Do expanded seven-day NHS services improve clinical outcomes? Analysis of comparative institutional performance from the “NHS Services, Seven Days a Week” project 2013–2016

Background The cause of adverse weekend clinical outcomes remains unknown. In 2013, the “NHS Services, Seven Days a Week” project was initiated to improve access to services across the seven-day week. Three years on, we sought to analyse the impact of such changes across the English NHS. Methods Aggregated trust-level data on crude mortality rates, Summary Hospital-Level Mortality Indicator (SHMI), mean length of stay (LOS), A&E admission and four-hour breach rates were obtained from national Hospital Episode Statistics and A&E datasets across the English NHS, excluding mental and community health trusts. Trust annual reports were analysed to determine the presence of any seven-day service reorganisation in 2013–2014. Funnel plots were generated to compare institutional performance and a difference in differences analysis was performed to determine the impact of seven-day changes on clinical outcomes between 2013 and 2014, 2014–2015 and 2015–2016. Data was summarised as mean (SD). Results Of 159 NHS trusts, 79 (49.7%) instituted seven-day changes in 2013–2014. Crude mortality rates, A&E admission rates and mean LOS remained relatively stable between 2013 and 2016, whilst A&E four-hour breach rates nearly doubled from 5.3 to 9.7%. From 2013 to 2014 to 2014–2015 and 2015–2016, there were no significant differences in the change in crude mortality (2014–2015 p = 0.8, 2015–2016 p = 0.9), SHMI (2014–2015 p = 0.5, 2015–2016 p = 0.5), mean LOS (2014–2015 p = 0.5, 2015–2016 p = 0.4), A&E admission (2014–2015 p = 0.6, 2015–2016 p = 1.0) or four-hour breach rates (2014–2015 p = 0.06, 2015–2016 p = 0.6) between trusts that had implemented seven-day changes compared to those which had not. Conclusions Adverse weekend clinical outcomes may not be ameliorated by large scale reorganisations aimed at improving access to health services across the week. Such changes may negatively impact care quality without additional financial investment, as demonstrated by worsening of some outcomes. Detailed prospective research is required to determine whether such reallocation of finite resources is clinically effective. Electronic supplementary material The online version of this article (doi:10.1186/s12913-017-2505-8) contains supplementary material, which is available to authorized users.


Background
In December 2012 National Health Service (NHS) England published "Everyone counts: planning for patients 2013/ 2014" [1], where plans for seven-day access to NHS services in the United Kingdom were outlined. This led to the establishment of an "NHS Services, Seven Days a Week" Forum (the "Forum") chaired by the National Medical Director aimed at initially improving access to diagnostic, urgent and emergency services across the seven-day week.
The need for these changes were argued to be fourfoldto improve excess weekend mortality, to increase costefficiency, to move the NHS in line with the retail sector, and to improve the patient "customer" experience [1][2][3]. The Healthcare Financial Management Association (HFMA) subsequently carried out a costing analysis using a voluntary sample of eight variably sized NHS trusts at different stages of implementation (Table 1) [2]. Of these, Salford Royal, Aintree University Hospitals, Guy [2,[5][6][7]9]. Similarly, across England, other NHS trusts started reorganising services to reduce the differences in care provision across the seven-day week (full list in Additional file 1: Table S1).
Three years on, various changes to the English NHS, including reformation of consultants' and junior doctors' contracts have been deemed necessary for a fully comprehensive seven-day service [10,11]. Their basis has been primarily the excess 30-day mortality associated with weekend admissions demonstrated in two analyses of national Hospital Episode Statistics (HES) datasets from 2009 to 2010 and 2013-2014 by Freemantle et al., despite the authors stating it would be "rash and misleading" to attribute this to deficits in service provision [12,13]. Given the three-year interval from the original Forum report, we aimed to review changes in clinical outcomes across the English NHS between 2013 and 2016, comparing NHS Trusts that had introduced seven-day  [16]. The SHMI is the ratio of the observed to expected number of deaths either in-hospital or within 30 days of discharge, with the latter based on the national baseline for England, calculated from a risk-adjusted model for patient case-mix including age, gender, admission method, time of year, Charlson Comorbidity Index [17] and diagnosis [18]. The four-hour waiting time target was introduced by the Department of Health in 2004 to improve efficiency with the aim of getting 98% of patients admitted, transferred or discharged from A&E within 4 h of attending [19]. This target was relaxed to 95% in 2010. Annual reports for 2012-2013 and 2013-2014 for each individual Trust were reviewed to determine if any seven-day service reorganisation had occurred during the period of interest. All forms of service reorganisation were considered, including weekend outpatient clinics and seven-day changes to mental and community health care, as these could have knock-on effects on admissions, inpatient bed availability, length of stay and A&E outcomes. Specialist mental health and community hospitals were excluded from all analyses ( Fig. 1), whilst emergency admission and four-hour breach rate data from specialist A&Es (e.g. paediatric A&E units) was included. Due to the expected lag time from when seven-day service reorganisation was initiated in any given trust to its full implementation, we analysed outcomes over both 2014-2015 and 2015-2016 to determine their full impact.

