Variations in mortality across the week following emergency admission to hospital: linked retrospective observational analyses of hospital episode data in England, 2004/5 to 2013/14

Article history

The research reported in this issue of the journal was funded by the HS&DR programme or one of its preceding programmes as project number 12/128/48. The contractual start date was in March 2014. The final report began editorial review in September 2016 and was accepted for publication in April 2017. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The HS&DR editors and production house have tried to ensure the accuracy of the authors’ report and would like to thank the reviewers for their constructive comments on the final report document. However, they do not accept liability for damages or losses arising from material published in this report.

Declared competing interests of authors

none

Copyright statement

© Queen’s Printer and Controller of HMSO 2017. This work was produced by Han et al. under the terms of a commissioning contract issued by the Secretary of State for Health. This issue may be freely reproduced for the purposes of private research and study and extracts (or indeed, the full report) may be included in professional journals provided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising. Applications for commercial reproduction should be addressed to: NIHR Journals Library, National Institute for Health Research, Evaluation, Trials and Studies Coordinating Centre, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.

2017 Queen’s Printer and Controller of HMSO

Background

There are long-standing concerns that patients admitted to hospital outside normal working hours, when staffing levels are lower and some services are not available, suffer higher complication and mortality rates than patients admitted at times when the hospital is fully operational. 1–4 Although there is evidence for this phenomenon in all out-of-hours periods – principally at night and at weekends – it has become known as the ‘weekend effect’ because the seminal studies in this area concentrated on weekend admissions.

Payers and policy-makers have responded to these findings with health-care system reform on the assumption that reduced service provision underlies the higher risk of adverse outcomes at weekends and that more consistent hospital care throughout the week will reduce or remove the disparity. 5,6 In England, evidence on the weekend effect has been used to justify controversial attempts to introduce a ‘7-day service’ across the NHS, an approach that is supported by evidence suggesting that emergency weekend admission is not associated with poorer outcomes when there is consistent access to early diagnosis and treatments. 7–9

Although the weekend effect has been described internationally, it is of particular concern to the NHS, which was founded on fundamental principles of equity. Equity and Excellence: Liberating the NHS stated that the 2010 government’s long-term vision is of a NHS that eliminates discrimination and reduces inequalities in care. 10 If patients’ outcomes are strongly dependent on the time that they are admitted to hospital, and if time of presentation is partly socially determined, this has implications for equity across the service. Extending the hours for which the NHS is fully operational has the potential to improve access and outcomes for patients. However, extending fully operational hours could also decrease efficiency by raising input costs and providing services that are relatively underutilised.

However, although the weekend effect is well described it is poorly understood. Previous studies have been limited by the availability, completeness and content of routine admissions data, which has restricted investigations to day of admission and prevented direct measurement of severity of illness. It is therefore not clear whether or not the weekend effect extends to other out-of-hours periods, or how far excess mortality for out-of-hours admissions reflects a different presenting population with higher severity of illness, and how much is explained by poorer availability and quality of services. It is also not known if service changes could bring out-of-hours mortality rates down to levels comparable to daytime and weekday rates and what the cost implications of such changes would be. The NHS therefore faces difficult decisions in how it should respond to its responsibility to provide health care 24 hours a day, 7 days a week (24/7).

Research aims

Our original intention was to explore how extending fully operational hours affected costs and patient outcomes through a case study of an acute trust – Salford Royal NHS Foundation Trust (SRFT) – that had extended its hours of full operation, culminating in the provision of consultant-led care from 07.00 to 23.00 hours (see Appendix 1). This would enable us to determine whether or not rebalancing fully operational hours would represent a cost-effective use of NHS resources.

However, preliminary analyses of data from SRFT and interviews with local clinicians indicated that the population of patients admitted out of hours differed substantially from the population admitted on weekdays in terms of severity of illness. Furthermore, it was clear that standard risk adjustment models using routinely collected administrative data were not able to adequately adjust for these differences. If this pattern were repeated in other acute trusts, it would suggest that the weekend effect is strongly determined by selection into the admitted population and that our proposed analyses of patient outcomes would be biased.

Following consultation with the Study Steering Group, our research aims were reframed to address more fundamental questions about the causes of the weekend effect and the likely cost-effectiveness of restructuring NHS services to address excess mortality for out-of-hours admissions. We expanded the research to include all acute trusts in England, while continuing to use detailed data from SRFT. We also took advantage of newly emerging sources of data, which allowed for time of admission (and, therefore, out-of-hours mortality) to be measured more precisely.

Research questions

We conducted five linked studies addressing the following research questions.

1. What are the potential costs and benefits of introducing 7-day services?

2. Does the weekend effect extend to nights? Does it persist over time?

3. Do higher mortality rates for patients admitted to hospital out of hours reflect a lower probability of admission?

4. Are higher mortality rates for out-of-hours admissions explained by greater severity of illness?

5. What is the relationship between staffing levels and out-of-hours mortality?

Although variations in provision of care, patient characteristics and patient outcomes are also evident for elective admissions to hospital, examining these variations was beyond the scope of this study and we have therefore focused on emergency admissions.

Analysis of national data

For studies 1, 3 and 4, we used a combination of national patient-level data from Hospital Episode Statistics (HES) for patients attending accident and emergency (A&E) departments and all patients admitted to hospital in an emergency, in addition to data from the Office for National Statistics (ONS) on all-cause mortality within 30 days of admission.

Analysis of data from the Salford Royal NHS Foundation Trust

For studies 2 and 5, we used inpatient and staffing data derived from SRFT. SRFT is a large teaching NHS hospital in north-west England with a team of > 6000 staff to provide a comprehensive range of emergency care, specialist treatments and some community services. SRFT has a long-term commitment to improve equality in patient access and outcomes by delivering standardised safe care 7 days a week, particularly for emergency patients. Changes in the organisation of non-elective services have been implemented over time, including consultant presence ‘on the shop floor’ from 08.00 to midnight every day, acute physicians working in the emergency assessment unit from 08.00 to 20.00 hours 7 days a week, as well as radiology services available 24/7 for all core procedures.

Patient and public involvement

To ensure that the aims of the study reflected patient priorities and outcomes of importance to patients, we regularly engaged with the public to ensure that findings were communicated to the groups most likely to be affected. Engagement was achieved through consultation with the Salford Royal NHS Foundation Trust Membership (SRFTM) and the Salford Citizen Scientist Project. The SRFTM has 14,545 patient members and 6179 staff members and provides feedback to clinicians, managers and researchers via surveys, focus groups, co-design events and public meetings. The Salford Citizen Scientist Project is a forum bringing together researchers and local residents to improve health research and health-related services.

We held regular engagement events, located at SRFT.

• At month 3, we held an open day for patients, carers and staff to discuss study aims and objectives.

• At month 15, we held two ‘Medicine for Members’ events to discuss emerging findings, priorities for members and potential changes to analyses:

  • a seminar for 60 SRFTM members with a floor debate

  • two ‘citizens’ juries’ with 15 SRFTM members, where members could put questions to two ‘expert witnesses’ (TD and RM).

• At month 18, we held a focus group with 15 SRFTM members to disseminate preliminary findings and to elicit patient and public recommendations for providers.

Findings from these PPI events are detailed in Appendix 2.

Following the first event, two patient group representatives were recruited to sit on the Advisory Panel and attended three Advisory Group meetings during the course of the research programme.

Background

The English NHS is moving towards providing comprehensive 7-day hospital services in response to higher death rates for emergency weekend admissions. However, providing the same level of services every day of the week may not be the most cost-effective way of distributing limited health-care resources. It is not yet known whether or not changing to a 7-day service will improve outcomes and what the costs of any such reorganisation will be. In this study we examine the evidence base being used to support the case for 7-day services and, using national statistics and the results of published studies, we estimate the potential benefits of introducing such service extensions across England compared with the costs of doing so.

