Eight prospective randomised trials (n=514). None of the trials were blinded.
The average study quality score was 7.5 with an intra-class correlation of 0.41 (P<0.0001).
The Mantel-Haenszel log OR, which summarises the mortality data across all 8 trials was -0.77 (95% CI: -0.14, -1.40, P=0.016). The probability of dying was 6.6% in the antibiotic-treated patients compared to 13.3% in the patients. The homogeneity of variance, Q, was 3.60 (P=0.82).
Removing the study by Luiten et al., which used systemic and oral antibiotics, gave a Mantel-Haenszel log OR of -1.16 (95% CI: -0.19, -2.15, P=0.016) and Q was 3.19 (P=0.78). Removing the trial by Sainio et al. gave a log OR of -0.56 (95% CI: 0.12, -1.24, P=0.10).
Three trials using ampicillin were analysed as a separate group. The combined log OR was 0.40 (95% CI: -1.71, 2.52, P=0.71) and Q was 0.20 (P=0.91).
Analysis of 4 studies using broad-spectrum antibiotics in patients with severe pancreatitis gave a Mantel-Haenszel log OR of -1.38 (95% CI: -0.32, -2.44, P=0.008) and Q of 1.41 (P=0.70). Patients who received antibiotics had a 5.3% probability of dying compared to a 18.2% incidence in the control group.