Counts are described in Chapter 9, Section 9.2.5, and their meta-analysis is discussed in Chapter 9, Section 9.4.8. Data that are inherently counts may be analysed in several ways. The essential decision is whether to make the outcome of interest dichotomous, continuous, time-to-event or a rate. A common error is to treat counts directly as dichotomous data, using as sample sizes either the total number of participants or the total number of, say, person-years of follow-up. Neither of these approaches is appropriate for an event that may occur more than once for each participant. This becomes obvious when the total number of events exceeds the sample size, leading to nonsensical results. Although it is preferable to decide how count data will be analysed in advance, the choice is often determined by the format of the available data, and thus cannot be decided until the majority of studies have been reviewed. Review authors should generally, therefore, extract count data in the form in which they are reported.


Sometimes detailed data on events and person-years at risk are not available, but results calculated from them are. For example, an estimate of a rate ratio or rate difference may be presented in a conference abstract. Such data may be included in meta-analyses only if they are accompanied by measures of uncertainty such as a 95% confidence interval: see Section 7.7.7. From this a standard error can be obtained and the generic inverse variance method used for meta-analysis.