Transfusion strategy trials excluding patients transfused outside the trial study period are more likely to report a trend favoring restrictive strategies: a meta-analysis

Objectives: Some large, randomized trials investigating red cell transfusion strategies have significant numbers of transfusions administered outside the trial study period. We sought to investigate the potential impact of this methodological issue. Study Design and Setting: Meta-analysis of randomized controlled trials (RCTs) comparing liberal vs restrictive transfusion strategies in cardiac surgery and acute myocardial infarction patients. The outcome of interest was 30-day or in-hospital mortality. Results: In cardiac surgery, the pooled risk ratio for mortality was 0.83 (95% confidence interval [CI] 0.62-1.12, P = .22) times lower in the restrictive group when compared to the liberal group in trials applying a transfusion strategy throughout the patient's entire perioperative period, and 1.33 (95% CI 0.84-2.11, P = .22) times higher in the restrictive group in trials not applying transfusion strategies throughout the entire perioperative period. When combined, the risk ratio for mortality was 0.98 (95% CI 0.73-1.32, P = .89). In patients with acute myocardial infarction, the risk ratio for mortality was 0.72 (95% CI 0.40-1.28, P = .26) times lower in the restrictive group when compared to the liberal group in 1 trial excluding patients administered the intervention prerandomization and 1.19 (95% CI 0.96-1.47, P = .11) times higher in the restrictive group in 1 trial including patients receiving the intervention prerandomization. When combined the risk ratio for mortality was 1.00 (0.62-1.59, P = .99). Conclusion: Though not statistically significant, there was a consistent difference in trends between RCTs administering significant numbers of transfusion outside the trial study period compared to those that did not. The implications of our results may extend to RCTs in other settings that ignore if and how frequently an investigated therapy is administered outside the trial window.