Rigorously Investigating the ‘Weekend Effect’ on Short-Term Mortality for Patients Admitted With Stroke
To the Editor:
We were very interested to read the article by Saposnik et al1 regarding the subject of a ‘weekend effect’ on short-term mortality for stroke patients. Although we appreciate the importance of this topic, we have a few comments/questions for the authors. First, Saposnik et al use predetermined cut-offs2 to define ‘weekend.’ There is no justification as to why this dichotomization of the week should be predictive of patient death. In our opinion, collapsing the available information regarding the day of the week and time of day (perhaps hospitals are understaffed and resources are limited during late night hours) into 2 groups (‘weekday’ and ‘weekend’) might potentially misclassify the study’s “exposure” of interest. It seems that a more unrestricted assessment of the ‘weekend effect’ could have been obtained by modeling separate terms for day of the week and time of day and an interaction between the 2. This would allow for a better assessment of exactly when any ‘weekend effect’ might occur. Secondly, in the statistical analysis section, the authors state that “the presence of potential interactions between age and sex, hospital type (teaching status, location), and intensive care unit (ICU) admission were tested by adding interaction terms to the regression model.” No other interactions are mentioned—in particular, no interactions between the ‘weekend effect’ and any other factor are mentioned. In Table 3, however, the effect of ‘weekend’ differs with the levels of facility type, hospital location, most responsible provider, and admission to ICU. How is this effect modification possible if no interactions with the ‘weekend effect’ were included in the model? Finally, an (unadjusted) Kaplan-Meier plot is presented to illustrate the difference in time to death between weekend (as defined by the authors) and weekday (also as defined) admissions. As the authors have noted elsewhere in the article, however, the ‘weekend effect’ is modified by the inclusion of other factors listed in Table 1. (We leave it to others to decide if this list of factors is sufficient to adjust for any differences between ‘weekend’ and ‘weekday’ with regards to patient mortality.) The authors state that “patients admitted on weekends had 13% higher odds of dying compared with patients admitted during weekdays,’ whereas the unadjusted ‘weekend effect’ in Table 2 yields an odds ratio of 1.17 (95% CI, 1.06 to 1.29). Indeed, it is on the adjusted odds ratio that the authors base their conclusions regarding the effect of ‘weekend’ on risk of death in their discussion section. Thus, the reporting of an unadjusted curve, along with the probability value for the accompanying log-rank statistic, is not appropriate. An adjusted regression model (with an accompanying plot based on this model) would have been more correct for this analysis.