How to Read and Interpret a Forest Plot in Meta-Analysis

Q: What does the diamond at the bottom of a forest plot mean?

The diamond represents the pooled (overall) effect estimate from combining all the individual studies in the meta-analysis. Its horizontal position marks the pooled point estimate, and its width represents the 95% confidence interval of that pooled estimate. If the diamond does not touch or cross the vertical line of no effect, the pooled result is statistically significant; if it touches or crosses that line, the pooled result is not statistically significant.

Q: Why are the squares in a forest plot different sizes?

The size of each square is proportional to that study's weight in the meta-analysis — larger squares indicate studies that contributed more to the pooled estimate, usually because they had a larger sample size or more precise (narrower) confidence interval. A large study with a wide confidence interval and a small study with a narrow confidence interval can sometimes receive similar weight, since weight is driven by precision, not sample size alone.

Q: What does it mean if a study's confidence interval crosses the line of no effect?

It means that individual study's result was not statistically significant on its own — the data from that study are consistent with no effect, a beneficial effect, and a harmful effect all being possible, given its sample size and variability. This is extremely common for individual studies within a meta-analysis and is not a flaw; the whole purpose of pooling many such studies is to gain the statistical power that a single smaller study cannot achieve alone.

Q: What is heterogeneity and why does it matter in a forest plot?

Heterogeneity refers to genuine variability in the true effect size across the different studies included in a meta-analysis, beyond what would be expected from chance (random sampling error) alone. It is usually reported as the I² statistic, a percentage from 0% to 100%: below 25% is generally considered low heterogeneity, 25-50% moderate, and above 50-75% substantial to considerable. High heterogeneity means the studies may not all be estimating the same true effect — due to differences in populations, interventions, or outcome definitions — and is the primary reason a random-effects model is chosen over a fixed-effect model, and a reason to interpret the pooled estimate more cautiously.

Q: Should I trust a forest plot with very few studies?

A forest plot with only 2 or 3 studies can still be technically valid, but the pooled estimate and especially the heterogeneity statistic (I²) become very unstable with so few data points, and the confidence interval of the pooled result is often wide. Interpret meta-analyses with very few included studies more cautiously, and pay close attention to whether the individual study estimates are pointing in a broadly consistent direction rather than relying on the pooled diamond and p-value alone.

Q: Can a forest plot be statistically significant but not clinically meaningful?

Yes. With a large enough combined sample size across the pooled studies, a very small, clinically trivial effect can still produce a narrow confidence interval that excludes the line of no effect, making it statistically significant. Always look at the actual magnitude of the pooled effect estimate — the odds ratio, risk ratio, mean difference, or hazard ratio value itself — and judge whether that specific magnitude would matter to a patient or clinician, rather than relying on statistical significance alone to judge clinical importance.

A forest plot looks intimidating the first time you see one — rows of squares, lines of different lengths, and a diamond sitting at the bottom — but it is really just a picture of a debate between studies, resolved into one answer. This guide explains every single piece of a forest plot in plain English, walks through two fully labeled example plots you can study line by line, and covers fixed vs random-effects models, heterogeneity, common misreadings, and forest plots for every major effect measure (OR, RR, MD, SMD, HR) — everything you need to read one confidently in a paper or build one for your own meta-analysis.

What Is a Forest Plot?

A forest plot is a graph used in meta-analysis to display the results of multiple individual studies on the same research question, side by side, along with a single combined ("pooled") result at the bottom. Each row represents one study; a vertical line running through the whole plot marks the point of "no effect"; and each study's result is drawn as a square with a horizontal line through it, showing both its estimated effect and how precise that estimate is.

The name comes from its appearance — a column of horizontal lines can resemble a forest of trees viewed from a distance, an image that stuck in the epidemiology literature decades ago and never left.

Take-Home Points A forest plot is a visual summary of a meta-analysis: one row per study, plus one combined result at the bottom. Every visual feature on it (position, line length, square size) carries specific statistical meaning, covered piece by piece below.

Why Is a Forest Plot Used?

A forest plot solves a real problem: a systematic review might identify ten different studies on the same question, each with a different sample size, a slightly different estimate, and its own uncertainty — reading ten separate results in a table is hard to synthesize at a glance. A forest plot lets a reader instantly see, in one image, whether the studies broadly agree with each other, how much each one contributed to the final answer, and what the combined evidence concludes.

It also makes transparent exactly what a meta-analysis is doing mathematically: pooling several imprecise individual estimates into one more precise combined estimate, which is the entire logic behind why meta-analyses are considered high-quality evidence — more studies, thoughtfully combined, generally narrow the uncertainty around the true effect.

