Forest plot interpretation is the skill of reading a meta-analysis at a glance: each horizontal row is one study, the square is its point estimate, the line through it is the confidence interval, and the box size shows the study’s weight. At the bottom, a diamond marks the pooled estimate, with its width showing the combined confidence interval. A vertical line of no effect runs down the middle, and where everything sits relative to that line tells you the direction and certainty of the result.
Anatomy of the plot, element by element
Reading a forest plot well means knowing what every mark encodes. The square is the study’s observed effect, and its area is proportional to how much the study contributes to the pool, so a large precise trial draws a big box and a small one a tiny box. The whiskers are the confidence interval: long whiskers mean an imprecise estimate, short ones mean a precise estimate. The diamond at the foot is the pooled result, and you read its centre for the effect and its width for the uncertainty.
Reading direction and significance
The vertical reference line is the value at which the intervention does nothing: 1 for ratio measures such as the odds ratio or risk ratio, and 0 for differences such as the mean difference. If a study’s confidence interval crosses that line, its result is not statistically significant on its own. The same logic applies to the diamond: a pooled diamond that touches the line of no effect means the combined result is not significant, however many studies point one way.
Spotting heterogeneity by eye
Before trusting the diamond, scan the scatter. If the squares line up on the same side with overlapping intervals, the studies agree and pooling is comfortable. If they sit on both sides of the line with intervals that barely overlap, you are looking at heterogeneity, and the single pooled number may be papering over real disagreement. The plot is the visual companion to the formal statistics covered in heterogeneity in meta-analysis, and a good reviewer reads both together. Most plots print I-squared and the Q statistic in a corner; you can reproduce those figures from your own data in the heterogeneity calculator to confirm the visual impression matches the numbers.
The prediction interval, the most overlooked mark
Under a random-effects model the diamond shows the uncertainty around the average effect, but it says nothing about how widely the true effect varies between settings. A prediction interval, sometimes drawn as a second wider bar beneath the diamond, estimates the range in which the effect of a future study would plausibly fall. It is almost always wider than the confidence interval, and when heterogeneity is high it can straddle the line of no effect even though the diamond clears it. A pooled estimate that looks decisively positive but whose prediction interval crosses into harm is a result to report cautiously, because the next study could easily land the other side of the line.
Reading subgroup rows and the test for subgroup differences
Many plots stack the studies into subgroups, each with its own pooled diamond, then report a test for subgroup differences. A small p-value there suggests the effect genuinely differs between subgroups, which is more informative than the overall diamond alone. Read these rows with care: a subgroup contrast is an observational comparison across studies, so it generates a hypothesis rather than confirming one, and it should have been pre-specified to carry weight.
Weights and the model behind the plot
The box sizes also reveal the model. Under a fixed-effect versus random-effects choice, a fixed-effect plot lets a few large studies dominate with big boxes, while a random-effects plot evens the weights so smaller studies carry more. If the pooled estimate shifts noticeably between the two, that sensitivity belongs in your write-up. The plot is one output of the wider workflow in how to do a meta-analysis.
Common misreadings to avoid
Three mistakes recur. First, treating a wide diamond as a strong result because its centre looks favourable; width is uncertainty, and a wide diamond is weak evidence. Second, ignoring the scale: on a log axis, distances are not linear, so a value of 2 and a value of 0.5 are equidistant from 1. Third, reading significance without checking how trustworthy each study is; a tidy plot built from biased studies is still a biased result. And the plot says nothing about publication bias, which needs a separate funnel plot to assess.
A worked reading of a small plot
Picture five trials of a treatment, all reporting a risk ratio. Four sit just left of the line at 1, with confidence intervals that each cross it, so none is significant alone. The fifth, a large trial with a big box, sits well left at 0.70 with a tight interval that clears the line. The pooled diamond lands at 0.82 with an interval from 0.71 to 0.95, so the combined result is significant. The honest reading is that the significant pool rests heavily on one dominant study, the four smaller ones merely lean the same way, and a leave-one-out check that drops the large trial would test whether the conclusion holds. That is the kind of fragility a robustness analysis of the pool is built to expose, and it is invisible to anyone who reads only the diamond.
Putting it together
A confident reading runs in order: check the direction against the line of no effect, gauge precision from the whisker lengths, scan the scatter for heterogeneity, read any prediction interval and subgroup rows, note the weights and model, then read the diamond last. Done that way, a forest plot communicates an entire synthesis in a few seconds, which is exactly why it is the centrepiece of nearly every meta-analysis. If you want the plot built and read against the underlying data rather than taken at face value, that is part of our statistics service.