The weekend effect: now you see it, now you don’t

Those who yearn for a world in which policies are based on a systematic and dispassionate assessment of the evidence can find little encouragement in the ongoing debate about the safety of hospitals at weekends. Almost nothing is clear in this tangled tale. It began when the health secretary, Jeremy Hunt, claimed on the BBC Today programme that there were 6000 avoidable deaths each year and a lack of weekend cover by consultants was a key factor. Yet the evidence to support this claim was elusive. Sometimes the Department of Health pointed to a 2012 paper on hospital mortality. Other times, and contrary to the government’s code of practice on use of statistics, they mentioned a (then) yet to be published paper in The BMJ. The problem was that, while both did identify an increase in deaths among those admitted at weekends, neither attributed it to a shortage of medical staff. Both identified numerous possible explanations, including various data artefacts. Indeed, the second paper stated explicitly that “to assume that [these deaths] are avoidable would be rash and misleading.” Yet, notwithstanding this considerable uncertainty, the government sought major changes in hospital staffing, somehow shifting its attention away from consultants to doctors in training. In an unprecedented move, the editor of this journal wrote to the health secretary asking him to desist from further misleading claims based on The BMJ paper by Freemantle and colleagues. The subsequent government decision to impose a new contract was mired in confusion. The only good thing to have come out of this process is that it has stimulated a series of studies that seek to resolve the uncertainties identified by the authors of the initial papers and in accompanying commentaries. In a linked paper, Li and Rothwell (doi:10.1136/bmj.i2648) used data from a population based stroke register to evaluate the quality of administrative data on patients admitted to hospital with stroke. The choice of stroke is appropriate because it is a condition for which specialist management in the acute stage can considerably improve survival. Although few will be surprised, the authors identify substantial problems with the data. Only three quarters of new strokes could be identified from the administrative data, and more than a third of episodes were incorrectly coded as admissions for acute stroke. Crucially, many people with apparent new strokes admitted during the week were actually patients undergoing investigations or procedures related to earlier strokes; these patients were much less likely to die. An analysis limited to patients with genuine new strokes found no weekend effect. This study provides support for two of the alternative explanations for the weekend effect, data artefact and case mix. Three other recent studies have filled other gaps. Aldridge and colleagues examined the work of consultants at weekends. Again, using crude data, they found a increase in mortality at the weekend but, while noting several limitations such as a low response rate among consultants, were unable to show any association between the intensity of consultant input to patient care and mortality. Bray and colleagues also studied stroke outcomes using a clinical database that overcame many limitations of administrative data. Using sophisticated adjustment for case mix, they found no weekend increase in mortality but did find complex variation in the use of investigations and treatment, with patients admitted on weekday nights faring worst. Finally, Meacock and colleagues examined the important question of whether the threshold for admitting patients is higher at weekends, finding that it is. As suspected, patients getting over this higher weekend threshold are sicker and more likely to die. Once again, the weekend effect disappears after appropriate adjustment. Collectively, these studies answer some of the outstanding questions. They show that at least part of the weekend effect is data artefact and, consistent with evidence that was available when the health secretary made his initial statement, any remaining association betweenweekend admission andmortality does not seem to be due to hospital medical staffing. However, they also raise other questions. The reported weaknesses of administrative data cast further doubt on the use of measures such as hospital standardised mortality rates. And to the extent that a weekend effect does exist, what is the appropriate response? The available evidence points to a need for