91828 Chap.4 Forest Plots and Funnel Plots
Forest Plots and Funnel Plots
Forest plots compress effect direction, precision, study weight and a pooled estimate onto one scale. This chapter teaches a four-pass reading that begins with the measure and null, then examines individual intervals, relative weights and clinical consistency.
It also treats funnel-plot asymmetry as a prompt for investigation rather than proof, connecting visual patterns back to search completeness, study size and methodological heterogeneity.
What this chapter covers
- 01
Effect scale and null
- 02
Point estimates and intervals
- 03
Statistical weight
- 04
Pooled diamond
- 05
Clinical heterogeneity
- 06
Funnel-plot asymmetry
Worked example · free
Interpret a pooled risk ratio
- +1The point estimate indicates adverse-outcome risk under intervention is 81% of control risk.
- +1The estimated relative reduction is 1 - 0.81 = 0.19 or 19%.
- +1The interval excludes the ratio null of 1 and represents the plausible relative range under the model.
- +1Control event risk is required to calculate absolute risk reduction and NNT.
- +1Study bias, heterogeneity and applicability still constrain the pooled interpretation.
Key terms
- Forest Plot
- A forest plot shows study effect estimates, intervals, weights and often a pooled result on one scale.
- Null Line
- The null line marks no effect, usually one for ratios and zero for differences.
- Pooled Diamond
- A pooled diamond shows the summary estimate at its centre and confidence interval across its width.
- Funnel Plot
- A funnel plot compares effect size with precision or study size to explore small-study patterns.
Forest Plots and Funnel Plots FAQ
What does a confidence interval crossing one mean?
For a ratio measure, it means values compatible with the data include the null and effects on both sides, at the chosen confidence level. It does not prove equivalence or prove that treatment has no effect; the study may be imprecise.
Does a larger square mean a better study?
It usually means greater statistical weight, often because the estimate is more precise. Weight does not guarantee low bias, clinical relevance or good reporting. A large biased study can strongly influence a pooled estimate and deserves careful appraisal.
Does funnel asymmetry prove publication bias?
No. Missing studies related to results are one explanation, but clinical heterogeneity, methods that vary with study size, selective outcomes, chance and an unsuitable scale can also produce asymmetry. Investigate the search and study characteristics before assigning a cause.
How do I tell which side favours treatment?
Read the outcome and comparison before the geometry. A lower ratio favours treatment for an undesirable event but may favour control for recovery. Check axis labels, transformations and group definitions, then state direction in words rather than relying on left or right alone.
What does the pooled diamond width show?
The centre is the pooled point estimate and the horizontal width is its confidence interval. A narrow diamond indicates statistical precision under the model, not freedom from bias or clinical coherence. Interpret it alongside the contributing studies, heterogeneity and review methods.
Can I compare two studies by significance alone?
No. One interval excluding the null while another includes it does not prove their effects differ. Compare the estimates and intervals directly, examine an interaction or heterogeneity test when appropriate, and consider whether populations, outcomes or methods explain an apparent contrast.
Assessment move
Practise forest plots with a fixed four-pass routine until visual reading no longer depends on the authors' conclusion. First name the effect measure, its null value and which direction favours the intervention for the stated outcome. Second, read each point estimate and interval, identifying the most precise study from interval width rather than square size alone.
Third, inspect relative weight, the pooled diamond and the consistency of effects. Fourth, return to populations, interventions, comparators, outcome definitions and follow-up to decide whether one average remains clinically meaningful. Use examples with ratios and differences so one and zero are never confused as null values.
Cover the numerical labels and estimate approximate values from the scale, then uncover them to check accuracy. Write a full interpretation that includes outcome, comparison, magnitude, interval, relation to the null and a clinically meaningful threshold. Never calculate absolute benefit from a ratio without a control risk.
Create paired examples where one study is statistically significant and another is not, then explain why that contrast does not prove their effects differ. Use intervals or a direct interaction instead. For funnel plots, begin with axes and the expected relation between precision and scatter. List publication bias, clinical heterogeneity, method differences, chance and scale choice as competing explanations for asymmetry.
Connect each possibility to evidence you would seek in the search methods or study table. Avoid confident visual claims when few studies are available. For AT1 notes, translate every visual observation into a methods question: a dominant study prompts scrutiny of its validity; wide dispersion prompts an explanation; a one-sided gap prompts an unpublished-evidence audit.
Finish by recording the last search date and trial-register coverage, because apparent asymmetry is interpreted in the context of what the review attempted to find. A sound response uses the plot to locate uncertainty and the methods to explain it.