Recent evidence from England suggests that patients’ risk-adjusted probability of dying is increased by 11% [95% confidence interval (CI) 9% to 13%] if admitted to hospital on a Saturday and 16% (95% CI 14% to 18%) if admitted on a Sunday, compared with those admitted mid-week. 3 This weekend effect varies substantially by condition, from a zero additional risk for pneumonia to 16% for stroke, 28% for lung cancer and 37% for renal failure. 3 To interpret these weekend effects it is important to consider the baseline level of risk. In the review commissioned by the NHS Services, Seven Days a Week Forum,25 only one study, by Aylin et al. ,15 presented actual mortality rates for patients admitted at the weekend and during the week. Using data from 2005/6, Aylin et al. 15 reported in-hospital mortality rates for emergency admissions of 4.9% for weekday admissions and 5.2% for weekend admissions. This represents a 10% increase in relative risk, but only a 0.3 percentage point increase in absolute risk.

One of the main differences between hospital care at weekends compared with weekdays is the reduced availability of senior clinical staff; and this is often cited as an explanation for the observed weekend effect. 7,26 However, there is a lack of evidence that increasing the levels of consultant cover at weekends leads to reductions in mortality rates. 25

Available evidence on the impact of extending service provision comes from a small number of case studies of specific care pathways. For stroke, for example, improved outcomes, reduced length of stay and favourable evidence on cost-effectiveness have been found in specialised units configured to treat patients admitted for this condition every day of the week in London,27 though later work showed that the same results were not seen in another location, Greater Manchester. 28 The death rate within 7 days of admission is reported to have fallen from 10.0% to 7.3% for those admitted at the weekend after the reorganisation of stroke services in London. 25 However, these facilities provide a range of enhanced facilities in addition to 7-day services and it is therefore difficult to identify which aspects were responsible for the observed improvements in outcomes. In the London case study, the mortality rate for patients admitted during the week was reported to have fallen from 8.0% to 6.4%,25 meaning that the relative weekend effect was reduced but not eliminated.

Funding new interventions imposes costs on health systems, reducing the resources available for existing services and potentially resulting in a net loss of health benefits. The costs of extending normal operational hours must therefore be weighed against the predicted benefits. Increasing the level of consultant cover during the weekends will require a redistribution of the existing workforce and/or additional training and recruitment. Diverting consultant cover away from weekdays towards weekends would be expected to affect the quality of services and outcomes for patients admitted during the week. The introduction of 7-day services might therefore narrow the gap between weekday and weekend mortality, but at the cost of higher weekday rates.

Results

Summary statistics for emergency hospital admissions are given in Table 1. The crude 30-day mortality rate was 3.70% for patients admitted during the week and 4.05% for those admitted during the weekend, resulting in an excess death rate of 0.35 percentage points (Table 2). If the crude mortality rate observed during the week applied to patients admitted during the weekend, this would translate into an annual estimate of 4355 excess deaths occurring nationally at weekends (see Table 2). After applying the risk-adjusted odds ratio (OR) from Freemantle et al. ,3 this figure rose to an estimated 5353 excess deaths. The risk-adjusted OR reported in Aylin et al. 15 is very similar to that obtained using our crude mortality rates and produces a similar estimate of excess weekend deaths of 4376. Depending on the figures used, this translates into a potential health gain of between 29,727 and 36,539 QALYs per year if all excess deaths were to be averted. Using the NICE threshold, the NHS should spend no more than £595–731M to achieve a health gain of this size. Using the upper bound of the 95% CI for the OR reported by Freemantle et al. 3 of 1.145 increases these gains by 14%.

These calculations represent the maximum possible gains from introducing 7-day services for three reasons. First, they represent the number of deaths that would be averted if the weekend effect were to be completely eradicated by extending services. Second, the methodology probably overestimates the potential QALY gains from an averted death as it assumes that those surviving would enjoy the same quality of life and life expectancy as the general population, conditional on their age and sex. 17 Third, our calculations represent the scenario where benefits to patients admitted at the weekend are achieved without any detrimental effect on outcomes for those admitted during the week.

Although the potential benefits of extending services appear large, they must be compared with the additional costs of doing so. The NHS Services, Seven Days a Week Forum estimated these costs for eight ‘successful Foundation Trusts with an interest in seven day services’. 29 The costs were estimated using a costing template and interviews with finance staff, managers and clinicians, followed by two workshops to agree methodology and overall findings. The cost estimates were highly variable across the trusts and included some cost savings associated with reduced length of stay and reduced readmissions where these were identified by the trusts. Caution was emphasised in generalising the results, but overall it was estimated that the costs of implementing 7-day services would be 1.5–2.0% of total hospital income, equivalent to a 5–6% increase in the cost of emergency admissions.

According to Department of Health (DH) accounts, national expenditure on hospitals was £71.3B in the financial year 2013/14. 30 Application of the estimates29 suggests that implementing 7-day services would cost between £1.07B and £1.43B per annum. This cost exceeds our estimates of the maximum amount that the NHS should spend to eradicate the weekend effect by a factor of 1.5 to 2.4, or between £339M and £831M.

Conclusion

Recent initiatives to extend normal hours of hospital operation and to provide more comprehensive 7-day services have been implemented in response to alarming statistics on the gap in mortality rates between patients admitted at the weekend and those admitted on weekdays. These statistics, however, are insufficient by themselves to justify a policy change towards extending normal hours of operation into the weekend. There is as yet no clear evidence that 7-day working will, in isolation, reduce the weekend death rate; that lower weekend mortality rates can be achieved without increasing weekday death rates; or that such reorganisation is cost-effective.

Our analysis indicates that the estimated cost of implementing 7-day services exceeds the maximum amount that NICE would recommend the NHS should be prepared to spend on eradicating the observed weekend effect. Comprehensive roll-out of 7-day services across the NHS is therefore unlikely to be a cost-effective use of resources, particularly as our estimates of potential health benefit represent the upper limit of what is achievable. Given the lack of evidence supporting the impact of service extension on patient outcomes, the benefits actually realised would likely be much lower. Furthermore, the consequences for patients admitted during the week also need to be considered, as care for these patients may deteriorate if resources are redistributed. Against that must be balanced other potential gains from a 7-day service, including improvements in morbidity, improvements in efficiency and overall improvements in mortality – rather than reductions in excess weekend mortality – as result of rebalancing of patient care across the week.

More – and more nuanced – evidence is required before a policy of providing full 7-day services can be supported. For example, our analysis considered only mortality and associated QALYs as an outcome, which are increasingly recognised as limited measures of outcomes. 31 There may be other detrimental effects on quality and outcomes for patients admitted at the weekend that improved weekend services could address. Although the policy debate to date has focused on the excess mortality rates observed for patients admitted in an emergency to hospitals during the weekend, there are likely to be wider consequences, such as the impact on elective activity currently undertaken during the week and the impact on primary and community services that are also limited at weekends. It is possible that selected service extensions – for specific specialties and at certain times of day – could prove to be cost-effective, but substantial commitments of NHS resources should not be made until these can be identified and robust evidence provided.

Background

Admission to hospital at weekends has consistently been found to be associated with poorer patient outcomes, for both emergency and non-emergency conditions. 3,32–35 However, much of the evidence on the effect of weekend admission has been limited in key respects. First, it tends to be cross-sectional and there is therefore limited information on the trends in disparities over time. Second, it is often based on routinely available data that provides the day but not the hour of admission. As a result, weekends have effectively been defined as 00.00 hours Saturday to 23.59 hours Sunday (thereby excluding Friday evenings and Monday mornings) and separate analysis of night-time admissions has not been possible. This is an important omission, as it is likely that if reduced service availability at weekends adversely impact on outcomes, there is also an increased likelihood of adverse outcomes for patients admitted at night, when services are also reduced. Some studies have utilised specialist data to address this issue; for example, finding that at night there is an increased risk of mortality following in-hospital cardiac arrest36 or coronary artery bypass grafting,37 and a reduced chance of receiving timely intervention following stroke.

