Using about five million hospital discharges from all acute NHS hospitals in England for a year (08/09), we found that after adjustment for a range of risk factors, admission on the weekend was associated with an increased risk of death, which was more pronounced in the elective setting. In the emergency setting the odds increased by 9% (Odds Ratio: 1.09) and in the elective setting the odds increased by 32% (Odds Ratio: 1.32).
Several large database studies have found an increased risk of death for weekend admissions. Aylin et al's  study of all emergency admissions to English NHS acute hospitals for the year 2005/2006 (n = 4.3 million emergency admissions), reported a 10% increase in the odds of death following weekend admissions versus weekday admission after statistical adjustment for known confounders . This effect size is consistent with our emergency admissions model despite differences in the case-mix adjustment scheme, whereby they used the Clinical Classification System (CCS)  and we adopted HRG codes, although they both ultimate rely on the same HES database. Bell and Redelmeier  in a large study (3.8 million emergency admissions in Ontario, Canada) noted that weekend emergency admissions were associated with significantly higher mortality rates for 23 of the 100 leading causes of death and were not associated with significantly lower mortality rates for any of these conditions. A more recent large database study to examine the weekend effect is that of Freemantle et al. , who also considered elective and emergency admissions, but not as separate analyses. They used time to event survival models and, unlike other studies in the field, also investigated the risks to patients of being exposed to weekend care irrespective of the day of admission. Freemantle et al.  found an increased risk of death for weekend admissions (in elective and emergency settings, Hazard Ratio 1.14 to 1.16 for weekend admissions), but paradoxically they also found that being in hospital over the weekend, irrespective of day of admission, was associated with a reduced risk of death, suggesting that the mechanisms underlying the increased risk for weekend admissions are likely to be complex. Freemantle et al.  also incorporated elective admissions into their study but did not undertake separate analyses for elective and emergency admissions and so their findings are not specific to these settings. Our finding that the weekend effect is more pronounced in the elective setting as opposed to the emergency setting appears to be novel.
Numerous smaller studies have investigated the effect of acute weekend admission on hospital mortality . For specific acute conditions or settings [3–20]. A recent review of these studies concluded that the evidence for a weekend effect was mixed - some studies found an effect whilst others did not2. Where studies have found an increased risk of death associated with weekend admission, two major explanations have been forwarded [2, 4], which are not mutually exclusive: that patients admitted over the weekend are sicker than their weekday counterparts, and/or that patients admitted over the weekend experience poorer quality of care.
Evidence to support the "sicker patients at weekends" hypothesis stems from the non-uniform daily incidence of acute conditions (eg acute myocardial infarction, stroke) reported in some studies [8, 10] but not in others . Where weekend and weekday incidence patterns differ this can lead to selection bias . One possible mechanism for differential incidence is that patients with less severe acute illness may avoid a weekend admission, choosing instead to present on a weekday. This would mean that weekend patients are sicker but this mechanism can potentially operate both ways - patients with less severe illness may choose to present on a weekday as a way of avoiding a weekend admission. We did find some evidence of differential levels of complex elderly admissions between weekend and weekday admissions in the elective and emergency setting, although we expect that the statistical model will adequately control for such differences providing the underlying relationship between patient sickness and mortality is the same for weekend and weekday admission . Whilst Becker has called for further condition specific studies as a way to minimise selection bias in the study of the weekend effect , our study along with others [2–18] suggest that there is generic dimension to the weekend effect, which according to our findings, is more marked in the elective setting. Furthermore, since the sicker patient's hypothesis relies on non-uniform daily incidence of acute conditions, this explanation is less applicable to the elective setting where admission is planned.
Direct evidence that emergency admissions admitted on the weekend may experience suboptimal care has been reported in some studies (examining specific acute conditions such acute myocardial infarction, stroke) but not others [6, 8]. Where suboptimal care has been found it was mostly due to delay in treatment . However the extent and nature of possible suboptimal care in the elective weekend versus weekday setting remains somewhat under explored, perhaps because previous work has focused primarily on emergency admissions.
Our study, like several other studies [3–20] is void of any direct measurement of care and it would be premature to presume that apparently adequate case-mix adjustment safely rules out the "sicker weekend patients" hypothesis and rules in the "quality of care" hypothesis . For instance, it is worth noting that if patients admitted over the weekend have less comprehensive clinical coding (perhaps because case-notes are less complete for weekend admissions), then such differential measurement error could potentially undermine the case-mix adjustment in our model, because the "sickness" of weekend admissions may have been systematically underestimated. Or, for emergency admissions, if provision for palliative care in the community is reduced over the weekend then hospitals will see an increased mortality over the weekend not because of poorer quality of care but because the mortality burden is displaced at the weekends. Indeed it is also worth noting that case-mix adjustment itself is not without risks of bias  and as recommended by Nicholls  we systematically checked for interaction effects between our primary covariate (weekend admission) and the other covariates in the model. Nevertheless, further studies to determine the relationship between the processes of care for weekend versus weekday admissions especially in the elective setting appear justifiable even though we know from a systematic review that the links between excess mortality and quality of care are not always reliable .
There are some issues with our approach to modelling which merit discussion. We did not combine elective and emergency admissions into the same model as Freemantle et al.  have done. Our rationale was that elective and emergency admissions are generated by different processes and the material differences in coefficients for the same covariate would appear to support. Furthermore, as recommended by Jones  and noted in another study where emergency admissions were analysed,  we excluded zero stay emergency admission that were discharged alive, from our data but note that other studies using NHS data [13–18] did not make special allowances for these zero stay admissions. During model development we found that model stability and healthy goodness-of-fit statistics were achieved after including an interaction between "withcc" (with comorbidities and/or complications), but this was less surprising because HRG codes which include a "with cc" description may also include an age criterion (e.g. aged < 69 years). We did not exclude bank holidays from the analysis, because there are only eight bank holidays in England and exclusion (or indeed special attention e.g. as a covariate) was not likely to seriously impact on our findings. We developed an adequately fitting model using an arbitrary 10% random sample, but our overall model calibration statistics were statistically significant for the emergency admissions (but not the elective admissions), presumably because of the very large sample sizes involved. Our approach to case-mix adjustment did not incorporate any specific disease groups because this was not our primary hypothesis. This does not mean that disease specific models are not warranted. Nevertheless we emphasise that since desktop based exercises, such as ours, are generally void of any direct measures of care their ability to inform efforts to improve care over the weekends is limited [2, 4]. Since the formulation of interventions to address deficiencies in care over the weekends requires painstaking detective work to determine the causal relationships between processes of care on the weekend and mortality, we would therefore prioritise this type of challenging research activity, commencing with the elective setting.