Statistical analysis
As the units of analysis for each outcome were individual trusts, the various outcome metrics were summarised as means (standard deviation (SD)). Funnel plots were generated to graphically compare institutional performance against other NHS Trusts across England. This method of analysis plots a clinical outcome against its precision, the latter usually proportional to the sample size [20]. Larger institutions exhibit less variation, so funnel-shaped "control" limits can be generated around the target outcome to compare like-sized institutions against each other. In these analyses, control limits corresponding to 95th (±2 SDs) and 99.8th centiles (±3 SDs) were set. Over-dispersion was accounted for by a random effects model for SHMI [18] and a generalised linear model for LOS, admission and breach rate data, both with a 10% trim (detailed methodology in Additional file 1) [20]. The difference in each of the four outcomes for each individual trust between 2013 and 2014, 2014-2015 and 2015-2016 was determined, and an unpaired t-test was used to compare the difference in differences. NHS Trusts undergoing significant restructuring through mergers with other Trusts (Fig. 1)

Results
Baseline characteristics of trusts (Fig. 1, Table 2) Outcome data from a total of 159 trusts were analysed, with 79 (49.7%) of these instituting seven-day changes in 2013-2014. Between 2013 to 2016, 11 NHS trusts underwent mergers to form six new NHS trusts, resulting in a final number of 154 trusts by the end of 2016. Additionally, South London Healthcare NHS Trust was dissolved in October 2013, with its three constituent hospitals being taken over by King's College Hospital NHS Foundation Trust, Lewisham and Greenwich NHS Trust, and Oxleas NHS Foundation Trust respectively (the lattermost Trust was not included in any analysis as this is a community hospital Trust with no A&E). These trusts were excluded from all pairwise difference in differences analyses. Eighteen specialist hospitals were excluded from the analysis of SHMI data due to the nature of the patients seen. Fifteen of these hospitals also did not have an A&E department and were excluded from analysis of A&E admission and four-hour breach rate data.

Summary Hospital Mortality Indicator (SHMI)
Overall annual crude mortality rates across the whole of the English NHS for the periods of April Trusts were evenly distributed throughout the SHMI funnel plots for all periods between 2013 to 2016 regardless of whether seven-day service reorganisation had occurred or not (Fig. 2 a-  such that it fell from above to below the +3 SD control limit between the two periods despite not undergoing any significant seven-day reorganisation. The mean differences in SHMI over the three periods were not significantly different between trusts that had implemented seven-day changes (n = 70) and those which had not  There was no apparent change in pattern of distribution of trusts on the funnel plots over the three periods, regardless of seven-day changes, with the Heart of England NHS Foundation Trust remaining above the +3 SD control limit despite seven-day reorganisation (Fig. 2 g-i). A&E admission rates did not significantly change over the three periods for trusts that had undergone such changes (n = 72) in comparison to those that had not (n = 61; mean difference:

