Response

We thank Perel et al for their thoughtful reading of our study on early blood transfusions in sepsis.¹  Perel et al correctly note that our analysis was at the hospital rather than the patient level and propose 2 additional analyses: a patient-level analysis examining the association of blood transfusions on the study outcomes and a patient-level subgroup analysis that excludes patients who received transfusions after day 2 of their hospitalization. These proposals are worthwhile extensions to our work provided that appropriate variables are used for multivariable adjustment, most notably a patient’s hemoglobin concentration over time. Unfortunately, hemoglobin data were absent from our database. Hence, we performed a hospital-level analysis, which mitigates some bias inherent in patient-level analyses.² 

A key finding of our study that would be lost in a patient-level analysis was the considerable variation in early transfusion rates at the hospital level after adjustment for many potential...