The analysis section of a Cochrane review protocol may be more susceptible to change than other protocol sections (such as criteria for including studies and how methodological quality will be assessed). It is rarely possible to anticipate all the statistical issues that may arise, for example, finding outcomes that are similar but not the same as each other; outcomes measured at multiple or varying time-points; and use of concomitant treatments.
However the protocol should provide a strong indication as to how the author will approach the statistical evaluation of studies’ findings. At least one member of the review team should be familiar with the majority of the contents of this chapter when the protocol is written. As a guideline we recommend that the following be addressed.
Ensure that the analysis strategy firmly addresses the stated objectives of the review (see Section 9.1.2).
Consider which types of study design would be appropriate for the review. Parallel group trials are the norm, but other randomized designs may be appropriate to the topic (e.g. cross-over trials, cluster-randomized trials, factorial trials). Decide how such studies will be addressed in the analysis (see Section 9.3).
Decide whether a meta-analysis is intended and consider how the decision as to whether a meta-analysis is appropriate will be made (see Sections 9.1.3 and 9.1.4).
Determine the likely nature of outcome data (e.g. dichotomous, continuous etc) (see Section 9.2).
Consider whether it is possible to specify in advance what intervention effect measures will be used (e.g. risk ratio, odds ratio or risk difference for dichotomous outcomes, mean difference or standardized mean difference for continuous outcomes) (see Sections 9.4.4.4 and 9.4.5.1).
Decide how statistical heterogeneity will be identified or quantified (see Section 9.5.2).
Decide whether random-effects meta-analyses, fixed-effect meta-analyses or both methods will be used for each planned meta-analysis (see Section 9.5.4).
Consider how clinical and methodological diversity (heterogeneity) will be assessed and whether (and how) these will be incorporated into the analysis strategy (see Sections 9.5 and 9.6).
Decide how the risk of bias in included studies will be assessed and addressed in the analysis (see Chapter 8).
Pre-specify characteristics of the studies that may be examined as potential causes of heterogeneity (see Section 9.6.5).
Consider how missing data will be handled (e.g. imputing data for intention-to-treat analyses) (see Chapter 16, Sections 16.1 and 16.2).
Decide whether (and how) evidence of possible publication and/or reporting biases will be sought (see Chapter 10).
It may become apparent when writing the protocol that additional expertise is likely to be required; and if so, a statistician should be sought to join the review team.