Perioperative blood transfusion and cancer recurrence: meta-analysis for explanation.

BACKGROUND: Meta-analysis was used to explain disagreements across observational studies in regard to the association between perioperative transfusion and cancer recurrence. STUDY DESIGN AND METHODS: Observational studies published in English from 1982 through 1994 were retrieved. Five or more articles published in complete form were identified for each of six cancer sites: colorectum, breast, head and neck, lung, prostate, and stomach. Necessary information for building a 2 x 2 contingency table could be extracted from 60 studies. Summary relative risks (RR) reflecting the "average" adverse transfusion effect were computed for each cancer site by the random-effects method. Seven study characteristics were examined as potential explanations for the disagreements among the published studies.
RESULTS: Before any adjustment for the effect of confounding, computed crude summary RRs suggested a significant (p < 0.05) deleterious transfusion effect in all cancer sites, except for breast. The RR of an adverse outcome was 1.49 in colorectal cancer (95% CI, 1.23-1.79) and ranged from 1.06 in breast cancers to 3.62 in head and neck cancers. The disagreements among published studies were most marked in the case of colorectal and gastric cancers. These discrepancies could be explained, in part, by study design, because prospective investigations had not produced a significant unadjusted transfusion effect (RR = 1.18; 95% CI, 0.93-1.51 in the case of colorectal cancer).
CONCLUSION: A reduction in the size of the computed unadjusted transfusion effect (of an appropriate magnitude to adjust for the effect of confounding) might eliminate the significance of the average adverse effect in most studied cancer sites. Whether the entire unadjusted transfusion effect should be ascribed to the effect of confounding or whether a true, deleterious transfusion effect also exists can be resolved only by randomized controlled trials.