Ten RCTs were used to assess absolute mortality rate difference (N=1,329).
Four RCTs were used to assess mortality rate difference in Gram-negative infection (N=413; N=251 after stated exclusions).
Four RCTs were used to assess mortality rate difference in Gram-positive infection (N=306; N=142 after stated exclusions).
Five RCTs were used to assess 'low-dose' steroids (N=530).
Five RCTs were used to assess 'high-dose' steroids (N=799).
Five RCTs were used to assess gastrointestinal bleeding (N=648).
Six RCTs were used to assess secondary infection (N=1,017).
Four RCTs were used to assess hyperglycaemia (N=529).
Absolute mortality rate difference (treatment minus control): global effect, -0.2% (95% confidence interval, CI: -9.2, 8.8%; test for heterogeneity, P < 0.001).
Mortality rate differences were 4.8% (95% CI: -0.3, 9.9; test for heterogeneity, P > 0.2) after excluding the trial by Schumer, and 3.5% (95% CI: -2.0, 9), after excluding the trial by Schumer and the trial with a different protocol.
Mortality rates difference after excluding patients who died on the first day was -1.7% (95% CI: -11.0, 7.6).
Type and dosage of steroid mortality rate difference: 'low-dose' regime (less than 20 g equivalents of hydrocortisone during the first 24 hours of the study), -1.9% (95% CI: -20.0, 16.2); 'high-dose' regime, 3.6% (95% CI: -2.5, 9.8).
Gram-negative infections:
mortality rate difference was -5.6% (95% CI: -21.4, 10.1), and after excluding the trial by Schumer and the trial with a different protocol, -2.9% (95% CI: -14.7, 8.9).
Gram-positive infections:
mortality rate difference was 1.8% (95% CI: -15.1, 18.6), and after excluding the trial by Schumer and the trial with a different protocol, 9.2% (95% CI: -4.4, 22.7).
Adverse effects of steroids:
gastrointestinal bleeding rate difference was 2.3% (95% CI: -0.7, 5.4), and after excluding the study in which steroids were administered for 6 days, 1.1% (95% CI: -3.2, 5.4);
secondary or 'super infection' rate difference was 0.4% (95% CI: -4.4, 5.2); and
hyperglycaemia difference was 0.2% (95% CI: -4.0, 4.4).
The quality assessment scores of the primary studies range from 41 to 97 out of a maximum of 100.