• E C Vamvakas.
• Department of Pathology and Laboratory Medicine, Cedars-Sinai Medical Center, Los Angeles, CA, USA. vamvakase@cshs.org
• Vox Sang. 2008 Oct 1;95(3):165-73.

AbstractArticles on the appropriateness of intention-to-treat (ITT) vs. as-treated (AT) analyses of randomized controlled trials (RCTs) have highlighted issues relevant to drug trials, such as approaches suitable for 'explanatory' vs. 'pragmatic' RCTs. These considerations are less relevant to transfusion medicine RCTs, especially those of red blood cell transfusion therapies where the main issue is whether to include in the analysis randomized patients who did not receive transfusion. This article discusses issues pertinent specifically to transfusion medicine RCTs, and the thesis presented here is that the primary analysis of any transfusion medicine RCT must be based on the ITT principle. This is because the rationale for randomization is to avoid selection bias and to generate experimental arms that are comparable with respect to all, known and unknown, confounding factors. Only the entire arms produced by random allocation of subjects are so comparable, and only an ITT analysis of these entire arms fulfills the purpose of randomization. Withdrawals of randomized patients compromise the comparability of the groups and the rationale for randomization. Deviations from the ITT principle may be valid only when other conditions are met to ensure that non-adherence to ITT will not bias the results. For RCTs of red blood cell transfusion therapies, such conditions include that the RCTs be double-blind and that transfusion criteria should be applied consistently. Nonetheless, the rationale for ITT can be reversed in equivalence and non-inferiority trials where the finding of no difference is the objective of the research; thus, both ITT and AT analyses should be presented in these settings.

4 keywords...

hide…