RevMan provides a flexible framework for producing forest plots in the ‘Data and analyses’ section of a Cochrane review. Components of a Cochrane forest plot are described in Box 11.3.a, and an example from RevMan is given in Figure 11.3.a, using results from a review of compression stockings to prevent deep vein thrombosis in airline passengers (Clarke 2006). A tutorial on the use of RevMan is available within RevMan (available also from here).
RevMan offers multiple options for changing the analysis methods (e.g. between fixed and random-effects meta-analyses, or using different measures of effect; see Chapter 9 (Section 9.4), and graphics (e.g. scale of axes and ordering of studies). One forest plot for each dataset entered into RevMan is automatically incorporated into the full published version of the Cochrane review. Default analyses are displayed unless options are overridden. The defaults are Mantel-Haenszel odds ratios for dichotomous data, fixed-effect meta-analyses of mean differences for continuous data, Peto odds ratios for ‘O–E and Variance’ outcomes and fixed-effect meta-analyses for generic inverse variance outcomes (see Chapter 9, Section 9.4). The author should override any default settings that do not correspond with results reported in the text when setting up or editing outcomes in RevMan. This ensures that the results displayed are consistent with what is described in the text. In addition, the scale of the axis should be selected so that the point estimates (and most, if not all, of the confidence intervals) are visible in the plot.
A past convention in the Cochrane Database of Systematic Reviews (CDSR) has been that dichotomous outcomes have focused on unfavourable outcomes, so that risk ratios and odds ratios less than one (and risk differences less than zero) indicate that an experimental intervention is superior to a control intervention. This would result in effect estimates to the left of the vertical line in a forest plot implying a benefit of the experimental intervention. The convention is no longer encouraged since it is not universally appropriate. A much superior approach is to make it transparent which side of the line indicates benefit of which intervention by labelling the directions on the axis on the forest plots. RevMan allows authors to specify the labels used for ‘experimental’ and ‘control’ groups in each outcome. These labels are then used in the CDSR. Thus it is essential to know which way figures are constructed and should be interpreted. This is particularly important for measurement scale data where it is not always apparent to a reader which direction on a scale indicates worsening health.
Forest plots should not be generated that contain no studies, and are discouraged when only a single study is found for a particular outcome. To display outcomes that have been investigated only in single studies, authors can use a forest plot using a subgroup for each outcome (ensuring that the option to pool the data is disabled). Otherwise results of single studies may more conveniently be presented in an Additional table (see Section 11.6).