Thirteen trials were identified (659 patients). However, not all trials were used in all stages of the meta-analyses. Twelve trials (622 patients) reported patient characteristics separately for each treatment group, and 11 trials reported further descriptive variables.
The pooled RR of death is 0.62 (95% CI: 0.38,1.05). The results using a fixed-effect model are statistically significant, showing a beneficial effect of glucocorticoid treatment.
The results are suggestive of publication bias; studies with a high weight (larger sample size) showing no therapeutic effect and those with a low weight (smaller sample size) showing an effect. No association was found between the effect size and the quality score, the type of corticosteroid, the daily dose or duration of therapy.
There was heterogeneity between trials, with the smallest trials having the greatest imbalance between treatment and control groups with respect to patient characteristics (age, male gender, encephalopathy, ascites, bilirubin and albumin). In particular, the following imbalance was associated with a high therapeutic effect: a low ratio of bilirubin, percentage male, or percentage with ascites.
Accounting for prognostic variables, the overall effect of glucocorticosteroid treatment is not significant - the RR is 0.78 (95% CI: 0.51,1.18). Poorer prognosis was significantly-associated with higher prevalence of encephalopathy and a high bilirubin concentration.