A total of 68 RCTs (n=4,539) and 75 cohort studies (n=30,326) were included in the review.

PAD.

Eight RCTs (n=1,191) were identified; their methodological quality was generally poor. PAD reduced the probability of receiving an allogeneic blood transfusion by 63% (8 RCTs; RR 0.37, 95% confidence interval, CI: 0.26, 0.54; Q statistic, p=0.002), but increased the probability of any transfusion by 29% (7 RCTs; RR 1.29, 95% CI: 1.12, 1.48; Q statistic, p=0.005). PAD did not appear to influence the probability of infection (3 RCTs; RR 0.70, 95% CI: 0.34, 1.43) or thrombosis (2 RCTs; RR 0.82, 95% CI: 0.21, 3.13).

Forty-two cohort studies (n=24,473) were identified. PAD reduced the probability of receiving an allogeneic blood transfusion by 69% (42 cohort studies; RR 0.31, 95% CI: 0.27, 0.35; Q statistic, p<0.001), but increased the probability of any transfusion by 91% (35 cohort studies; RR 1.29, 95% CI: 1.60, 2.28; Q statistic, p<0.001). PAD did not appear to influence LOS (6 cohort studies; WMD -0.74, 95% CI: -1.86, 0.39) or infection rates (2 cohort studies; RR 2.87, 95% CI: 0.72, 11.46).

ANH.

Thirty RCTs (n=1,295) were identified; their methodological quality was poor. ANH reduced the probability of receiving an allogeneic blood transfusion by 31% (25 RCTs; RR 0.69, 95% CI: 0.56, 0.84; Q statistic, p<0.001) and reduced the volume of allogeneic blood used per person by 1.9 units (17 RCTs; WMD 1.9, 95% CI: 1.1, 2.7). ANH did not appear to influence LOS (3 RCTs; WMD 0.21, 95% CI: -1.26, 1.68), mortality (8 RCTs; RR 1.16, 95% CI: 0.19, 7.15), infection (2 RCTs; RR 4.94, 95% CI: 0.61, 40.19) or nonfatal myocardial infarction (3 RCTs; RR 3.43, 95% CI: 0.15, 19.74).

Seven cohort studies (n=355) were identified. ANH reduced the probability of receiving an allogeneic blood transfusion by 55% (7 cohort studies; RR 0.45, 95% CI: 0.29, 0.70) and reduced the volume of allogeneic blood used per person by 2.8 units (7 cohort studies; WMD 2.8, 95% CI: 1.7, 4.0).

CS.

Thirty RCTs (n=2,125) were identified; their methodological quality was poor. CS reduced the probability of receiving an allogeneic blood transfusion by 42% (26 RCTs; RR 0.58, 95% CI: 0.47, 0.73; Q statistic, p<0.001) and reduced the volume of allogeneic blood used per person by 0.91 units (17 RCTs; WMD 0.91, 95% CI: 0.51, 1.31; Q statistic, p<0.001). CS did not appear to influence LOS (5 RCTs; WMD -1.28, 95% CI: -2.65, 0.08), mortality (11 RCTs; RR 1.53, 95% CI: 0.65, 3.61), infection (9 RCTs; RR 0.75, 95% CI: 0.41, 1.37) or nonfatal myocardial infarction (5 RCTs; RR 0.58, 95% CI: 0.28, 1.19).

Twenty-six cohort studies were identified. CS reduced the probability of receiving an allogeneic blood transfusion by 43% (26 cohort studies; RR 0.57, 95% CI: 0.46, 0.69) and reduced the volume of allogeneic blood used per person by 0.96 units (15 RCTs; WMD 0.96, 95% CI: 0.47, 1.461).

Other outcomes were reported in the paper.