How to Read a Forest Plot (And What Most People Miss)
The forest plot is the most recognisable output of a meta-analysis - and one of the most frequently misread. Here is what the visual is actually telling you, and what it conceals.
The first recognisable forest plot appeared in a 1978 paper by Richard Peto and colleagues on antiplatelet therapy. The name - attributed to Richard Lewis, who noted that the plots looked like a forest of horizontal lines - stuck. Decades later, the forest plot is the standard visual language of meta-analysis, reproduced in Cochrane reviews, HTA submissions, and journal articles in every clinical specialty. Most readers learn to look at the diamond at the bottom and judge whether it crosses the vertical line. Most readers are missing the majority of what the plot is trying to communicate.
The Basic Structure
A forest plot displays one row per included study. On the left: the study identifier and, usually, some summary of the study design or sample size. In the centre: a horizontal line (the confidence interval) with a square at its midpoint (the point estimate). The size of the square is proportional to the study's weight in the meta-analysis - larger squares come from larger or more precise studies. On the right: the numerical values of the effect estimate and confidence interval.
The vertical line at the centre of the plot marks the null value - for ratio measures (odds ratio, risk ratio, hazard ratio), this is 1.0; for absolute measures (mean difference, standardised mean difference), this is 0. A confidence interval that does not cross this line represents a statistically significant result at the level of alpha corresponding to the width of the interval (usually 95%). A confidence interval that crosses it does not.
The Diamond
At the bottom of the plot sits a diamond representing the pooled estimate. The centre of the diamond is the point estimate; the left and right tips are the confidence interval boundaries. The diamond is visually compelling, and this is partly a problem - a narrow diamond looks precise, and a precise-looking diamond can seduce readers into confidence that the underlying evidence does not support.
The width of the diamond reflects the precision of the pooled estimate, which increases with more studies and larger sample sizes. It says nothing about whether those studies should have been pooled, whether there is clinical heterogeneity that makes the pool meaningless, or whether the confidence interval on the pooled estimate is the relevant uncertainty for any specific clinical decision.
Heterogeneity Statistics: I², tau², and the Prediction Interval
Below or beside the diamond, most forest plots report heterogeneity statistics. The Chi² test for heterogeneity provides a p-value, but it has low power with few studies and high power with many - it is one of the least informative numbers on the plot. I² estimates the proportion of variability attributable to between-study heterogeneity rather than sampling error; values above 50% are often described as "moderate" and above 75% as "considerable," though these thresholds are not absolute.
Tau² (or its square root, tau) estimates the actual variance in true effects across studies. This is the number that matters most for understanding whether the pooled estimate is likely to apply in a new setting. The prediction interval - which is derived from tau and the pooled estimate - tells you the range of effects you would expect in a new study, and a prediction interval that crosses the null even when the pooled estimate is significant is one of the most important pieces of information a forest plot can provide.
The Cochrane Handbook chapter on heterogeneity provides the statistical detail behind these measures and guidance on interpreting them in practice.
What the Visual Conceals
Several things that matter enormously for interpreting a meta-analysis are invisible in a standard forest plot. The quality of the individual studies - their risk of bias, their design, their population characteristics - is not encoded in the visual. Two studies with identical effect estimates and confidence intervals may have very different methodological quality; they will look identical on the plot. Subgroup or sensitivity analyses that change the pooled estimate substantially are not visible on the main forest plot; they require reading the full review.
The choice of effect measure matters and is not always salient. Relative measures (risk ratios, odds ratios) look different from absolute measures (risk differences, number needed to treat), and choosing between them is not a neutral decision. A relative risk of 0.75 looks like a 25% reduction regardless of whether the baseline risk is 1% or 50%; the absolute benefit is very different in the two cases, and the forest plot does not show you which situation applies.
Reading a Forest Plot Well
Reading a forest plot well requires looking beyond the diamond. Start with the individual study estimates: are they pointing in the same direction? Do any single studies dominate the pool by virtue of their large squares? Are there outlier studies whose confidence intervals do not overlap with the others? Then look at the heterogeneity statistics - not just I², but tau² and, if reported, the prediction interval. Ask whether the pooled estimate is being driven by a small number of large studies or represents a consistent signal across many. Look at the study characteristics table alongside the plot to assess whether the included studies are clinically comparable.
At NousLab, we build evidence synthesis outputs that include not just forest plots but the contextual annotations that make them interpretable - prediction intervals, risk of bias overlays, and subgroup breakdowns that put the main result in its proper perspective. See how we present evidence to support decision-making.
