Using the "Uniform Scale" to facilitate meta-analysis where exposure variables are qualitative and vary between studies - methodology, examples and software [version 1; peer review: 1 approved with reservations]

Meta-analyses often combine covariate-adjusted effect estimates (odds ratios or relative risks) and confidence intervals relating a specified endpoint to a given exposure. Standard techniques are available to do this where the exposure is a simple presence/absence variable, or can be expressed in defined units. However, where the definition of exposure is qualitative and may vary between studies, meta-analysis is less straightforward. We introduce a new "Uniform Scale" approach allowing expression of effect estimates in a consistent manner, comparing individuals with the most and least possible exposure. In 2008, we presented methodology and made available software to obtain estimates for specific pairwise comparisons of exposure, such as any versus none, where the source paper provides estimates for multiple exposure categories, expressed relative to a common reference group. This methodology takes account of the correlation between the effect estimates for the different levels. We have now extended our software, available in Excel, SAS and R, to obtain effect estimates per unit of exposure, whether the exposure is defined or is to be expressed in the "Uniform Scale". Examples of its use are presented.