Forest plots also make a meta-analysis auditable in a way a paragraph of prose cannot. A reader can check, without re-running any statistics, whether the pooled conclusion actually reflects what the individual studies show, whether one enormous study is silently dominating the result, or whether the studies genuinely disagree with each other beneath a seemingly tidy pooled number — this transparency is a large part of why systematic reviews sit at the top of most evidence hierarchies used in clinical guideline development.

Take-Home Points Forest plots exist to make a meta-analysis's logic visible at a glance: how much each study agrees, how much each contributed, and what the combined evidence says.

Every Part of a Forest Plot, Explained

Below is a fully labeled, realistic example forest plot from a meta-analysis of five randomized trials comparing a new drug to placebo, using the odds ratio (OR) as the effect measure. Study through each element using the table underneath it, then look back at the plot until every piece makes sense.

Example Forest Plot — Odds Ratio, Fixed-Effect Model

Study estimate (size = weight) Pooled estimate (diamond) Line of no effect (OR = 1)

Heterogeneity: Chi² = 3.10, df = 4 (p = 0.54); I² = 0%
Test for overall effect: Z = 4.12 (p < 0.001)

Element	What It Shows You
Study names	The first author and year of each included study, one per row, usually ordered alphabetically, by year, or by weight.
Effect size	The point estimate for that study (here, an odds ratio) — the center of its square, and the number printed in the right-hand column.
Confidence interval	The horizontal line through each square — its left and right ends mark the lower and upper bounds of that study's 95% CI.
Squares	Mark each study's point estimate. Their size is proportional to that study's weight in the pooled analysis.
Horizontal lines	Represent each study's confidence interval — longer lines mean less precise (more uncertain) individual results.
Diamond	Represents the pooled (combined) effect across all studies. Its center is the pooled estimate; its width is the pooled 95% CI.
Weight	The percentage contribution of each study to the pooled estimate — driven mainly by precision, not just sample size.
Line of no effect	The vertical reference line at the value representing "no difference" (OR/RR/HR = 1, or MD/SMD = 0).
Overall effect	The pooled estimate and its CI, shown as the diamond and printed in the bottom row — the meta-analysis's headline result.
Heterogeneity	A statistic (usually I²) shown in a caption below the plot, describing how much the true effect appears to vary across studies.

Take-Home Points Every visual feature of a forest plot maps to a specific number: position = effect size, line length = precision (CI width), square size = weight, and the diamond = the pooled answer the whole plot is building toward.

Fixed-Effect Forest Plot

A fixed-effect model assumes that every included study is estimating the exact same true underlying effect, and that any differences between study results are due purely to random sampling error (chance), not genuine differences between the studies. This is a reasonable assumption when the included studies are clinically and methodologically similar — same population, same intervention, same outcome definition.

The example plot above uses a fixed-effect model: notice the low heterogeneity (I² = 0%, meaning essentially none of the variation between study results is beyond what chance alone would explain) and the correspondingly narrow diamond, since a fixed-effect model only needs to account for within-study sampling error when calculating the pooled CI.

Take-Home Points Choose a fixed-effect model, and expect a narrower pooled diamond, when studies are clinically similar and heterogeneity (I²) is low.

Random-Effects Forest Plot

A random-effects model instead assumes that the true effect genuinely varies from study to study — perhaps due to different populations, dosages, or follow-up durations — and that each study is estimating its own slightly different true effect, drawn from a distribution of plausible true effects. This model adds an extra layer of uncertainty (between-study variance) on top of ordinary sampling error, which is why random-effects pooled estimates almost always have a wider confidence interval than a fixed-effect analysis of the same data.

The example below shows a random-effects meta-analysis of five trials measuring the mean difference (MD) in systolic blood pressure with a new therapy versus control. Notice the substantial heterogeneity (I² = 73%) and the visibly wider diamond compared to the fixed-effect example above, even though both plots pool five studies of a similar size.

Example Forest Plot — Mean Difference, Random-Effects Model

Study estimate (size = weight) Pooled estimate (diamond) Line of no effect (MD = 0)

Heterogeneity: Tau² = 8.42; Chi² = 14.70, df = 4 (p = 0.005); I² = 73%
Test for overall effect: Z = 3.71 (p < 0.001)

Take-Home Points Random-effects models are the safer default whenever studies differ meaningfully in population or methods, or when heterogeneity (I²) is moderate to high — they produce a wider, more conservative pooled confidence interval than a fixed-effect model would on the same data.

Interpreting Statistically Significant Results

A pooled result is statistically significant when the diamond does not touch the vertical line of no effect — its entire width lies on one side of that line. In the fixed-effect OR example above, the diamond (0.64, 95% CI 0.53–0.78) sits entirely to the left of OR = 1, meaning the drug significantly reduced the odds of the outcome compared to placebo.

Individual studies can also be significant on their own — notice Okafor et al. 2020 (0.48, 95% CI 0.29–0.79) does not cross the line either — but a pooled result being significant does not require every individual study to be significant on its own; several individually non-significant studies can still combine into a significant pooled result once their precision is pooled together.

