Introduction A meta-analysis was performed using individual patient data from the five most recent randomised controlled trials (RCTs) which evaluated corticosteroids in severe alcoholic hepatitis (Maddrey discriminant function (DF) ≥32 or encephalopathy). This approach overcomes limitations associated with the use of literature data and improves the relevance of the study and estimates of effect size.

Aims To compare 28-day survival between corticosteroid- and non-corticosteroid-treated patients and to analyse the response to treatment using the Lille model.

Methods Individual patient data were obtained from five RCTs comparing corticosteroid treatment with placebo (n=3), enteral nutrition (n=1) or an antioxidant cocktail (n=1).

Results 221 patients allocated to corticosteroid treatment and 197 allocated to non-corticosteroid treatment were analysed. The two groups were similar at baseline. 28-day survival was higher in corticosteroid-treated patients than in non-corticosteroid-treated patients (79.97±2.8% vs 65.7±3.4%, p=0.0005). In multivariate analysis, corticosteroids (p=0.005), DF (p=0.006), leucocytes (p=0.004), Lille score (p<0.00001) and encephalopathy (p=0.003) were independently predictive of 28-day survival. A subgroup analysis was performed according to the percentile distribution of the Lille score. Patients were classified as complete responders (Lille score ≤0.16; ≤35th percentile), partial responders (Lille score 0.16–0.56; 35th–70th percentile) and null responders (Lille ≥0.56; ≥70th percentile). 28-day survival was strongly associated with these groupings (91.1±2.7% vs 79.4±3.8% vs 53.3±5.1%, p<0.0001). Corticosteroids had a significant effect on 28-day survival in complete responders (HR 0.18, p=0.006) and in partial responders (HR 0.38, p=0.04) but not in null responders.

Conclusion Analysis of individual data from five RCTs showed that corticosteroids significantly improve 28-day survival in patients with severe alcoholic hepatitis. The survival benefit is mainly observed in patients classified as responders by the Lille model.