For this study, we analysed emergency admissions to SRFT, a large teaching hospital in north-west England providing a complete range of acute services, including specialist tertiary care. SRFT has maintained detailed electronic patient records since 2004, including precise time of admission, and has been a pioneer of extending normal hours of operation and providing enhanced weekend services. In 2011 it opened an ‘emergency village’, providing a consultant-led enhanced A&E service 7 days a week, and also extended services such as radiology, pathology and pharmacy across the weekend. 25

Using the enhanced data available from SRFT, we examined disparities in patient outcomes associated with out-of-hours admissions across the full range of clinical specialties. We aimed to answer the following three key questions:

1. Is the weekend effect found when weekends are more appropriately defined (i.e. 19.00 hours Friday to 06.59 hours Monday)?

2. Does the weekend effect vary over time?

3. Is there a ‘night-time’ effect, with increased risk of mortality for patients admitted at night-time compared with daytime?

Results

We extracted a total of 244,639 emergency spells admitted to SRFT between April 2004 and March 2014, excluding maternity patients and children. About one-third of these (80,971 spells) were admitted during weekends and holidays. Descriptive statistics, as shown in Table 3, suggested that patients in the weekday admission group had similar characteristics to the weekend and holiday admission group, in terms of age, gender, ethnicity distribution and socioeconomic composition. Each spell-level record was associated with around six different diagnosis codes (both primary and secondary diagnoses) and 1.5 operation codes (both primary and secondary procedures) in both groups. Each patient, on average, had one emergency admission in the past year, regardless when he/she was admitted. The Charlson Comorbidity Index, measured as a continuous variable, was only marginally higher for weekend and holiday admissions (5.15 for weekend and holiday admissions vs. 5.07 for weekday admissions).

The differences in crude mortality rates between the two groups were more noticeable. The weekend and holiday admission group had a higher unadjusted risk of death for each outcome measure. Within 7 days from admission, 2.91% of patients admitted out of hours were recorded as died (both in and out of hospital), compared with 2.42% of the weekday group. The probability of death in 30 days was 11.5% higher if the admission was outside normal working hours (6.39% for weekend and holiday admissions vs. 5.72% for weekday admissions). The result is similar when taking into account only in-hospital death. There was a 0.5 percentage point difference between our admission groups, representing a 10.6% higher risk of death for out-of-hour patients.

In Table 4 we report the association between the risk of death and weekend and holiday admissions after adjusting for patient characteristics and complexity. Each mortality rate was significantly higher for this admission group when averaged across the entire study period. The risk of dying in 30 days increased by 0.4 percentage points if the emergency patients were admitted at weekends or holidays, compared with if they were admitted in normal working hours. The risk of dying in 7 days and the risk of dying in hospital were both increased by 0.3 percentage points for the focal admission group.

Consistent with the literature, patients’ age, gender and ethnic group were significant predictors for mortality. The risk of dying was increasing with each additional year of age when admitted. Male patients had higher probability of dying in 30 days than female patients. There was no significant difference between male and female patients for 7-day and in-hospital mortalities. Although white patients had a higher risk of death than Asian and black ethnicities, but had better outcome than the group containing all other ethnic backgrounds.

In general, more complex patients were associated with poorer outcomes. A 1-unit increase in the Charlson Comorbidity Index, representing a greater extent of comorbidity, would increase the risk of dying in 30 days by 0.2 percentage points when other confounders were held constant. This impact was 0.1 percentage point for the risk of dying in the first week and the risk of dying in hospital. More diagnosis codes and the inclusion of palliative care within an admission spell were found to be significant predictors for worse outcomes. However, the total number of procedures within a spell and the previous emergency admissions had varying impact across our mortality measures, although the estimated marginal effects were generally statistically significant.

We obtained a c-statistic of around 0.9 in the estimation of each mortality rate, suggesting that our model was a good prediction of our outcome measures. We examined the extent to which this weekend effect varied over time by analysing patients admitted in each financial year. The results in Table 5 show that admission at weekends and holidays were associated with a higher risk of 30-day mortality in 9 out of 10 years. However, in only 3 out of 10 years was this association statistically significant with a p-value of < 0.05. This may reflect a lack of power in the analysis, given the limited number of admissions in each year. These years include 2005/6, with a 0.7 percentage point marginal effect of out-of-hours admission, 2007/8, with a marginal effect of 0.5 percentage points, and 2012/13, with an marginal effect of 0.8 percentage points. The 7-day mortality rate was elevated for weekend and holiday admissions, at a statistical 95% confidence level in 2005/6, 2011/12 and 2012/13. A higher risk of dying in hospital was found for out-of-hours patients in 2011/12, with this impact only weakly significant at a 90% confidence level.

Applying our second definition of the exposure variable, we found evidence of inequality in the outcomes as each out-of-hours admission group had an independent and statistically significant association with higher risk of death (Table 6). Averaged across the study period, patients admitted on weekdays but after 19.00 hours were more likely to die by 0.3 percentage points than patients admitted during the day. This difference was constant across the three mortality variables. Weekend admissions during daytime had similar impact on the outcome variables compared with the same reference group. Adjusted 30-day mortality increased by 0.4 percentage points for this group, whereas both 7-day and in-hospital mortalities rose by 0.3 percentage points. Admissions to hospital overnight at weekends had the worst outcomes throughout the week. There was an additional 0.6 percentage point risk of dying in 30 days for these patients. The probability of dying in 7 days increased by 0.4 percentage points, whereas the chance of dying in hospital similarly increased by 0.5 percentage points.

Table 7 contains the results from the estimation of each financial year. There was evidence suggesting that, in most of these years, patients’ mortality was affected by which day of the week and what time of the day they were admitted. The pattern of this disparity, however, varied depending on the mortality variable as well as on the financial year under study. In general, patients’ chance of survival was more likely to be affected by the admission time and date in 2004/5, 2006/7 and 2012/13, with stronger correlation between higher mortality rates and out-of-hours admissions. However, these are not the only years that showed inequality in patient outcomes. In 2008/9, both 7-day and 30-day mortalities had a substantial increase for the weekend night admission group. In 2013/14, the highest risk of dying in 7 days and in hospital was observed for those admitted overnight on weekdays after adjusting for patient case mix.

Dividing a week into 14 short periods, the results showed that patients admitted overnight were exposed to greater risk of mortality than day admissions (Table 8). The adjusted mortality rates were the highest for admissions overnight on Saturdays and Sundays, with the marginal effects on the risk of dying in 30 days being 0.8 and 0.6 percentage points, respectively. Moreover, emergency patients admitted overnight from Tuesdays to Fridays were also affected, although to a lesser extent. Among day admissions, only Sundays were associated with a higher risk of dying in 7 days than on Wednesdays, at 90% confidence level. We did not observe a similar association for any other day of the week.

Conclusions

We found that, in addition to a weekend effect, there is a night-time effect; mortality rates are elevated for patients admitted at night and the highest risk of death is for patients admitted at night on weekends. We also found that the size of weekend and night-time effects varies over time. They are present in most years, but are not always statistically significant. However, this variation and lack of statistical significance may be attributable to a lack of power for analyses of individual years. Our finding of increased risk of mortality at night-times across the range of emergency admissions suggests that more patients may be affected by service variations across the week than estimated in previous studies based on weekend/weekday comparisons. However, as with previous studies, we cannot exclude the possibility that reduced capacity out of hours leads to the selection of a sicker patient population and that this explains increased risk of mortality for both weekend and night-time admissions. We explore the issue of patient selection in studies 3 and 4.

