Data linkage studies can help to explain the weekend effect

www.thelancet.com Published online May 9, 2017 http://dx.doi.org/10.1016/S0140-6736(17)31195-9 1 Is the weekend effect a worrying manifestation of systematic deficiencies in the quality and safety of health care, or instead simply an artefact, constructed from confounding, bias, and the imperfections of existing methods for making sense of observational data? This question has not only motivated a huge number of research studies, but also entered the arena of politics and policy, contributing in England to the first industrial protest by junior doctors in 40 years. Although the consequences are hard to measure, popular awareness of the weekend effect could plausibly have fostered anxiety in the minds of patients and families about the safety of seeking health care at the weekend. Although most studies agree that mortality rates are higher for patients admitted to hospital at the weekend than during the week, the extent to which this effect reflects the care provided versus the characteristics of the patients admitted has been unclear. Studies have largely used administrative data gleaned from discharge coding, which contain limited information on illness severity. The study by A Sarah Walker and colleagues in The Lancet provides new insights by drawing on a large database of linked administrative and laboratory data. When the researchers limited their analysis to variables from administrative data (such as diagnosis codes), relative mortality risk was similar to that previously found: 8–9% higher at the weekend. However, the results of common biochemical and haematological tests explained 33% (95% CI 21–70) of the excess mortality on Saturdays and 52% (lower 95% CI 34) on Sundays, significantly reducing the size of the weekend effect when those variables were included. Striking differences were found in the blood test results between weekends and weekdays, with proportionately fewer patients with normal results being admitted at the weekend. This result suggests that part of the explanation for the weekend effect is proportionally fewer admissions for minor illness at the weekend, resulting in a cohort of patients with higher average risk of mortality compared with the weekday cohort. That said, even after adjusting for laboratory data, risk of mortality remained 18% higher for patients admitted between 1100 h and 1500 h on weekends. Further investigation is warranted, especially because another study has recently found similar levels of raised mortality on weekday afternoons. The authors modelled the relationship between mortality and hospital workload (eg, admission rates and bed occupancy). Their reasoning was that if the weekend effect is linked to reduced staffing levels or access to services, then the weekend effect should be diminished when controlling for busy days—eg, when the hospital is fuller than average. In fact, controlling for workload did not affect excess mortality following weekend admission. Although this result is intriguing, workload data are difficult to substitute for direct data on staffing levels and, in our view, stronger evidence is available from elsewhere regarding weekend mortality and staffing levels. Walker and colleagues’ contribution might be better located within the growing literature on hospital workload. At a time when occupancy rates are running at around 95% for general and acute beds in the English National Health Service, more studies should follow their approach and consider the implications of busy hospitals for quality and safety. Based on this study and other evidence, it now seems likely that many of the highest profile studies of the weekend effect overstated the problem by relying too heavily on administrative data. With this consideration in mind, it is striking that the oldest laboratory data used in this latest study were collected over a decade ago, accumulating steadily within the electronic depths of clinical databases. In other words, the data used by Walker and colleagues predate some of the most controversial studies into the weekend effect, and Data linkage studies can help to explain the weekend effect