Take-Home Points Significance is a visual test: does the diamond (or an individual study's line) touch the vertical reference line? If no part of it touches, the result is statistically significant at conventional thresholds.

Interpreting Non-Significant Results

A result is not statistically significant when its line (for a study) or diamond (for the pooled estimate) touches or crosses the line of no effect — the data are consistent with the true effect being null, protective, or harmful, and the study or meta-analysis cannot distinguish between those possibilities with confidence.

A non-significant individual study is extremely common and expected within a meta-analysis — it is not a sign the study was flawed, simply that it alone did not have enough precision to rule out chance. A non-significant pooled result is more informative: it suggests that even after combining all available evidence, the true effect remains uncertain, which may reflect a genuinely small or absent effect, or simply that even the combined sample size remains too small.

Take-Home Points Non-significant individual studies are normal and expected — that's the reason meta-analysis exists. A non-significant pooled result is a more meaningful signal that current evidence is genuinely inconclusive.

Clinical vs Statistical Significance

A statistically significant pooled result only tells you the combined evidence is unlikely to reflect pure chance — it says nothing about whether the size of that effect actually matters to a patient. With enough pooled studies and a large enough combined sample size, even a tiny, clinically trivial effect can become statistically significant.

Practical Example

A meta-analysis pools 40,000 patients across 22 trials and reports MD = -1.2 mmHg (95% CI [-1.9, -0.5], p = 0.001) for systolic blood pressure with a new drug versus placebo. This is statistically significant — the CI excludes 0 — but a 1.2 mmHg reduction is very unlikely to be clinically meaningful for an individual patient's cardiovascular risk. Always look at the actual magnitude of the diamond's position, not only whether it crosses the null line.

Take-Home Points Read the number, not just the significance star. Ask whether the pooled effect's actual size would change clinical practice, independent of whether its confidence interval happens to exclude the null value.

Common Interpretation Mistakes

Mistake 1: Judging Only Whether the Diamond Crosses the Line

Reducing the whole plot to a binary "significant or not" ignores the actual magnitude of the pooled effect, its precision, and how consistent the individual studies were with each other.

✓ Fix: Report and interpret the pooled point estimate and its CI, not only a significance verdict.

Mistake 2: Ignoring Heterogeneity

Presenting a pooled estimate with high I² as if it were a single, well-agreed-upon answer glosses over the fact that the included studies may not be estimating the same true effect at all.

✓ Fix: Always report the I² statistic alongside the pooled estimate, and interpret high heterogeneity as a reason for caution, not a footnote to skip.

Mistake 3: Treating a Large Square as More Trustworthy Than a Small One

Square size reflects statistical weight (precision), not study quality — a large, poorly designed study can still receive a large square despite being at high risk of bias.

✓ Fix: Assess study quality (risk of bias) separately from weight; a forest plot alone does not tell you which studies were methodologically sound.

Mistake 4: Confusing the Scale for OR/RR/HR with the Scale for MD/SMD

Odds ratios, risk ratios, and hazard ratios are centered on 1 (no effect) and usually plotted on a logarithmic scale; mean differences and standardized mean differences are centered on 0 and plotted on a linear scale. Misreading which applies leads to misjudging distances on the plot.

✓ Fix: Always check the line of no effect's value (1 or 0) and the axis scale before comparing distances visually.

Mistake 5: Assuming Statistical Significance Equals Clinical Importance

A very large pooled sample size can make a trivially small effect statistically significant, which does not make it clinically actionable.

✓ Fix: Interpret the actual magnitude of the pooled estimate against a clinically meaningful threshold, not the p-value alone.

Forest Plots for OR, RR, MD, SMD, and HR

The same forest plot layout is used for every effect measure in meta-analysis — only the line of no effect's value and the axis scale change.

Effect Measure	Used For	Line of No Effect	Typical Scale
Odds Ratio (OR)	Case-control studies; binary outcomes in logistic regression	1	Logarithmic
Risk Ratio (RR)	Cohort studies, RCTs; binary outcomes, cumulative risk	1	Logarithmic
Mean Difference (MD)	Continuous outcomes measured on the same scale across studies	0	Linear
Standardized Mean Difference (SMD)	Continuous outcomes measured on different scales/instruments across studies	0	Linear
Hazard Ratio (HR)	Time-to-event (survival) outcomes, from Cox regression	1	Logarithmic

Choosing between MD and SMD depends entirely on whether the underlying studies used the same measurement instrument. If every trial measured pain using the identical 0-10 numeric rating scale, MD is the natural, directly interpretable choice — a pooled MD of -1.5 means "1.5 points lower on that specific scale." If different trials used different pain scales entirely (a 0-10 scale in one study, a 0-100 visual analogue scale in another), those raw units cannot be pooled directly, and SMD re-expresses each study's effect in standard deviation units instead, sacrificing some direct interpretability in exchange for making otherwise incompatible studies combinable.

