Why have meta-analyses of randomized controlled trials of the association between non-white-blood-cell-reduced allogeneic blood transfusion and postoperative infection produced discordant results?

Intention-to-treat analyses of randomized controlled trials (RCTs) of the association between non-white-blood-cell (WBC)-reduced allogeneic blood transfusion (ABT) and postoperative infection were reported as the reason why meta-analyses of RCTs of this association have produced discordant results. We examined three possible reasons for disagreements between meta-analyses: (i) sources of medical heterogeneity and integration of RCTs despite extreme heterogeneity; (ii) reliance on as-treated (vs. intention-to-treat) comparisons; and (iii) inclusion (or not) of the three most recent RCTs. When nine RCTs reported up to 2002 were combined despite extreme heterogeneity, both intention-to-treat and as-treated comparisons found an association between non-WBC-reduced ABT and postoperative infection [summary odds ratio (OR) = 1.38, 95% confidence interval (CI) 1.03-1.85, P < 0.05; and summary OR = 1.56, 95% CI 1.06-2.31, P < 0.05, respectively]. When 12 RCTs reported up to 2005 were integrated despite extreme heterogeneity, both intention-to-treat and as-treated comparisons found no association of non-WBC-reduced ABT with postoperative infection (summary OR = 1.24, 95% CI 0.98-1.56, P > 0.05; and summary OR = 1.31, 95% CI 0.98-1.75, P > 0.05, respectively). In both analyses, the separate integration of four RCTs transfusing red blood cells (RBCs) or whole blood filtered after storage showed an association between non-WBC-reduced ABT and postoperative infection, whereas the separate integration of six (or nine) RCTs, reported through 2002 or 2005, and transfusing prestorage-filtered RBCs showed no association, whether intention-to-treat or as-treated comparisons were used. Thus, the published meta-analyses have produced discordant results because they did (or did not) investigate medical sources of heterogeneity and did (or did not) include the most recent RCTs. Intention-to-treat and as-treated comparisons produced concordant results.