with possible evidence of harm during a transition phase.