• Neil Blumberg, Hongwei Zhao, Hongkun Wang, Susan Messing, Joanna M Heal, and Gary H Lyman.
• University of Rochester Medical Center, Rochester, New York 14642, USA. Neil_Blumberg@urmc.rochester.edu
• Transfusion. 2007 Apr 1; 47 (4): 573-81.

BackgroundThe scientific method requires that only experiments actually and correctly performed be used to draw conclusions. The intention-to-treat principle requires that all patients, even those not or improperly treated, be included. In clinical trials and meta-analyses investigating leukoreduced blood transfusions to reduce postoperative infections, the intention-to-treat principle and the scientific method have been variably applied.Study Design And MethodsClinical trials and meta-analyses were retrieved from the literature, and their scientific and statistical methods were assessed. A meta-analysis emphasizing the scientific method was created, restricted to patients who actually received transfusions, given that patients who did not receive transfusions cannot benefit from leukoreduction.ResultsNine clinical trials and 11 meta-analyses were identified. In 2 of the trials and all but 1 of the meta-analyses, conclusions were based on data that included many (>10%) patients who had been randomly assigned but not received a transfusion or data not derived from investigative results. Limiting the meta-analysis to patients who actually received transfusions (n = 3093), demonstrated that leukoreduced transfusions significantly reduced the odds of postoperative infection (summary odds ratio, 0.522; 95% confidence interval, 0.332-0.821; p = 0.005).ConclusionWhen data restricted to patients receiving transfusions are analyzed, and no data absent from the actual investigations are introduced, leukoreduced transfusions substantially and significantly reduce the odds of postoperative infection by approximately 50 percent. These results demonstrate the importance of including only scientifically valid data in clinical trials and meta-analyses. The intention-to-treat principle should never lead to inclusion of data not actually derived from experimental results.

10 keywords...

hide…