Twelve studies were included in the review (n=497): four observational studies without controls (n=39); six observational studies with historical controls (n=314); and two RCTs (n=144). Jadad scores for the two RCTs were 1 and 3.
No statistically significant between-group difference for mortality was found for the two RCTs (relative risk 0.6, 95% CI: 0.27 to 1.34). Evidence of statistical heterogeneity was found with Cochran Q (p=0.045), but not with Χ2 (p=0.21). In the non-randomised studies with historical controls, a statistically significant effect in favour of immunosuppressive therapy was found (relative risk 0.55, 95% CI: 0.39 to 0.77). Evidence of statistically significant heterogeneity was found (Cochran Q p=0.001, Χ2 p=0.001 and I2=72.9%). The survival rate in the four uncontrolled studies was 74.4% (95% CI: 58.9% to 85.4%). Visual inspection of the forest plot indicated heterogeneity.
The number needed to treat for observational studies was three (95% CI: 3 to 4) and for RCTs was five (95% CI: 3 to 14); when combined, one out of four patients (95% CI: 3 to 5) were treated successfully with immunosuppressive therapy.
The funnel plot showed evidence of significant publication bias for the outcome of mortality in all the controlled trials.