Background

The finding that patients admitted to hospital in an emergency at the weekend have a higher rate of mortality than those admitted during the week is now well documented. 3,15,26 However, the cause of this weekend effect is not known. The phenomenon has been attributed to reduced availability of senior clinical staff and access to investigative services in hospitals at weekends,7 but there is no causal evidence establishing this link.

Differences in the severity of illness of patients admitted to hospital at the weekend compared with during the week may not be captured fully by the case-mix variables available in administrative data sets. 38 The number of patients admitted to hospital in an emergency is markedly reduced at weekends. 15,35 This may be because the population is less likely to seek emergency care, A&E departments are less likely to admit patients and/or the limited availability of services in the community leads to fewer direct admissions to hospital. Higher death rates among the restricted number of patients who are admitted at weekends might reflect a higher average severity of illness among those who are admitted rather than excess avoidable deaths caused by poorer quality of care on admission.

In this study we perform a detailed analysis of the variation by day of the week in the number of admissions and subsequent mortality, stratifying patients by their route of access to hospital. We exploit previously underutilised data on A&E attendances to investigate whether or not higher mortality among the population of patients admitted to hospital reflects a more stringent admission threshold. We then examine the extent to which the limited availability of services in the community at weekends leads to fewer direct admissions and whether or not there is a higher mortality rate among the restricted number of patients who are admitted via this route.

Results

Conclusions

Patients admitted to hospital as emergencies at the weekend are known to have a higher rate of death than patients admitted during the week. However, we did not find any increase in mortality rates for the whole population attending A&E departments at weekends. The weekend effect was only apparent in the subset of patients who are admitted to hospital and was far stronger for patients directly admitted from the community – who were admitted in far smaller numbers at weekends – than for patients admitted via A&E. These findings raise the strong possibility that a sicker population of patients is admitted to hospital at weekends and that this selection effect is at least partly responsible for the weekend effect.

Elevated mortality rates among the population of patients admitted to hospital in an emergency at weekends are driven by a reduction in the number of patients admitted to hospital at the weekend rather than an increase in the number of deaths. There was a 7% reduction in admission numbers through A&E at weekends, which was not fully explained by the patient characteristics that we could control for (although adjusting for patient characteristics did increase the effective odds of being admitted at the weekend – see notes in Table 11). Hospital staff appear to apply a more stringent admission threshold at weekends to patients seeking emergency care through A&E. This raises the possibility that the patient population admitted at weekends is, on average, more severely ill than the population admitted on weekdays and that this increased severity is not completely captured by standard risk adjustment using administrative data.

The weekend effect is greatest among the patients directly admitted to hospital, for whom the relative risk of mortality was 21% higher at the weekend. However, the number of admissions through this route was 61% lower at weekends than on weekdays and these admissions represent just 11% [1317/(1317 + 10,526)] of total emergency admissions on a typical weekend day. The decrease in the number of direct admissions at weekends is not matched by an increase in A&E attendances or admissions, indicating that patients are not simply switching between the two routes into hospital at weekends. The concentration of the weekend effect where we see a substantial restriction in the patient flows again raises the strong possibility that it is attributable to greater unmeasured severity of illness rather than lower quality of care at admission.

There may be concern that patients who are directly admitted could experience different quality of care on arrival at hospital. A small proportion (6%) of patients attending A&E have been referred there by a GP and they are therefore part of the same patient pool as direct admissions in that they initially sought GP care. However, on arrival at A&E these patients would be expected to receive the same care as those who self-refer to A&E. In an attempt to shed further light on our findings we performed some supplementary analysis on this group of patients. We found that the flows of patients referred to A&E by a GP behaved in much the same way as the direct admissions. The number of A&E attenders referred by a GP dropped by 68% at weekends, as did the number of admissions through A&E for this patient group. Attending A&E on a weekend following GP referral was associated with a significantly higher risk-adjusted probability of mortality (OR 1.168, 95% CI 1.096 to 1.245). These findings suggest that direct admission to hospital at the weekend is not a cause of elevated mortality, but instead an indicator of an inherently different patient group. If the cause of elevated weekend mortality among direct admissions was lower quality of care on admission rather than referral of sicker patients, we would not expect to see a weekend effect among patients referred to A&E by a GP.

The central findings of our analysis are (1) that there are fewer deaths following emergency hospital admission at the weekend; (2) that the mortality rate following an A&E attendance is the same regardless of whether patients present at the weekend or on a weekday; and (3) that, conditional on their characteristics as measured in HES, patients are less likely to be admitted to hospital at the weekend. Our interpretation of these findings is that the patients who are most likely to die form a larger proportion of the admitted patient population at the weekend. Hospitals apply more selective filters for admission at the weekend, admitting cases that are, on average, more urgent. Put differently, as we do not know what the appropriate level of admission is, hospitals have the capacity to admit more low-risk patients during the week.

An alternative interpretation could be that hospitals randomly select a smaller proportion of patients for admission at the weekend and the poorer quality of care that these patients receive on admission causes them to die at a higher rate. However, to generate the same overall death rate for all A&E attenders (both admitted and not admitted), this interpretation would require not only that hospitals effectively select the additional patients they choose not to admit at weekends at random (or based on criteria uncorrelated with severity of presentation), but also that the population of patients who present at weekends have a lower underlying risk of death. These offsetting phenomena would need to be even stronger for direct admissions. We think it more plausible that admission decisions at the weekend take some account of the severity and urgency of patients’ conditions. As described by McKee, the comparison of weekend to weekday admissions is ‘confounded by indication’. 38 There are some clues on this in the characteristics of patients attending A&E, including the higher proportion of patients arriving by ambulance at the weekend (see Figure 2).

Previous studies have consistently found higher mortality rates for patients admitted at weekends, both before and after risk adjustment. Although we have also found higher mortality rates among the patients who are admitted at weekends, our study differs in two important respects. First, we widened our focus to include all patients attending A&E departments, including those not admitted, in order to avoid possible selection effects in the admitted population. Second, we assessed direct admissions and admissions via A&E separately, in order to gain a better understanding of variations in patient flows throughout the week. Using this approach we found substantial falls in the number of patients admitted to hospital in an emergency on weekends, attributable to a 61% reduction in the number of direct admission and a 5% relative reduction in the risk-adjusted probability of admission following an A&E attendance. These increased thresholds for admission at weekends are likely to have biased previous studies on weekend mortality.

Background

Concerns about the adequacy of administrative data to account fully for fluctuations in the severity of patients admitted to hospital at different times of the week have been raised,39,40 suggesting that more accurate adjustment for severity of illness could explain the weekend effect. The majority of studies identifying a weekend effect have relied on analyses of routinely collected inpatient administrative databases and have therefore adjusted for patient risk accounting for demographic characteristics, primary diagnosis and comorbidity measures based on reported ICD-10 codes. 34 Although these risk adjustment models exhibit high explanatory power in predicting mortality, they cannot account fully for severity of illness. 41 Studies based on data from specialist clinical databases and audits for specific diseases or clinical departments have used richer risk adjustment models that account for variations in severity of illness within a diagnostic and comorbidity group. These studies have found little or no significant difference in death rates by day of admission. 42 However, these studies have focused on patients with specific conditions.

In this study, we use national administrative inpatient data merged with previously underutilised data on attendances at A&E departments. These contain information on whether or not the patient arrived by ambulance, an additional indicator of severity not available in the inpatient data. 43 Despite ambulance services being available 24 hours each day of the week, the proportion of patients arriving by ambulance varies substantially across the week. We examine how the inclusion of information on arrival by ambulance in the risk adjustment model affects the observed pattern of mortality across the week.

