Seven studies were included and combined to determine the following outcome measures.
Good outcome versus all other grades (6 studies; 1,086 participants). Good or fair outcome versus other grades (4 studies; 1,016 participants).
Overall mortality (6 studies; 1,132 participants). Deficit or mortality attributed to vasospasm or DID (7 studies; 1,202 participants). Mortality attributed to vasospasm or DID (6 studies; 1,048 participants).
Cerebral infarction rate, as assessed by CT (3 studies; 921 participants).
Deficit or mortality attributed to rebleeding (3 studies; 842 participants).
Mortality attributed to rebleeding (4 studies; 958 participants).
All results are expressed as ORs with 99% CIs, significance level, and homogeneity test (Q) given in parentheses.
Good outcome versus all other grades: 1.86 (99% CI: 1.07, 3.25, p=0.004; p(Q)=0.11).
Good or fair outcome versus all other grades: 1.67 (99% CI: 1.13, 2.46, p=0.0007; p(Q)=0.43).
Overall mortality: 0.73 (99% CI: 0.42, 1.25, p=0.1; p(Q)=0.22).
Deficit or mortality attributed to vasospasm or DID: 0.46 (99% CI: 0.31, 0.68, p<0.00001; p(Q)=0.64).
Mortality attributed to vasospasm or DID: 0.50 (99% CI: 0.26, 0.97, p=0.007; p(Q)=0.55).
Cerebral infarction rate: 0.58 (99% CI: 0.38, 0.90, p=0.001; p(Q)=0.96).
Deficit or mortality attributed to rebleeding: 0.80 (99% CI: 0.2, 3.02, p=0.67; p(Q)=0.08).
Mortality attributed to rebleeding: 0.82 (99% CI: 0.3, 2.4, p=0.62; p(Q)=0.19).
Other than the first 3 analyses all other analyses were regarded as exploratory.
Publication bias. A funnel plot shows that the meta-analysis may be affected by publication bias. Further analysis to gauge the magnitude of this bias suggests that with the exception of the cerebral infarction analysis, the meta-analyses that showed positive treatment effects were unlikely to have been affected by publication bias.
Jackknife analysis. The main analysis (good outcome versus all other grades) was tested to determine if the single large trial unfairly dominatef the meta-analysis, resulting in unrepresentation of the smaller trials. The Jackknife analysis suggests that the meta-analysis is independent of the single large trial (OR range: 1.60 - 2.15).
Meta-regression. The main meta-analysis showed some degree of heterogeneity. This was explained by studies that recruited patients with more severe SAH which reported lower overall odds of good outcome (p=0.005).
Fixed-effect models. All meta-analyses were repeated using the fixed-effect model. There were no significant differences between the two models with the exception that the homogeneity test for the primary analysis in the fixed-effect model fell below the nominal level of 0.10.