We read with interest the study involving more than 250000 patients in 364 hospitals by Raghunathan et al, 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, Raghunathan et al 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) 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...

You do not currently have access to this content.