The attendance data also enable us to use information on the time of arrival at hospital, in addition to the day of arrival. Studies have shown that patterns of outcomes differ throughout the whole week and between daytime and night-time, and that either being admitted overnight or receiving care overnight tends to be associated with worse outcomes. 39,44 This suggests that the description of the weekly variation in care quality as a weekend effect is an oversimplification. If fluctuations in staffing levels are the cause of the weekend effect, we would expect to see the patterns of patient mortality align with the typical shift patterns worked by senior medical staff. We examined numbers and rates of hospital admissions and deaths across the week for patients presenting to emergency services through two routes: (1) hospital A&E departments, which are open throughout the week; and (2) services in the community, for which availability is more restricted at weekends.

Results

Association between crude risk of mortality and day and time of attendance

Of all 3,027,946 patients admitted to hospital following attendance at A&E, and for whom the admitted patient record can be matched with a corresponding A&E record, the majority (1,907,920; 63.0%) attended A&E during daytime, whereas the remaining 1,120,026 (37.0%) attended at night (Table 14). The crude mortality rates are higher for patients attending A&E during the day than at night. This reflects the higher proportions in the highest risk quintile (as measured using the standard risk adjustment model) among patients arriving during the day than for those arriving at night (Figure 1).

TABLE 14 - Association of day and time of attendance with risk-adjusted mortality within 30 days of admission

Notes

Standard risk adjustment includes controls for age, gender, ethnicity, primary diagnosis, Elixhauser conditions, admission source, deprivation quintile, month and admitting hospital. Daytime is defined as 07.00 to 18.59 and night-time is defined as 19.00 to 06.59 the following day.

Table reproduced from Anselmi et al. 13 This is an Open Access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.

Admission	Number of admitted patients		Crude mortality rate (%)	Model
				1: standard risk adjustment		2: standard risk adjustment plus mode of arrival		3: standard risk adjustment (mode of arrival: ambulance)		4: standard risk adjustment (mode of arrival: non-ambulance)
	Total	Arrived by ambulance		OR	95% CI	OR	95% CI	OR	95% CI	OR	95% CI
Day and time
Wednesday day	273,739	157,675	3.66	Reference		Reference		Reference		Reference
Wednesday night	158,297	106,248	3.48	1.055	1.016 to 1.096	0.988	0.951 to 1.025	0.980	0.943 to 1.020	1.093	0.952 to 1.255
Thursday day	277,620	160,146	3.75	1.028	0.996 to 1.060	1.027	0.996 to 1.060	1.030	0.997 to 1.065	1.015	0.920 to 1.121
Thursday night	159,197	107,511	3.46	1.052	1.013 to 1.092	0.986	0.950 to 1.024	0.977	0.940 to 1.016	1.117	0.975 to 1.281
Friday day	278,570	159,339	3.71	1.001	0.970 to 1.033	1.007	0.976 to 1.039	1.006	0.973 to 1.041	1.015	0.921 to 1.120
Friday night	165,502	112,742	3.28	1.026	0.989 to 1.066	0.962	0.927 to 0.999	0.958	0.921 to 0.996	1.006	0.874 to 1.158
Saturday day	251,464	153,602	4.00	1.058	1.025 to 1.092	1.031	0.999 to 1.064	1.028	0.995 to 1.063	1.097	0.988 to 1.219
Saturday night	164,312	113,400	3.32	1.073	1.033 to 1.114	0.997	0.960 to 1.035	0.992	0.954 to 1.032	1.075	0.927 to 1.246
Sunday day	256,531	154,123	4.12	1.089	1.056 to 1.124	1.061	1.028 to 1.095	1.055	1.020 to 1.090	1.186	1.070 to 1.314
Sunday night	155,299	105,371	3.59	1.096	1.055 to 1.137	1.019	0.981 to 1.058	1.011	0.972 to 1.051	1.147	0.991 to 1.326
Monday day	294,526	164,155	3.72	1.010	0.979 to 1.042	1.017	0.986 to 1.049	1.014	0.981 to 1.048	1.056	0.960 to 1.161
Monday night	160,510	106,782	3.43	1.008	0.971 to 1.046	0.947	0.912 to 0.983	0.940	0.904 to 0.977	1.052	0.918 to 1.205
Tuesday day	275,470	157,148	3.68	1.004	0.973 to 1.036	1.007	0.975 to 1.039	1.008	0.975 to 1.043	0.994	0.901 to 1.097
Tuesday night	156,909	105,004	3.41	1.027	0.988 to 1.066	0.964	0.929 to 1.002	0.971	0.933 to 1.010	0.871	0.753 to 1.006
Average	216,282	133,089	3.66
Mode of arrival
Ambulance	1,863,246		5.46			Reference
Non-ambulance	1,152,854		0.78			0.339	0.330 to 0.348
Unknown	11,846		1.20			0.456	0.380 to 0.548
n	3,027,946			3,027,946		3,027,946		1,863,246		1,152,854
c-statistic				0.907		0.911		0.877		0.951

FIGURE 1.

Percentage of admissions by quintile of mortality risk and day and time of attendance. Figure reproduced from Anselmi et al. 13 This is an Open Access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.

Figure reproduced from Anselmi et al. 13 This is an Open Access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.

Over the whole week, 20.9% of daytime arrivals were in the highest risk quintile, compared with 18.5% for night-time arrivals. Daytime arrivals on Sundays contained the highest proportion of patients in the highest risk quintile (21.59%), followed by daytime arrivals on Saturdays (21.38%) and Fridays (20.89%). The proportions of patients in the highest risk quintile among those arriving at night-time were similar across the days of the week, with the highest proportions on Mondays (18.97%) and Wednesdays (18.9%) and the lowest proportion on Saturdays (17.62%).

As mortality is substantially higher for patients arriving by ambulance, and as the proportion of patients arriving by ambulance follows patterns similar to crude mortality rates, mode of arrival could be a proxy for severity of illness. The inclusion of a proxy for patient severity in the set of risk adjusters reduces the differences in risk-adjusted mortality throughout the week. After adjusting for ambulance arrival or when carrying out the analysis for the subsamples of patients arriving by ambulance or by other mode, mortality was significantly higher only for patients attending on Sunday daytime.

Association between risk of mortality and mode of arrival

The majority (1,863,246; 61.5%) of patients admitted to hospital after attendance at A&E were brought into hospital by ambulance (see Table 14). Patients admitted after being brought into A&E by ambulance were at a higher risk of death within 30 days: 27.4% of patients admitted after being brought into A&E by ambulance were in the highest risk quintile and 52.8% were in the two highest risk quintiles (Figure 2). Only 8.1% of patients who were admitted after arriving at A&E in other ways were in the highest risk quintile, whereas the majority were in the lowest risk quintile (31.9%) and over half of them (61.2%) were in the two lowest risk quintiles.

FIGURE 2.

Percentage of admissions by quintile of mortality risk and mode of arrival. Figure reproduced from Anselmi et al. 13 This is an Open Access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.

Figure reproduced from Anselmi et al. 13 This is an Open Access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.

Association between day and time of attendance and mode of arrival

The proportion of admitted patients brought into A&E by ambulance was higher at night-time (average 67.6%) than daytime (average 57.1%), with the highest proportions on Saturday night (69.0%) and Friday night (68.1%) and the lowest proportion on Monday night (66.5%) (Figure 3). During daytime, the proportion of admitted patients brought in by ambulance was substantially higher on Saturdays (61.1%) and Sundays (60.1%), compared with other days of the week.

FIGURE 3.

Percentage of admissions by mode of arrival and day and time of attendance. Figure reproduced from Anselmi et al. 13 This is an Open Access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.

Figure reproduced from Anselmi et al. 13 This is an Open Access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.

Association between risk-adjusted mortality and day and time of attendance

Using standard risk adjusters (excluding arrival mode), mortality was significantly higher than Wednesday daytime for patients arriving at A&E during the week at night-time on Wednesday (1.6 percentage points, OR 1.055, 95% CI 1.016 to 1.096) and Thursday (1.5 percentage points, OR 1.052, 95% CI 1.031 to 1.092) (model 1). Mortality increased through the weekend from Saturday daytime (1.6 percentage points, OR 1.058, 95% CI 1.025 to 1.092) to Sunday night-time (2.6 percentage points, OR 1.096, 95% CI 1.055 to 1.137). On every day, night-time arrival was associated with higher mortality. Mortality increased from Friday night-time to Sunday night-time before decreasing on Monday daytime.

