Nineteen trials (1,370 patients) were included in the review. Another trial of 75 patients with TIPS was included in the sensitivity analysis. Sequence generation and allocation concealment were adequate in 16 trials; 14 trials were double blind and had no missing outcome data. Three trials were considered to have a risk of other bias.
A statistically significant effect was found with Rifaximin for the prevention of overt hepatic encephalopathy (RR 1.36; 95% CI 1.06 to 1.65; two RCTs). The significance of this result was reduced when the trial on TIPS was included (RR 1.24, 95% CI 1.00 to 1.53; Ι²=50%).
Rifaximin was statistically significant for full resolution of hepatic encephalopathy (RR 1.34, 95% CI 1.11 to 1.62; 11 RCTs; Ι²=54%; NNT six). The significance of this result was unaltered in the sensitivity analysis including only trials with a low risk of bias. There was no evidence of publication bias.
There were statistically significant improvements in manifestations of hepatic encephalopathy (RR 1.00, 95% CI 1.00 to 1.23 – random effects; and RR 1.19, 95% CI 1.11 to 1.27 – fixed effect). The number of trials in the analysis was not reported. The statistical significance remained in the fixed-effect sensitivity analysis only. There was evidence of publication bias. Sequential analysis showed that additional information was needed to support or refute improved hepatic encephalopathy manifestation.
Rifaximin significantly reduced mortality (RR 0.64, 95% CI 0.43 to 0.94; 12 RCTs; Ι²=0). The sequential analysis revealed that further trials were necessary to confirm this finding.
There were no differences between Rifaximin and control for serious adverse events (13 RCTs). Where it was possible to combine data (four RCTs), the result was not statistically significant.
The results for secondary outcomes, and of further sensitivity and subgroup analyses, were reported.