Expressing findings from meta-analyses of continuous outcomes in terms of risks

Meta-analyses of clinical trials with continuous outcome data typically report the effect of an intervention as either a mean difference or a standardized mean difference. These results can be difficult to interpret, and re-expressing the effect size in terms of risk may facilitate understanding and applicability. We describe three methods for obtaining risks in such situations. Two of these methods involve direct transformation of a standardized mean difference to an odds ratio. The third entails estimation of risks in the two groups for a specific cut point. We extend this third approach to a completed meta-analysis by expressing the finding in the format of a single 'meta-study'. We compare the methods in two examples of meta-analyses and in a series of simulation studies that examine their properties in individual studies and in meta-analyses. These simulations show that the methods for expressing meta-analysis results from continuous outcomes are sensitive to underlying distributions, sample sizes and cut points but are remarkably robust to the presence of heterogeneity across studies. We offer suggestions of situations in which the various methods may safely be applied. In particular, if the underlying distribution is approximately normal, then estimation of risks for a specific cut point may be used for large sample sizes; direct transformations may be preferable otherwise. However, if the standard deviations in the two groups are notably different, then none of the methods have good properties. Furthermore, absolute risks are safely estimated after direct transformation only if they are in the region of 20% to 80%.