Risk-adjusted mortality was substantially lower for patients arriving by ‘other’ modes (–2.4 percentage points, OR 0.339, 95% CI 0.330 to 0.348) and unknown modes (–2.0 percentage points, OR 0.456, 95% CI 0.380 to 0.548) than for patients arriving by ambulance (model 2). Accounting for arrival mode changed the estimated effects of the time of arrival on mortality. After controlling for arrival mode, Sunday daytime was the only time period associated with a higher mortality risk (1.7 percentage points, OR 1.061, 95% CI 1.028 to 1.095) than Wednesday daytime.

The patterns of mortality among patients arriving by ambulance (model 3) were similar to those for all patients after adjusting for arrival mode. Among patients arriving by ambulance and patients arriving by other modes (model 3), mortality was only significantly higher for patients arriving on Sunday daytime (2.3 percentage points, OR 1.055, 95% CI 1.020 to 1.090; and 0.1 percentage points, OR 1.186, 95% CI 1.070 to 1.314, respectively).

Figure 4 shows how the estimated effects of time and day of arrival at A&E on mortality were affected by the addition of mode of arrival to the risk adjustment models. Once the mode of arrival is included in the model (Figure 5), the increase in risk of mortality associated with admission after attendance at weekend and at night-time reduced, compared with Wednesday daytime. The trend towards increased risk of mortality after night-time arrivals reversed, with lower mortality for patients arriving overnight from Monday to Saturday. Figure 6 shows that the same pattern is observed among patients arriving by ambulance. There is more variation in mortality among patients arriving by other modes, but only Sunday daytime differs significantly from Wednesday daytime. Additional work is needed to identify the reasons for the higher mortality rates for Sunday daytime admissions. Postponement of procedures and investigations until Monday for these patients, combined with the high number of handovers of care on Monday mornings, may mean that urgent care is delayed.

FIGURE 4.

Risk-adjusted mortality within 30 days of admission by day and time of attendance (standard risk adjustment). Logistic regression models including controls for age, gender, ethnicity, diagnosis, admission method, admission source, Elixhauser conditions, deprivation score, month and trust. Figure reproduced from Anselmi et al. 13 This is an Open Access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.

Figure reproduced from Anselmi et al. 13 This is an Open Access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.

FIGURE 5.

Risk-adjusted mortality within 30 days of admission by day and time of attendance (standard risk adjustment plus mode of arrival). Logistic regression models including controls for age, gender, ethnicity, diagnosis, admission method, admission source, Elixhauser conditions, deprivation score, month, trust and mode of arrival. Figure reproduced from Anselmi et al. 13 This is an Open Access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.

Figure reproduced from Anselmi et al. 13 This is an Open Access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.

FIGURE 6.

Risk-adjusted mortality within 30 days of admission by day and time of attendance, by mode of arrival at A&E. Mode of arrival (a) ambulance; and (b) non-ambulance. Figure reproduced from Anselmi et al. 13 This is an Open Access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.

Figure reproduced from Anselmi et al. 13 This is an Open Access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.

Comparison with other studies

For comparison with previous studies, which divide the week into week and weekend, we conducted pooled analyses comparing week with weekend, day with night and weekday daytime with other broad out-of-hours periods – weekday night-time, weekend daytime, weekend night-time (Table 15). These analyses show that the weekend effect is reduced from an OR of 1.059 (95% CI 1.044 to 1.075) to 1.036 (95% CI 1.021 to 1.052) when arrival mode is added to the regression model. This suggests that arrival by ambulance partly – but not entirely – explains the weekend effect. Dividing the week into four periods (weekday, weeknight, weekend day and weekend night) shows that this is largely attributable to the weekend daytime period, which retains an elevated mortality rate even after accounting for arrival by ambulance. This is in turn attributable to elevated mortality rates for Sunday daytime. The night-time effect is reduced from an OR of 1.020 (95% CI 1.006 to 1.035) to 0.959 (95% CI 0.946 to 0.973) when arrival mode is added to the regression model, which suggests that greater severity of illness for night-time admissions may explain the higher mortality rates for these patients.

TABLE 15 - Comparison of association of day and time of attendance with mortality

Standard risk adjustment includes controls for age, gender, ethnicity, primary diagnosis, Elixhauser conditions, admission source, deprivation quintile, month and admitting hospital.

Day and time	Model
	1: standard risk adjustmenta		2: standard risk adjustmenta plus arrival mode		3: standard risk adjustmenta (patients arriving by ambulance)		4: standard risk adjustmenta (patients arriving by other mode only)
	OR	95% CI	OR	95% CI	OR	95% CI	OR	95% CI
Daytime and night-time during week and at weekend
Weekday day (base)	–	–	–	–	–	–	–	–
Weekday night	1.025	1.007 to 1.042	0.958	0.942 to 0.974	0.954	0.937 to 0.971	1.009	0.948 to 1.075
Weekend day	1.065	1.045 to 1.085	1.034	1.015 to 1.054	1.030	1.010 to 1.050	1.122	1.055 to 1.195
Weekend night	1.075	1.050 to 1.101	0.996	0.973 to 1.020	0.990	0.966 to 1.014	1.093	0.993 to 1.203
Week and weekend
Weekday (base)	–	–	–	–	–	–	–	–
Weekend	1.059	1.044 to 1.075	1.036	1.021 to 1.052	1.033	1.017 to 1.049	1.112	1.054 to 1.173
Daytime and night-time
Day (base)	–	–	–	–	–	–	–	–
Night	1.020	1.006 to 1.035	0.959	0.946 to 0.973	0.956	0.942 to 0.970	0.996	0.944 to 1.052

In study 4 we noted that no weekend effect was apparent for all attenders at A&E, but our standard model included mode of arrival and referral sources as covariates. Excluding these covariates from the regression model made little difference to the odds of mortality (see footnote b, Table 9). This suggests that selection is only important for the admitted population and makes little difference to the population attending A&E.

Conclusions

We analysed the association between risk-adjusted mortality and day and time of arrival for patients admitted to hospital after attendance at A&E. When controlling only for the risk adjusters available in inpatient administrative records as in previous studies,3,15,26 mortality was found to be significantly higher for patients admitted to hospital throughout the weekend and on Wednesday and Thursday night-time. Patients arriving at A&E by ambulance had a higher risk of mortality. Higher proportions of patients admitted overnight and at weekends arrived by ambulance.

Accounting for mode of arrival substantially altered the temporal patterns of risk-adjusted mortality. Only admissions on Sunday daytime were associated with an increased risk of mortality than Wednesday daytime. Admissions on Monday and Friday night-time were associated with significantly lower mortality than Wednesday daytime. None of the night periods was associated with higher mortality.

Accident and emergency records are known to be less complete and accurate than inpatient records. 45 We relied mainly on the better-quality data extracted from inpatient records. We used only information on arrival time and mode of arrival from the A&E record. No specific concerns about these variables have been raised by the HSCIC in its audits of the quality of A&E records. We were restricted by data availability to deaths that occur in hospital and could not examine out-of-hospital deaths. However, > 80% of all deaths within 30 days of an emergency admission occur in hospital and the proportion is higher at weekends, leading to a small bias towards finding a weekend effect.

