LAUSR

Limitations Concerning the Association of Physician Sex and Patient Outcomes To the Editor In their Original Investigation in a recent issue of JAMA Internal Medicine, Tsugawa et al1 reported on an association between physician sex and patient outcomes, based on an analysis of very large administrative and social networking data sets. An important limitation of this study is the method used to link physicians with hospitalization events, which was based on attributing the care episode to the doctor accounting for the majority of that episode’s cost. This attribution assumes that the amount of care provided is directly proportional to spending. Given the importance of care transitions,2 one could argue that the hospitalization should instead be attributed to the physician in charge at the time of admission or even more so at the time of discharge, who may have a greater impact on 30-day outcomes. On average, the physician to whom care was attributed was responsible for just 51.1% of spending, leaving open the question of whether they were responsible at the time of either admission or discharge. This slim majority of spending also suggests that in cases where only one other physician was involved in the hospitalization, care would have been almost evenly shared. While secondary analyses of big data can prove useful in hypothesis-generating research, the risk of identifying spurious correlations must be carefully considered.3 With very large sample sizes like the one used in this study, statistical tests will often identify significant differences.4 This may magnify the potential impact of any particular feature of the study design, including the method for attributing physicians to hospitalizations used in this case. The use of a negative control, such as the association between physician handedness and patient outcomes at 30 days, may have increased the confidence that the sex difference identified in this study is both true, and of an important magnitude.