Data and measures

We obtained anonymised patient-level data from the SSNAP,31 for all patients in England with a primary diagnosis of stroke (ischaemic stroke or primary intracerebral haemorrhage) between 1 January 2014 and 31 December 2014. SSNAP collects data on clinical characteristics, care quality (from the time of admission up to 6 months after stroke) and outcomes for all stroke patients admitted to acute care hospitals in England.32–34 During our study period the case ascertainment in the SSNAP, which is calculated as the proportion of all acute stroke patients admitted to hospitals, for England was estimated to be 90%.35 We excluded patients treated at hospitals in Wales from our analysis because for Wales the case ascertainment was estimated to be 60%.33

The following quality of care indicators were measured from time of hospital admission (or onset of stroke symptoms for those who were already in hospital): brain scan within 1 hour and within 12 hours, dysphagia screen within 4 hours, assessment by a nurse trained in stroke management within 24 hours, administration of intravenous thrombolysis to eligible patients, door-to-needle time within 1 hour in patients receiving thrombolysis, assessment by a stroke specialist consultant physician within 12 hours* and within 24 hours, admission to a stroke unit within 4 hours, assessments by a physiotherapist within 24 hours* and within 72 hours, assessments by occupational therapist within 24 hours* and within 72 hours and assessments by speech and language therapist within 24 hours* and within 72 hours. These measures are quality indicators routinely reported by SSNAP; we also included measures (marked with a *) with more stringent time constraints to reflect the time-critical nature of acute stroke care. Outcomes were measured as whether or not the patient died within 3 days and disability using the modified Rankin Scale (mRS) score 0 to 2 versus 3 to 6 (moderate, moderately severe or severe disability or death) at the end of the inpatient stay. We also analysed mRS score 0 to 2 versus 3 to 5 at the end of the inpatient stay, excluding patients who died. Mortality data beyond hospital discharge were not available in SSNAP; we therefore measured mortality up to 3 days after admission to minimise the number of missed deaths. We analysed length of stay (LOS) in the HASU (in London only) and LOS in hospital. The denominators used for each measure were consistent with the SSNAP key indicators.36 Most outcomes were measured for all patients, but there were exceptions: patients who were medically unwell or refused to be screened were excluded from the dysphagia screen measure, only patients with ischaemic stroke who met the Royal College of Physicians guideline minimum threshold for thrombolysis were included in the thrombolysis rate, door-to-needle times included only those who received thrombolysis with a final diagnosis of stroke and patients who were persistently medically unwell, declined to be assessed or had no relevant deficit were excluded from the therapy performance measures.

To examine variations across the week we initially used a flexible specification of time of admission, measured in six 4 hour periods from 00:00 to 03:59, 04:00 to 07:59, 08:00 to 11:59, 12:00 to 15:59, 16:00 to 19:59, 20:00 to 23:59 for every day of the week (42 periods). We also created a more restrictive measure to examine broad trends across the week: Monday to Friday 08:00 to 19:59, Monday to Friday 20:00 to 07.59, Saturday and Sunday 08:00 to 19:59, Saturday and Sunday 20:00 to 07.59 (four periods) following Bray et al 16 who found variations across the week with both specifications.

Statistical analysis

We ran patient-level logistic regressions, regressing each measure against time period of admission. For LOS we used parametric survival models (modelled as time to event of discharge) assuming a lognormal survival distribution. We ran separate models for London and the rest of England. In every model we controlled for sex, age (continuous variable), ethnic group (six categories), type of stroke (infarction or primary intracerebral haemorrhage), comorbidities prior to admission (five options), mRS before stroke (0 to 2, 3 to 5), level of consciousness on arrival at the hospital (four categories), method of admission to the hospital (three categories), time from onset of stroke symptoms to admission (four categories), month of admission (12 categories) and hospital trust. When analysing mRS scores 0 to 2 versus 3 to 5 at the end of the inpatient spell we additionally controlled for the number of days after admission at which the mRS score was measured. We were unable to do this for the analysis of mRS score 0 to 2 versus 3 to 6 as date of death was not available. We tested for statistically significant variations across the week using Wald tests and reported the results as joint p values under the null hypothesis that the regression coefficients for every time period relative to the omitted time period were zero. We calculated the average predicted probability of each outcome (predicted median LOS in the case of the LOS variables) in each time period controlling for the covariates. Patients admitted with a diagnosis of acute stroke in London who were not treated in a HASU were excluded (6% all London patients in our data set were not treated in a HASU). P values<0.05 were considered to be statistically significant. Data on National Institutes of Health Stroke Scale (NIHSS) score, a validated measure of stroke severity on a scale from 0 (no stroke symptoms) to 42 (severe stroke), were available for 93% patients in London HASUs and 77% patients in the rest of England. Due to the extent of missing NIHSS data, in our main analysis we controlled for stroke severity using level of consciousness on arrival at the hospital (one component of NIHSS); we then reran all analyses controlling for NIHSS on arrival at the hospital on the smaller sample instead of level of consciousness on arrival. The findings using NIHSS score on arrival were qualitatively the same and are presented in the online supplementary figures S1-S6.

