Wu and colleagues have conducted an elegant observational study evaluating the effect of intraoperative transfusion on 30-day mortality in elderly patients (≥ 65 years) undergoing non-cardiac surgery. Standardized, prospective data captured by a major surgical quality initiative (NSQIP) at Veterans Affairs hospitals in the United States was merged with other standardized patient data. Eight major types of surgery were evaluated including general, vascular, orthopedic and urologic. The primary analysis assessed 30-day mortality in groups of transfused and non-transfused patients stratified by preoperative hematocrit (< 24%, 24 to 29.9%, 30 to 35.9% and ≥ 36%). The investigators also assessed the impact of estimated blood loss calculated derived from an algorithm incorporating intra- and postoperative hematocrits and red cell usage. Propensity analysis was used to match patients including a number of baseline demographic and clinical characteristics, hospital, surgeon characteristics, and complexity of surgery. Close to 240,000 patients were identified from which 18,846 transfused patients were "matched" with 18,646 non-transfused patients. The odds of 30-day mortality were reduced by 40% in the transfused patients with preoperative hematocrits < 24% compared to the non-transfused, and no effect on mortality was observed for hematocrits between 24 and 29.9%. Overall, patients with hematocrits > 30% had higher odds of mortality (37 to 59%) except if intraoperative blood loss was estimated to be 500-999 mL, in which case their odds of mortality were reduced by 65% and 22%, respectively. For patients with an estimated blood loss of < 500 mL, those with hematocrits < 24% clearly benefited from intraoperative transfusion while those with hematocrits between 24 and 29.9% had a small but statistically non-significant decrease in survival. This is a well-conducted study with appropriate analysis and high-quality data. In light of the lack of rigorous clinical trials assessing transfusion triggers in the non-cardiac surgery population despite the transfusion of millions of units every year, the study by Wu and colleagues provides the next best evidence of when intraoperative transfusion may be beneficial or harmful. Indeed, randomized clinical trials are still required to definitively assess the effect of transfusion triggers as observational studies cannot escape bias introduced by confounding by indication and residual confounding of unmeasured peri-operative risk factors. – Dean Fergusson