“Lies, damned lies, and statistics” is a well known one-liner coined by Mark Twain to describe the fact that it is possible to substantiate several hypotheses with the same data set, as long as the “appropriate” statistical approach is used. In the study of Shawn and colleagues on the effects of blood transfusions on outcome, the results appear to be clear but, in fact, the protocol of the study was never designed to answer the question that is answered in the end. In their conclusion, the authors state that the wide use of blood products shows no discernible benefits. However, this is not what their protocol had been designed for. What they did, in fact, was demonstrate that patients that are transfused during cardiac surgery not only had a higher mortality rate, but also had lower preoperative hemoglobin levels. As a consequence, the statistical approach does not allow to infer from the data that the differences in mortality are due to transfusion, since the differences observed might also have been secondary to a systemic disease accompanied by anemia. However, the propensity matching used does not account for this phenomenon.
– Jens Meier