We used the time of attendance at A&E rather than the time of admission to hospital as our index time. We conducted a sensitivity analysis using the time of departure from A&E to approximate the time of admission to hospital (Table 16). The pattern of mortality across the week was largely unaffected, but admission on Sunday night-time became associated with a higher risk of mortality (OR 1.043, CI 1.005 to 1.081). We argue that the time of attendance is more appropriate, as this is the point at which hospitals become responsible for patient care. We used a classification of daytime versus night-time that was consistent with senior doctors’ working hours. We checked that our results were robust to using an alternative definition of daytime and night-time (08.00–19.59 hours and 20.00–07.59 hours the following day), as has been used in a previous study (Table 17). 36

Previous studies have been constrained by the absence of information on patients’ time of arrival at hospital and have been forced to classify the weekend crudely by dates as 00.00 hours Saturday to 23.59 hours Sunday and have been unable to adjust for patient severity of illness. Including arrival mode in the risk adjustment accounted for most of the differences in mortality between different days and different times, leaving Sunday daytime as the only period associated with a higher risk of death. Our results are in line with previous studies which have used clinical audit data and concluded that the elevation of mortality at weekends disappears once indicators of severity are controlled for. 39,40 In contrast to these studies, however, we analysed a large national data set and considered all clinical causes of admissions. Nonetheless, arrival by ambulance is an imperfect proxy for severity of illness. It is likely that there are further aspects of severity affecting the risk of mortality and differing between patients admitted at different times of the week that remain unmeasured.

Using arrival by ambulance as a measure of severity, we have shown that much of the weekend effect identified in previous studies is instead explained by variations in patient severity. With the exception of Sunday daytime, elevated mortality among admissions at weekends reflects a higher proportion of patients arriving by ambulance.

Background

Most staffing studies focus on nurses because of their well-documented work shifts. Higher levels of nurse staffing (as measured by the number of nurses per bed or patient-to-nurse ratio) is generally associated with better quality of care and patient outcomes. 46–48 Bond and Raehl49 investigated mortality rates in relation to hospital-based clinical pharmacy services and pharmacy staffing in 885 US hospitals. After controlling for the severity of illness, they found that the presence of a number of pharmacy services was associated with lower mortality rates. Bond et al. 50 examined the impact of staffing levels of a variety of health professional groups and found lower mortality rates were associated with higher staffing levels of medical residents, registered nurses, registered pharmacists, medical technologists and total hospital personnel. Conversely, higher mortality rates were associated with higher levels of hospital administrators and licensed practical/vocational nurses per occupied bed.

There is a lack of evidence, however, on the interaction between staffing levels and out-of-hours periods, that is, whether or not reduced staffing levels is partly responsible for the weekend effect. In this study, we aimed to estimate the impact of staffing level of nurses, both qualified and unqualified, on variation in patient outcomes associated with out-of-hours admissions.

Results

Staffing patterns

Roster data contained the number of nurses, both qualified and unqualified, working in the ASU for each hour of the day and each day of the study period. In Figures 7–10 the level of nurses by daytime and night-time for each day of the week are plotted over time. Overall staffing levels increased over time, with the largest increase for qualified nurses during the day. As expected, daytimes had the highest level of qualified nurses and levels during the daytime were higher during the week than at the weekend. The lowest level of qualified nurses was observed for night periods, with no significant difference between weekdays and weekend. The pattern was broadly similar for unqualified nurses.

FIGURE 7.

Staffing level of nurses: January 2009–July 2014 (qualified, daytime).

FIGURE 8.

Staffing level of nurses: January 2009–July 2014 (qualified, night-time).

FIGURE 9.

Staffing level of nurses: January 2009–July 2014 (unqualified, daytime).

FIGURE 10.

Staffing level of nurses: January 2009–July 2014 (unqualified, night-time).

Conclusions

In this study we found no evidence that patients admitted out of hours were processed more slowly than patients admitted during normal working hours, although they tended to receive fewer diagnoses and procedures. In general, there were more overnight admissions during the week and more daytime admissions at the weekend. The highest number of admissions was during the daytime on Saturdays and Sundays, and these periods carried the highest unadjusted risk of mortality.

Risk-adjusted mortality rates were higher for all out-of-hours periods, with the risk marginally higher for weekday night admissions. This is slightly different pattern from that observed for all emergency admissions (see Table 7), for which the weekend night period carried the highest risk of mortality. The increased out-of-hours mortality risk was reduced for 7-day and in-hospital mortality after adjusting for the number of nurses present in the first hour of admission. The marginal effect was greater for qualified nurses. This suggests that the increased risk of mortality for out-of-hours stroke admissions could be partly attributable to a reduced number of qualified nurses during those periods. Our analysis was, however, restricted by a lack of data on other clinical staff and information on availability of diagnostic services, so were unable to determine whether or not other staffing factors are associated with out-of-hours mortality. It is also worth noting that we assessed relative differences in survival across the week and their relationship to staffing levels, rather than the absolute effect of staffing levels on survival.

Summary of results

Conclusions

As well as patients admitted at the weekend, patients admitted at night also experience higher mortality rates. This effect was reduced for stroke patients in a large teaching hospital when more – and more experienced – nursing staff were present during the first hour of admission. Nationally, looking at all emergency admissions, we found that excess mortality out of hours was largely explained by a sicker population of patients being selected for admission. However, mortality rates were still higher than expected on Sunday daytimes when we accounted for severity of patient illness. We also found that the estimated cost of implementing 7-day services exceeds the maximum amount that NICE would recommend the NHS should spend on eradicating excess mortality at weekends.

Through our patient engagement work, we found that patient groups expressed concerns about potential variations in quality of care throughout the week and also expressed a strong preference for having the most experienced staff available outside normal working hours. However, after examining the evidence on cost-effectiveness of extending normal hours of operation and considering the potential unintended effects of such extensions, they also expressed reservations about implementing radical changes to current working arrangements in the absence of clear evidence for patient benefit. There was also a consensus that the NHS had greater priorities than providing 7-day services and that this issue had become unnecessarily politicised.

Implications for management and practice

The move towards 7-day services has begun with local implementation in acute trusts; hospitals are required to implement new clinical standards for ‘7-day services’, which include providing emergency admissions with a thorough clinical assessment by a suitable consultant within 14 hours of arrival and timely 24-hour access to consultant-directed interventions. These moves are being supported by other initiatives, including longer nursing shifts, extended contracts for community pharmacies and the opening of GP surgeries at evenings and weekends. The costs and benefits of these initiatives are largely unknown. Extending normal hours of operation could be cost-effective if it leads to improved access and better patient outcomes, particularly if these improvements benefit groups that are traditionally underserved and suffer worse health as a result.

Current initiatives to move towards 7-day hospital services are only likely to be successful in reducing mortality if reduced availability of services in hospitals on the day of admission is the major cause of the weekend effect. We have found some evidence that nursing levels may explain some of the excess mortality for stroke patients admitted during out-of-hours periods, and the potential cost-effectiveness of increasing staffing levels for these periods and for all types of staff should therefore be investigated. However, unless there is substantial investment in training and recruitment, increasing the level of nursing and clinician cover during the weekends will require a redistribution of the existing workforce, diverting cover away from weekdays.

Our findings at the national level across all indications for admission cast significant doubt over whether or not quality and availability of services is the major driver of excess out-of-hours mortality. Patients who attend A&E on weekends are at no higher mortality risk than patients who attend A&E on weekdays. A smaller proportion of attending patients are admitted at the weekend, and this higher threshold for admission is likely to mean that patients who are admitted via A&E at the weekend have, on average, greater severity of illness than patients admitted during the week. Reduced availability of primary care services at weekends means that a much smaller number of patients are admitted to hospital via this route and these patients are also likely to have greater severity of illness than their counterparts admitted during the week. When we accounted for mode of arrival at hospital – whether or not the patient arrived by ambulance – patterns of mortality changed substantially and most out-of-hours periods no longer carried an increased risk of mortality. However, admissions on Sunday daytime were still associated with an increased risk of mortality compared with Wednesday daytime. The reasons for this warrant further investigation.

