What to know
- Each square is one study; its size shows the study's weight and its line shows the confidence interval.
- The diamond is the pooled result; check whether it crosses the line of no difference (1 for ratios, 0 for differences).
- I² describes how much studies disagree; high heterogeneity means asking why before trusting the average.
01
What a forest plot shows
A forest plot is the standard picture of a meta-analysis. Each row is one study. A square marks that study's estimate, and a horizontal line through it shows its confidence interval, the range of effects compatible with the study's result. The Cochrane Handbook notes that the area of each square shows how much weight the study received. At the bottom, a diamond shows the pooled result: its centre is the combined estimate and its width is the confidence interval.
A vertical line marks no difference between the groups. For ratios such as risk ratios, odds ratios and hazard ratios it sits at 1; for differences, such as a difference in means, it sits at 0. Cochrane's own logo is a forest plot: its diamond sits clearly to one side of that line, showing that corticosteroids given before a premature birth can save newborn lives. Labels under the axis tell you which side favours which treatment, so read them before anything else.
02
Reading a forest plot step by step
The table below lays out an invented forest plot as rows. The numbers are hypothetical, chosen only to show how the parts fit together; the pooled line was calculated from them with the standard inverse-variance method, in which more precise studies count for more.
| Study | Risk ratio | 95% confidence interval | Weight | How to read it |
|---|---|---|---|---|
| Study A | 0.70 | 0.45 to 1.09 | 8% | Favours treatment, but the line crosses 1: compatible with no difference. |
| Study B | 0.85 | 0.70 to 1.03 | 43% | Narrow interval, large square: a precise study that carries much of the weight. |
| Study C | 1.10 | 0.60 to 2.02 | 4% | Points the other way, but the wide interval means it adds little information. |
| Study D | 0.78 | 0.64 to 0.95 | 41% | The whole line is below 1: on its own, this study found a lower risk. |
| Study E | 0.60 | 0.30 to 1.20 | 3% | The largest apparent effect, but very uncertain and lightly weighted. |
| Pooled (diamond) | 0.81 | 0.71 to 0.92 | 100% | About 19% lower risk on average; the diamond does not touch 1. |
- Start with the axis: which measure is used, where the line of no difference sits, and which side favours which group.
- Look at the diamond: where it sits and whether it crosses the line of no difference.
- Then the squares: which studies dominate the weight, and whether they agree with each other.
- Then the numbers: the heterogeneity statistics, usually printed under the plot, and the absolute effect if the review reports one.
Notice that three of the five studies cross the line on their own, yet the pooled interval does not. Combining studies increases precision. That is the main reason meta-analyses exist, and also why a pooled result needs the studies to be similar enough to combine.
Sources for this section: [1]
03
When studies disagree: heterogeneity and I²
If the confidence intervals of the studies barely overlap, the studies may be estimating different things. The I² statistic describes the percentage of the variability in estimates that is due to real differences between studies rather than chance. The Cochrane Handbook gives a rough guide: 0 to 40% might not be important, 30 to 60% may be moderate, 50 to 90% may be substantial and 75 to 100% considerable. It warns against relying on fixed thresholds, because the importance of heterogeneity also depends on the size and direction of the effects.
Real Cochrane reviews show why this matters. For varenicline against placebo, 41 studies with 17,395 people gave a risk ratio for quitting of 2.32 (95% CI 2.15 to 2.51) with I² of 60%: the studies varied, but the whole pooled interval sat well above 1 and certainty was rated high. For pulmonary rehabilitation after a COPD flare-up, eight studies gave an odds ratio for hospital readmission of 0.44 (0.21 to 0.91) with I² of 77%, and certainty was downgraded to moderate because the results were so inconsistent. In both cases the diamond is useful, but in the second you need to ask why the studies differ before trusting the average.
04
What the diamond does not tell you
The pooled result depends on the model. A fixed-effect analysis assumes every study estimates one shared effect, so its diamond is a typical effect. A random-effects analysis assumes the true effect varies between settings, so its diamond is an average effect: the Handbook stresses that it describes the centre of that distribution, not its width. Some reviews add a prediction interval, which shows the range of effects you might expect in a new setting. It is often much wider than the diamond.
A forest plot also cannot show problems that sit outside it. If small studies with negative results were never published, the plot will look more favourable than reality. If the included trials had a high risk of bias, a precise diamond can be precisely wrong. And a relative measure, such as a risk ratio, says nothing about absolute benefit until you combine it with how common the outcome is. Look for the review's risk-of-bias assessment, its certainty rating and any absolute effects it reports.
Compare absolute and relative risk
Sources for this section: [1]
05
A checklist for any forest plot
Use these questions when a review, guideline or news story shows a forest plot. Most answers are in the plot itself, the text under it, or the review's summary of findings table.
- Which outcome and which effect measure are shown, and where is the line of no difference?
- Which side of the line favours the treatment, according to the axis labels?
- Does the diamond cross the line of no difference, and how wide is it?
- Which studies carry most of the weight, and are they the most relevant to you?
- What is I², and do the studies point in the same direction?
- Was a fixed-effect or random-effects model used, and is a prediction interval reported?
- What do the risk-of-bias and certainty ratings say, and is an absolute effect given?
Sources
- Chapter 10: Analysing data and undertaking meta-analyses (Cochrane Handbook for Systematic Reviews of Interventions, version 6.5)Cochrane (Deeks, Higgins, Altman, McKenzie and Veroniki) · 2024Source accessed:
- Our name and logoCochraneSource accessed:
- Nicotine receptor partial agonists for smoking cessationCochrane (Livingstone-Banks and colleagues) · 2023Source accessed:
- Pulmonary rehabilitation following exacerbations of chronic obstructive pulmonary diseaseCochrane (Puhan, Gimeno-Santos, Cates and Troosters) · 2016Source accessed:
Revision history
- Initial English, Spanish and Japanese article prepared with automated assistance. The cited sources were checked for the complete article on 2026-10-28T03:07:45Z (UTC). The worked forest plot is hypothetical and its pooled line was recalculated with the inverse-variance method. Publication and preparation use the current date; no historical date was assigned. No independent statistical or clinical review is recorded.