This is an archived version of the Handbook. For the current version, please go to or search for this chapter here. Effect estimates and generic inverse variance meta-analysis

In some reviews, an overall estimate of effect will be sought from each study rather than summary data for each intervention group. This may be the case, for example, for non-randomized studies, cross-over trials, cluster-randomized trials, or studies with time-to-event outcomes. Meta-analysis can be applied to such effect estimates if their standard errors are available, using the generic inverse variance outcome type in RevMan (See Chapter 9, Section 9.4.3).  When extracting data from non-randomized studies, and from some randomized studies, adjusted effect estimates may be available (e.g. adjusted odds ratios from logistic regression analyses, or adjusted rate ratios from Poisson regression analyses). The process of data extraction, and analysis using the generic inverse variance method, is the same as for unadjusted estimates, although the variables that have been adjusted for should be recorded (see Chapter 13, Section 13.6.2).



On occasion, summary data for each intervention group (for example, numbers of events and participants, or means and standard deviations) may be sought, but cannot be extracted. In such situations it may still be possible to include the study in a meta-analysis using the generic inverse variance method. A limitation of this approach is that estimates and standard errors of the same effect measure must be calculated for all the other studies in the same meta-analysis, even if they provide the summary data by intervention group. For example, if numbers in each outcome category by intervention group are known for some studies, but only odds ratios (ORs) are available for other studies, then ORs would need to be calculated for the first set of studies and entered into RevMan under the generic inverse variance outcome type to enable meta-analysis with the second set of studies. RevMan may be used to calculate these ORs (entering them as dichotomous data), and the confidence intervals that RevMan presents may be transformed to standard errors using the methods that follow.


Estimates of an effect measure of interest may be presented along with a confidence interval or a P value. It is usually desirable to obtain a standard error from these numbers, so that the generic inverse variance outcome type in RevMan can be used to perform a meta-analysis. The procedure for obtaining a standard error depends on whether the effect measure is an absolute measure (e.g. mean difference, standardized mean difference, risk difference) or a ratio measure (e.g. odds ratio, risk ratio, hazard ratio, rate ratio). We describe these procedures in Section and Section, respectively. However, for continuous outcome measures, the special cases of extracting results for a mean from one intervention arm, and extracting results for the difference between two means, are addressed in Section 7.7.3.