Meta-analysis
Meta-analysis may be understood as “the statistical combination of the results of two or more studies” and may be used for various purposes.18 Earlier, we noted that available trials did not agree on whether steroids should be given to mothers delivering a premature baby to increase the chances of the baby’s survival, but a meta-analysis was used to resolve this question and determined that giving steroids to the mother would be beneficial to survival of the newborn.5 As illustrated in this case, a systematic review of existing evidence on a given question may be used in conjunction with meta-analysis to try to resolve conflicting claims regarding interventions.
While a common use of meta-analysis is to provide a quantitative synthesis of data collected from a systematic review of trials, meta-analysis may also be used to combine the results of observational analyses. For example, a study used interrupted time series analyses to determine whether each of 20 drug safety advisories in one of four countries was associated with a change in drug use. While the study found that the association of advisories with changes in drug use varied from one advisory to another, a random effects meta-analysis was used to show that on average advisories were associated with a decline in drug use of 5.8% in an 11-month post-advisory period.19
The results of a meta-analysis may be presented in a forest plot, which shows the findings of each included study or analysis and an overall result. For illustrative purposes, Figure 2 shows a meta-analysis of the effects of the Canadian drug safety advisories included in the meta-analysis mentioned above. It suggests the Canadian advisories were associated with a decline in drug use of 6.8%, which is reasonably consistent with the estimated average effect of advisories in all four countries. In practice, if we wanted to determine whether the association of advisories with changes in drug use differed by country, we would need to conduct a test of heterogeneity. You may wish to consult the Cochrane Handbook of Systematic Reviews of Interventions for further information on tests of heterogeneity and subgroup analyses.18
Meta-analysis methods typically use a variation on a weighted average of effect estimates from different studies or analyses.18 For example, studies may be weighted by the inverse of the variance of effect estimates from the individual studies, which weights according to the precision of each study. When conducting a meta-analysis, one methodological decision that needs to be made is whether a fixed effects or random effects approach to meta-analysis is appropriate. Broadly speaking, a fixed effects meta-analysis estimates a typical intervention effect based on the assumption that the results of individual studies differ only due to chance, while a random effects study estimates an average intervention effect and assumes that the results of individual studies vary due to both chance and genuine variation in effects.18
The study of drug safety advisories mentioned above used a random effects approach, which was likely appropriate because the drug safety advisories covered different topics and related to different populations.19 We would expect the effect of each advisory could differ for these reasons, although the overall result of the meta-analysis may capture the average effect of a regulatory advisory warning about an emerging drug risk.
Figure 2. Change in the rate of prescriptions following Canadian drug safety advisories
