Twenty RCTs (2,430 participants) were included.

Generally the methodological quality of the included studies was poor. Only 42% reported double-blinding, 21% had adequate allocation concealment, 21% reported the randomisation method and 26% had complete follow-up.

TXA versus aprotinin (10 trials).

TXA was inferior to aprotinin for reducing 24-hour blood loss (WMD 106 mL, 95% CI: 37, 176; 10 trials). However, there was no difference between TXA and aprotinin in the number of units of blood transfused (WMD 0.06 units, 95% CI: -0.18, 0.31; 5 trials). There was no difference in the number of people requiring transfusion using conventional meta-analysis (RR 1.08, 95% CI: 0.88, 1.32; 9 trials) or Bayesian methods (RR 1.11, 95% Bayesian CI: 0.92, 1.45), nor was there evidence of a difference between treatments for reoperation for bleeding using either conventional or Bayesian methods. For RBC transfusion, the estimated probability that TXA was non-inferior to aprotinin was 0.82 when using an RR threshold of 1.2 and 0.57 when using an RR threshold of 1.1. For reoperation the probabilities of non-inferiority were 0.92 and 0.90 when using non-inferiority limits of 20% and 10% respectively.

EACA versus aprotinin (6 trials).

EACA was inferior to aprotinin for reducing 24-hour blood loss (WMD 185 mL, 95% CI: 134, 235; 4 trials). However, there was no difference in the number of units of blood transfused (WMD -0.22 units, 95% CI: -0.52, 0.09). There was also no difference between treatments in transfusion rates using conventional meta-analysis (RR 1.14, 95% CI: 0.84, 1.55) or Bayesian meta-analysis (RR 1.08, 95% Bayesian CI: 0.73, 1.52). For RBC transfusion, the estimated probability that EACA was non-inferior to aprotinin was 0.76 when using an RR threshold of 1.2 and 0.54 when using an RR threshold of 1.1.

Data for the clinical outcomes were sparse, but the authors reported that there were no trends favouring any of the treatments.