See our guides on odds ratio, relative risk vs odds ratio, standardized mean difference (Cohen's d), and hazard ratio for the individual-study version of each of these measures, before they are ever pooled into a forest plot.

Take-Home Points Ratio measures (OR, RR, HR) center on 1 and use a log scale; difference measures (MD, SMD) center on 0 and use a linear scale. Always confirm which type you're looking at before interpreting distances on the plot.

Realistic Medical Example: Step by Step

Question: Does a new anti-emetic drug reduce postoperative nausea and vomiting (PONV) compared to standard care, based on the fixed-effect OR forest plot shown earlier?

Check the individual studies

All five squares sit to the left of OR = 1, suggesting a consistent direction of benefit even before pooling — a good sign of agreement between studies.

Check heterogeneity

I² = 0%, Chi² p = 0.54 — no meaningful heterogeneity detected, supporting the choice of a fixed-effect model and suggesting the studies are estimating a shared true effect.

Read the pooled diamond

OR = 0.64, 95% CI [0.53, 0.78] — entirely left of the line of no effect, so the pooled result is statistically significant.

Judge clinical significance

An OR of 0.64 represents a meaningful reduction in the odds of PONV, a common and distressing postoperative complication — plausibly large enough to matter clinically, not just statistically.

Write the conclusion

"Pooled analysis of five randomized trials showed the new anti-emetic significantly reduced the odds of postoperative nausea and vomiting compared to standard care (OR 0.64, 95% CI [0.53, 0.78], p < 0.001; I² = 0%)."

Practical Interpretation Checklist

Identify the effect measure and its null value

OR/RR/HR → line at 1; MD/SMD → line at 0.

Scan individual study direction

Are most squares on the same side of the null line, or scattered across both?

Check heterogeneity (I²)

Low (<25%) supports a fixed-effect model; moderate-to-high favors random-effects and more cautious interpretation.

Read the pooled diamond's position and width

Center = pooled estimate; width = pooled 95% CI.

Confirm statistical significance

Does the diamond touch the null line, or sit entirely on one side?

Judge clinical significance separately

Is the actual magnitude of the pooled estimate large enough to matter to a patient?

Note how many studies and how much total weight

Be more cautious with pooled results built from very few studies.

Frequently Asked Questions

What does the diamond at the bottom of a forest plot mean? +

The diamond represents the pooled (overall) effect estimate from combining all individual studies. Its horizontal position marks the pooled point estimate, and its width represents the pooled 95% confidence interval. If the diamond does not touch or cross the line of no effect, the pooled result is statistically significant; if it touches or crosses that line, it is not.

Why are the squares in a forest plot different sizes? +

Square size is proportional to that study's weight in the meta-analysis — larger squares indicate studies that contributed more to the pooled estimate, usually because they had a larger sample size or a more precise (narrower) confidence interval. A large study with a wide CI and a small study with a narrow CI can sometimes receive similar weight, since weight is driven mainly by precision.

What does it mean if a study's confidence interval crosses the line of no effect? +

It means that individual study's result was not statistically significant on its own — its data are consistent with no effect, a beneficial effect, and a harmful effect all being possible. This is extremely common for individual studies within a meta-analysis and is not a flaw; pooling many such studies is precisely how meta-analysis gains the statistical power a single smaller study lacks.

What is heterogeneity and why does it matter in a forest plot? +

Heterogeneity refers to genuine variability in the true effect size across studies, beyond what chance alone would explain — usually reported as I², from 0% to 100% (below 25% low, 25-50% moderate, above 50-75% substantial to considerable). High heterogeneity means the studies may not all be estimating the same true effect, is the primary reason for choosing a random-effects model, and is a reason to interpret the pooled estimate more cautiously.

Should I trust a forest plot with very few studies? +

A forest plot with only 2 or 3 studies can be technically valid, but the pooled estimate and especially I² become very unstable with so few data points, and the pooled CI is often wide. Interpret such meta-analyses cautiously, paying close attention to whether individual study estimates point in a broadly consistent direction rather than relying on the diamond and p-value alone.

Can a forest plot be statistically significant but not clinically meaningful? +

Yes. With a large enough combined sample size, a very small, clinically trivial effect can produce a narrow CI that excludes the null value, making it statistically significant. Always look at the actual magnitude of the pooled effect — the OR, RR, MD, SMD, or HR value itself — and judge whether that magnitude would matter to a patient, rather than relying on statistical significance alone.

Analyze the Studies Behind the Forest Plot

Use StatClinic's free calculators to compute the odds ratios, risk ratios, mean differences, and hazard ratios that feed into a meta-analysis — and get a journal-ready results sentence generated automatically. Free, no registration required.

Try StatClinic Free →