Patient and public involvement

Two stroke patient representatives contributed to the design of our study protocol and development of the research questions; they also contributed to discussions of interim findings presented at study steering committee meetings in June 2015 and July 2016, raising issues related to variation in quality of care and mortality, which we incorporated into our analysis. They were consulted on the methods for disseminating the outputs of this study and ensured that we were addressing questions and communicating lessons in a meaningful way. The findings of this research will be disseminated to the relevant patient community in an accessible way.

Principal findings

In our study, we found no evidence for an admission effect across the week on early outcomes in acute stroke patients admitted to a London HASU: 3-day mortality and modified Rankin Scale score at hospital discharge did not vary by day and time of admission in London HASUs. This is consistent with a recent study based on administrative data in the UK9 that found a steady reduction in in-hospital mortality difference between weekday and weekend stroke admissions in 2008 to 2014 across England and that this difference is no longer statistically significant in 2014.

There was also no variation by day and time of admission across the week in terms of rapid access to brain scanning, stroke nursing care and thrombolysis in London HASUs. Other quality of care measures did significantly vary across the week in London HASUs, and three patterns of variation were detected: by time of day but not day of the week, by day of the week but not time of day and by time of day and day of the week. LOS was longer among patients admitted to London HASUs at the weekend. In the rest of England there was variation in all measures by day and time of admission across the week, except for mortality at 3 days. We hypothesised there would be less variation across the week in care quality measures in London HASUs compared with the rest of England, and that this would translate into less variation in outcomes in London HASUs. The lower variation in care quality measures across the week in London HASUs was confirmed, but only with respect to ‘front door’ measures of acute stroke care. With respect to the health outcomes: there was no variation in mortality at 3 days and disability at hospital discharge by day and time of admission across the week in London HASUs. This is consistent with previous studies showing that timely access to thrombolysis is associated with good stroke outcomes.37 In the rest of England there was no variation in 3-day mortality by day and time of admission across the week (but there was in terms of disability after discharge), suggesting the lack of variation in outcomes in London HASUs may not be exclusively attributed to the lack of variation in ‘front door’ quality of care.

Strengths and limitations

The main strength of our study is the large national data set we have used containing detailed information on quality of care, outcomes and patient characteristics. We have examined whether time of admission was related to quality of care using a comprehensive set of indicators from across the acute stroke care pathway. Most of the measures were from a pre-existing set of national acute stroke care indicators, and those that were added had more stringent time constraints to reflect the time-critical nature of acute stroke care. Our outcomes were stroke mortality and disability, where previous studies have focused on mortality.2 4 5 7–10 The rich set of patient characteristics in the data set meant we could control for patient factors likely to affect quality of care and outcomes that vary by day and time of admission across the week and between London and the rest of England. There are several limitations. First, while case ascertainment in SSNAP was 90% during the time period of our study, these data might not be representative of all stroke patients. For example, not all hospitals receiving acute stroke patients in England participated in SSNAP, and the results may not be representative of hospitals who did not participate. Second, while analyses of hospital administrative data to investigate weekend effects in stroke have been undermined by evidence of variations in inaccurate coding across the week,15 in SSNAP data are inputted voluntarily by hospitals and we cannot exclude the possibility of inaccurate or selective reporting. Particularly problematic for our study would be if this bias was more likely to occur in London or the rest of England and/or if it was more likely to vary by time of admission. Third, we were unable to measure long-term outcomes as these were not available in SSNAP. Mortality data in SSNAP are currently only available for patients who are in hospital and therefore to reduce the risk of bias we measured mortality at 3 days after admission when most patients will still be admitted. Three-day mortality has been used in previous studies to evaluate the centralisation of acute stroke services in London,30 but the focus in our study on in-hospital mortality only is a further limitation. Similarly, long-term disability data are not reliably collected in SSNAP, and so this was measured by mRS at the end of the inpatient spell. Fourth, while the richness of our data set means we have been able to control for confounding factors we cannot exclude the possibility of confounding due to unobserved patient characteristics or staffing levels. Fifth, while the sample size of our study is large in both London and the rest of England, when evaluating quality of care and outcomes across the week the number of observations in each time period was considerably smaller in London. We cannot exclude the possibility that the smaller number of patients in London resulted in wider CIs around the adjusted predicted probabilities in each time period making it less likely to show significant variation in the measures evaluated.

