We read with interest the study involving more than 250000 patients in 364 hospitals by Raghunathan et al,1 who concluded that “early” transfusions during hospitalization for sepsis without shock were associated with increased costs but no difference in mortality rates. To address the possibility that sicker patients were more likely to receive transfusion,2 Raghunathan et al1 conducted their analyses at the hospital level.Although the regression models achieved acceptable predictive values, our concern remains because many of the variables included (Supplement 2)1 may lack plausible clinical significance and many other variables that are predictors of transfusion (eg, hemoglobin concentration) are missing. Moreover, the main conclusions of the study are derived from assessing correlations between the standardized transfusion rates and risk-adjusted mortality or cost at the hospital level, making the effective sample size 364 hospitals, rather than the 256 396 patients. The authors did not address the sample size and power of the study, leaving concerns about the potential risk of type II error. Comparison of the mortality rates between the patients who received early transfusions versus those who did not yields an unadjusted odds ratio of 1.677 (95% CI, 1.617–1.740, P <.001), favoring the latter group by a highly significant margin. Given the large sample size and available data, we think that patient-level analysis could have offered additional valuable insights.The patients who received early transfusions represented only just more than 40% of all patients receiving a transfusion, and they received less than one-third of all transfused blood in the study population.1 The outcome measures (mortality and costs) span the whole length of hospitalization, and transfusions happening beyond the first 2 days can also affect these outcomes.3 This problem could have been addressed by a subgroup analysis excluding patients who received any transfusions beyond the first 2 days.The limitations should not undermine important findings of this study, including one that we emphasize here. Raghunathan et al thoughtfully included the proportion of patients who had received 5 L or more of crystalloid solution by day 2 in their model, suggesting that “large volumes of crystalloid solution may result in hemodilution and consequent RBC transfusion.”1 According to their data, 20.5% of the patients who did not receive a transfusion versus 26.4% of the patients who did get a transfusion received 5 L or more of crystalloids (odds ratio, 0.719; 95%CI, 0.694–0.744; P <.001).1 A highly probable explanation for this significant association in nonbleeding cases in this study is a more pronounced decrease in hemoglobin concentration due to iatrogenic hemodilution, resulting in transfusions.4Significant hemodilution has been repeatedly reported to occur after fluid administration in patients with sepsis.5,6 Recently, an independent, albeit weak association between hemoglobin concentration and the volume of fluids administered during resuscitation from septic shock was observed by Maiden et al7 in a post hoc analysis of the data from the ARISE (Australasian Resuscitation in Sepsis Evaluation) trial. Although the volume infused accounted for less than 20% of the observed decrease in the hemoglobin concentration,7 we believe that the relationship between the amount of fluids infused and the observed decrease in hemoglobin concentration in the first 6 hours of the ARISE trial may be much stronger,8 particularly because the fluids that were administered before the enrollment in the ARISE trial (about 25 mL/kg on average) were not taken into consideration by Maiden et al.7One of the important messages from the study by Raghunathan et al1 is that patients with sepsis who receive more fluids as part of their resuscitation will experience iatrogenic decrease in hemoglobin concentration, leading to more transfusions. It is important therefore to note that in the absence of active bleeding, a consistent decrease in hemoglobin concentration in the context of fluid administration should be considered as a sign of iatrogenic hemodilution and should be taken into account before administering transfusions.