The first and most important step in planning the analysis is to specify the pair-wise comparisons that will be made. The comparisons addressed in the review should relate clearly and directly to the questions or hypotheses that are posed when the review is formulated (see Chapter 5). It should be possible to specify in the protocol of a review the main comparisons that will be made. However, it will often be necessary to modify comparisons and add new ones in light of the data that are collected. For example, important variations in the intervention may only be discovered after data are collected.
Decisions about which studies are similar enough for their results to be grouped together require an understanding of the problem that the review addresses, and judgement by the author and the user. The formulation of the questions that a review addresses is discussed in Chapter 5. Essentially the same considerations apply to deciding which comparisons to make, which outcomes to combine and which key characteristics (of study design, participants, interventions and outcomes) to consider when investigating variation in effects (heterogeneity). These considerations must be addressed when setting up the ‘Data and analyses’ tables in RevMan and in deciding what information to put in the table of ‘Characteristics of included studies’.