Discussion
The three-year lapse from the initiation of the "NHS Services, Seven Days a Week" Forum provided us with an opportunity to analyse outcomes from what was essentially a natural intervention trial involving all English NHS trusts, some of which had redesigned urgent, emergency and some elective services around seven-day working. The finding that for all outcomes studied, trusts instituting seven-day changes did not perform, on average, better than other institutions is unsurprising, particularly as NHS emergency services already operate 24 h a day, seven days a week. Our analysis also did not find any significant improvement in clinical outcomes over time in trusts which had actively reorganised services in 2013-2014 in comparison to those that had not. More worryingly, clinical outcomes such as A&E four-hour breach rates worsened across the English NHS despite approximately half of all trusts instituting seven-day changes.
In some organisations, despite active restructuring of services around seven-day care, there was a worsening or no change in mortality, length of stay and A&E outcomes. The reasons for this are probably multiple. Service reorganisation without sufficient additional investment or a cost-neutral budget could result in weekday care worsening at the expense of increasing weekend service provision so that overall outcomes through the seven-day week are worse or unchanged. Additionally, the association between poorer weekend clinical outcomes and service provision may potentially be non-causal, and therefore increasing weekend services may not result in improvement. Contrastingly, some trusts such as Blackpool Teaching Hospitals NHS Foundation Trust demonstrated reductions in SHMI without significant seven-day reorganisation.
The apparent association between weekend hospitalisation and increased mortality has been repeatedly demonstrated in several studies carried out in different countries [12,13,[21][22][23] and healthcare settings [24]. Other outcomes such as length of stay [25] and unplanned readmission rates [26] are also higher for patients admitted over the weekend. However, the definition of what constitutes a "weekday" and the methods of case-mix adjustment in these studies have been variable, [27] and it is well-recognised that the cause of this association remains unknown. Several factors have been blamed, including reduced availability of senior staff members, diagnostic and support services [3,13]. Other possibilities include the fact that sicker patients are more likely to present over the weekend and is incompletely adjusted for by various statistical models, or that publications surrounding the "weekend effect" are systematically biased to detect changes where there are none [28].
Recent evidence suggests that only 3% of total mortality is avoidable, [29] and several studies have now emerged suggesting that the "weekend effect" is unlikely to be improved by changes in the level of weekend staffing [30][31][32][33]. Our analysis supports this, supporting the assertion that the "weekend effect" is largely due to residual confounding and not care quality. Unlike previous studies which have not successfully determined its cause, [12,13,23] this analysis makes an attempt at dissecting out the effect of widespread reorganisation of clinical services around a seven-day working week on mortality, LOS and A&E quality of care, finding that these outcomes do not largely improve in trusts which have actively done so.
Despite heterogeneity in the measures taken by individual trusts to reorganise services, the changes in clinical outcomes observed between the two periods indicate these measures have also largely not improved such metrics as a whole across the English NHS. More worryingly, outcomes such as A&E four-hour breach rates were significantly worse on a background of increasing A&E attendances, suggesting increasing demands on a service where resources may have been misallocated or insufficient. We postulate therefore that large-scale changes to create a truly "seven-day NHS" such as that proposed (e.g. changes to the definition of unsociable hours working) may not lead to significant improvements, particularly in a cost-neutral setting resulting in limited weekday resources simply being moved to the weekend.

Study limitations
We recognise that our retrospective observational analysis utilising aggregated, trust-level data from publicly available national databases of institutional performance is limited by its quality and resolution, with less capacity for adjustment of confounding to account for differences in case-mix observed at individual trusts. As such, we could only analyse crude mean LOS and A&E outcome metrics at a trust level, and any differences in these outcomes between trusts are still subject to confounding from case-mix variability. Even metrics such as the SHMI, which attempts to account for this by adjusting for comorbidities at an individual patient level using HES data, do not completely adjust for factors such as the severity of a particular comorbidity or socioeconomic status [34]. Utilisation of aggregated trust-level data also means that there is a risk of over-interpretation at an individual level, a concept known as ecological fallacy. For instance, in this study we categorised trusts as having undergone seven-day service reorganisation on the basis of any increase in seven-day provision to test the hypothesis that this would lead to improvements in clinical outcomes. A patient suffering a stroke on the weekend may have benefitted (reduced mortality risk, reduced length of stay) from some forms of seven-day service reorganisation (such as a 24/7 thrombolysis service) but not others (such as extended weekend pharmacy opening times). However, at a trust level, the limited resolution of data means that determining the clinical effectiveness of specific changes in weekend service provision at an individual level is not possible. In order to do this, further detailed interventional studies need to be undertaken, looking at the effect of a specific change in service on a specific group of patients.
Additionally, individual trusts not included in the initial HFMA costing analysis during the period of interest may not have declared the presence of seven-day service reorganisation accurately via their annual reports. It is possible that some trusts may have demonstrated improved outcomes through such reorganisation that was not formally part of the "NHS Services, Seven Days a Week" project, and the lack of reporting would have diluted any comparative differences in institutional performance. However, it is far more likely that given the centrally-driven push for better seven-day care, any significant changes that could potentially lead to improvements in clinical outcome would have been declared explicitly. The heterogeneity of the various measures taken by individual trusts also means that reorganisation was not uniform; however, one would expect that such a global restructuring of services should at least have resulted in measurable change in clinical outcomes when examined across the entire English NHS. We also recognise that service reorganisation takes time, and that whilst our inclusion of the 2015-2016 analysis period somewhat mitigates for the time lag in the implementation of changes in health policy, further longitudinal analyses should be performed over consecutive years to determine their full effect.
In the pairwise difference in differences analysis of clinical outcomes, we excluded 11 trusts that had undergone significant restructuring by merging with other trusts, as it would have been difficult to determine if any changes were likely to be due to seven-day service reorganisation or the restructuring itself. However, we recognise that merging of trusts may have occurred to improve access to better seven-day services, and it is possible that clinical outcomes in these trusts could have improved as a result of the merge. We have also not examined other outcomes such as patient satisfaction in this analysis, although a similar pilot scheme to extend weekend GP services reported that demand is significantly lower than expected [35].