Seven-day services, as currently defined, do not extend to a full 24/7 service (i.e. the government proposes to extend normal hours of operation to weekends but not to night-times). In addition to the previously described weekend effect, we found similarly elevated mortality rates for overnight admissions. A consistent policy approach would therefore involve extensions of normal hours of operation into night-time periods. However, the same issues of selection and lack of cost-effectiveness (if reducing mortality is the primary aim) would affect such extensions.

It should be noted that our analyses focused on the hospital setting and that other parts of the patient pathway – including primary care and social care settings – will have an impact on variations in mortality across the week and have a vital role to play in a co-ordinated response to extending normal hours of service operation.

Patient and public involvement

We aimed to ensure that the aims and objectives of the study reflected patient and public priorities and engaged with the public in order to communicate findings to the groups most likely to be affected. This public engagement activity benefited the research programme by ensuring that (1) it addressed questions that matter to service users, (2) interpretations of results and findings were plausible and (3) dissemination was in appropriate formats.

Patients, medical staff and members of the public were involved in the development and dissemination stages of this research. We engaged throughout the project with the SRFTM (an advisory body of 20,000 patients and staff members); we recruited two lay members to the Project Advisory Group; we convened focus groups drawn from patients, carers and staff, at critical points of the project; and we hosted a public engagement day at SRFT (see Appendix 2). We held a final public engagement exercise in November 2016 to feed findings back to the membership.

Implications for practice

In light of our findings, we have the following recommendations.

• Caution should be exercised when interpreting trends in mortality following recent changes to weekend service provision. Extending services in hospitals and in the community at weekends may increase the number of emergency admissions, particularly for patients with less severe illness, and this could have the desired effect of securing lower hospital mortality rates. However, this would be a statistical phenomenon rather than a clinically meaningful improvement, as it would be achieved by admitting a less severely ill mix of patients rather than by reducing the absolute number of deaths.

• There are many important aspects of care quality beyond mortality. Indicators of quality, such as the time taken to receive necessary scans and procedures, have been shown to exhibit temporal variation throughout the week in condition-specific studies using clinical audit data. Temporal variations in quality indicators beyond mortality warrant further investigation.

• In addressing variations in patient outcomes across the week, a more nuanced approach, extending services for key specialties over critical periods – rather than implementing whole-system changes – is likely to be the most cost-effective.

Implications for research

In advancing the research agenda in relation to out-of-hours mortality and 7-day services we have the following recommendations.

• Most of the debate on 7-day services has focused on weekend care. A more granular approach to time classification is needed to fully capture the variations in the flow of patients and the services offered across times of the day in addition to days of the week (see studies 2 and 4). In particular, emergency admissions on Sunday daytime appear to have an increased risk of mortality that is not fully explained by the patient characteristics we could control for.

• In addition to the patient’s mode of arrival at hospital (see study 4), other information on the services that patients have accessed prior to their arrival at hospital should be included in future studies to further understand mortality risk. Such proxies will, however, always be inferior to direct measurement of severity of illness, which is required in order for clinicians and policy-makers to make valid comparisons of outcomes over time and across providers. Our results, therefore, add to the increasing body of evidence highlighting the limitations of the risk adjusters available in inpatient records when standardised mortality rates are interpreted as an indicator of hospital quality.

• Although we have found that mortality is not significantly increased for patients who are admitted during most out-of-hours periods once selection effects are accounted for (see studies 3 and 4), it is possible that risks are elevated for patients who are already in hospital and who deteriorate or require critical interventions during out-of-hours periods. Further research on critical periods other than the time of admission is required.

• In order to improve understanding of the underlying reasons behind variations in mortality across the week, more detailed analysis of mortality by cause and by patient group are required.

• Fewer than 4% of patients admitted to hospital die during their admission and most of these deaths are unavoidable. Focusing on mortality, therefore, tells us little about the care that most patients receive and other indicators of quality of care should be included in future studies of out-of-hours care.

• In light of our finding in study 1 on the likely failure of service-wide extensions to be cost-effective, future research should aim to identify candidate specialties and critical periods for which extending services is likely to be cost-effective.

Contributions of authors

Lu Han (Research Associate, Health Services Research), co-investigator, was responsible for data extraction and preparation, study design, statistical analysis, and preparation of reports and manuscripts.

Rachel Meacock (Research Associate, Health Economics), co-investigator, was responsible for data extraction and preparation, study design, statistical analysis, and preparation of reports and manuscripts.

Laura Anselmi (Research Associate, Health Economics), co-investigator, was responsible for data extraction and preparation, study design, statistical analysis, and preparation of reports and manuscripts.

Søren R Kristensen (Research Associate, Health Economics), co-investigator, was responsible for data extraction and preparation, study design, statistical analysis, and preparation of reports and manuscripts.

Matt Sutton (Professor of Health Economics) co-investigator, was responsible for design of the study and editing of reports and manuscripts.

Tim Doran (Professor of Health Policy), principal investigator, was responsible for overall management of the project, design of the study and editing of reports and manuscripts.

Stuart Clough (Programme Manager, SRFT) was responsible for securing data access and liaising with clinical and managerial staff.

Maxine Power (Director of Innovation and Improvement Science, SRFT) was responsible for project management at SRFT.

Publications

Meacock R, Doran T, Sutton M. What are the costs and benefits of providing comprehensive seven-day services for emergency hospital admissions? Health Econ 2015;24:907–12.

Meacock R, Anselmi L, Kristensen S, Doran T, Sutton M. Higher mortality rates amongst emergency patients admitted to hospital at weekends reflect a lower probability of admission. J Health Serv Res Pol 2016;22:12–19.

Anselmi L, Meacock R, Kristensen SR, Doran T, Sutton M. Arrival by ambulance explains variation in mortality by time of admission: retrospective study of admissions to hospital following emergency department attendance in England. Br Med J Qual Saf 2016;26:613–21.

Han L, Sutton M, Clough S, Warner R, Doran T. The impact of out-of-hours admission on patient mortality: longitudinal analysis in a tertiary acute hospital [published online ahead of print September 28 2017]. BMJ Qual Saf 2017.

Data sharing statement

Requests for access to data should be addressed to the corresponding author.

Disclaimers

This report presents independent research funded by the National Institute for Health Research (NIHR). The views and opinions expressed by authors in this publication are those of the authors and do not necessarily reflect those of the NHS, the NIHR, NETSCC, the HS&DR programme or the Department of Health. If there are verbatim quotations included in this publication the views and opinions expressed by the interviewees are those of the interviewees and do not necessarily reflect those of the authors, those of the NHS, the NIHR, NETSCC, the HS&DR programme or the Department of Health.

Original rationale

There are long-standing concerns that patients admitted to hospital at night and at weekends, when staffing levels are lower and some services are not available, suffer higher complication and mortality rates than patients admitted at times when the hospital is fully operational. It is not known, however, what service changes would be required to bring these rates down to levels comparable with daytime and weekday rates, and what the cost implications of such changes would be. The NHS therefore faces difficult decisions in how it should respond to its responsibility to provide health care 24/7. These decisions include the external margin (e.g. when services are open to patients) and the internal margin (e.g. when facilities are fully operational).

Extending availability increases overall costs and, at a time of resource constraint, these investment decisions must recognise the opportunity costs. 24/7 care could be cost-effective if it leads to improved access and better patient outcomes, particularly if these improvements are experienced by population and patient groups that are traditionally underserved and suffer worse health as a consequence. The evidence on equity in service use suggests that more deprived populations make greater use of unplanned services and less use of planned services. It is therefore feasible that extending opening hours differentially benefits the poor and may contribute towards reducing inequalities in health.

National Institute for Health Research patient and public involvement event: Medicine for Members, 14 July 2015

Medicine for Members: first session – 14.00–16.00, 21 July 2015

Medicine for Members: second session – 14.00–16.00, 15 November 2015