Comparison with other studies

There is a large literature examining weekend effects in healthcare across a range of clinical areas.38 In acute stroke there is conflicting evidence as to whether patients admitted at weekends have higher or lower quality of care and better or worse outcomes,1–8 but recent analyses have shown that care quality and outcomes in acute stroke vary across the week, and that comparing weekend versus weekday or in-hours versus out-of-hours effects is flawed as it does not take into account variations by day of the week and time of day.16 This study, using the same data set as ours but from an earlier time period and analysing the whole of England and Wales, found that quality of care varied across the entire week, not only between weekends and weekdays, with a number quality of care measures showing different patterns of variation over the week. While the findings mirrored our own for the rest of England, one noticeable difference was in mortality: Bray et al reported that patients admitted overnight on weekdays had lower odds of survival (0.90, 95% CI 0.82 to 0.99) compared with those admitted during the day at weekdays; this difference might be because our survival measure is not the same (3 vs 30 days) and/or because our extract of the SSNAP data set is more recent. What our study adds is analyses of variation in quality of care and outcomes in London HASUs separately following the centralisation of acute stroke services in London in 2010, which has been shown to increase the quality of care and outcomes on average across the week.29 30 Our findings were further expanded in Black et al 39 that aimed to identify factors influencing this variation.39

Implications

There are several implications of our study. The first is that London HASUs appear to operate a uniform service across the week with respect to some but not all aspects of acute stroke care. Performance standards originally set by Healthcare for London stipulated that London HASUs should operate a 24/7 service with respect to first assessment by a stroke nurse, rapid access to brain scans and administration of thrombolysis to eligible patients; our findings show that London HASUs do operate a 24/7 service with respect to these measures. However, for other less time-critical measures, such as senior stroke physician assessment within 24 hours and therapist assessments within 72 hours, we found significant variation by day and time of admission across the week in London HASUs. This suggests that some performance standards like ‘front door’ interventions may be emphasised more than others and analysis of qualitative data collected in Black et al 39 complemented our findings.39 The second implication is that there are differences in acute stroke care between London HASUs and the rest of England across the week, with less variation in quality of care and outcomes in London HASUs. The main differences were observed in nursing care, brain scanning and thrombolysis provision, and also with the type of variation observed for stroke consultant care. For these measures, our results show that the centralised model in London is more effective at providing constant care across the week. In terms of comparing London and the rest of England, four further issues are worth bearing in mind. First, our study focuses on patients admitted to London HASUs only, not other hospitals in London; our data suggest that 6% of acute stroke patients in London are not treated in a HASU. However, some of these patients will not have been eligible for HASU care because of greatly delayed presentation or identification of stroke, and others will have had a stroke after surgical procedures or in another context which precluded their admission to a HASU. Our focus on London HASUs was deliberate as the aim of our study was to evaluate the HASU model, but it means that our findings for London HASUs should not be generalised to all patients in London. Indeed, there is evidence that quality of care is lower for acute stroke patients in London not treated in a HASU compared with those who are.29 Second, and conversely, HASUs operate in many other parts of England using different models of care.31 40 In Greater Manchester, for instance, HASUs have also been shown to have higher quality of care than the rest of England excluding London.29 Hence the differences observed between London HASUs and the rest of England cannot be interpreted as a direct comparison of HASU versus non-HASU care, though if HASU-based care outside London was removed from the rest of England then the differences observed in this study are likely to be the same or greater. The third issue is that the London model may not apply to services operating in rural settings — in particular the greater travel times in rural areas make centralisation challenging.41 This means that potential benefits of the London model in terms of 24/7 care are unlikely to be achieved nationwide. The fourth issue is that the centralisation of acute stroke services in London was estimated to occur at an additional cost of £20 million, allocated to cover the increased cost per bed day in a HASU.28 With this additional level of funding it might be expected that the quality of care in London should improve, though whether it should produce less variation in quality of care and outcomes across the week in London compared with the rest of England depends on the relative levels of funding in both areas. There is some evidence that the reorganisation in London was cost-effective,42 43 but further analyses accounting for the size of the upfront investment, the relatively high costs per day of hyperacute stroke care, the impact on mortality and disability and the lifetime costs incurred by the NHS, social services and families caring for stroke survivors at different levels of disability would be helpful.

Future research

Further research would be beneficial to evaluate the impact of stroke admission at different times of the week on longer-term mortality and disability outcomes, and to investigate the relationship between quality of care and outcomes and if this relationship varies by time of admission. Further research would also be useful to investigate the reasons for the differences in variation found between London HASUs and the rest of England, and why for some standards care in London HASUs was constant across the week, irrespective of day and time of admission, but for others it was not. Performing follow-up studies to monitor attainment of key quality indicators and outcomes, complementary to the SSNAP clinical audit annual reports,33 44 would also be beneficial in order to get an overall picture of national trends and dynamics over time and look in detail at underlying reasons for that to understand what amendments to clinical guideline for stoke care ought to be proposed in the future. Also, accounting for the organisational factors at the stroke unit level could explain an important part of the variation in quality of acute stroke care and outcomes by day and time of admission in London HASUs and the rest of England. This research would help to further inform how acute stroke services ought to be designed in future to maximise patient outcomes in a cost-effective manner.