This page provides methodological interpretation of aggregate research output only. It does not convert a pooled result into individual treatment, diagnosis, or risk advice.
Read the forest plot
| Element | Meaning |
|---|---|
| Square or point | Study effect estimate |
| Horizontal line | Study confidence interval |
| Square size | Relative analysis weight, when encoded |
| Vertical line | Null value: often 0 for differences, 1 for ratios |
| Diamond | Pooled estimate; width commonly shows its interval |
| Heterogeneity output | Variation beyond sampling error under the selected model |
First identify the effect measure. A ratio axis is often logarithmic even when labels display ordinary ratios. Do not interpret equal visual distances around 1 as ordinary arithmetic distances without checking the scale.
An interval crossing the null is not evidence that the study found “no effect.” It describes uncertainty under the study analysis. The pooled diamond is not a vote count; weights depend on precision and the meta-analysis model.
Odds-ratio interpretation and odds ratio versus relative risk explain ratio scales. Bland–Altman plots answer a different measurement-agreement question.
Read heterogeneity carefully
I² describes the proportion of observed variation attributed to between-study heterogeneity under its definition, not the percentage of studies that disagree. Its uncertainty can be wide, especially with few studies. Inspect study populations, designs, outcomes, and effect definitions instead of using one I² label as a quality grade.1
Funnel plots do not diagnose publication bias alone
In a simple funnel plot, more precise studies cluster near the pooled estimate while less precise studies spread more widely. Asymmetry can arise from publication processes, true heterogeneity related to study size, selective analysis, poor methods, or chance. Sterne and Egger warn against equating asymmetry with publication bias.2
Tests for asymmetry also have limited power with few studies and can mislead under heterogeneity. State the test, assumptions, and number of studies.
A defensible reporting sentence
“The pooled ratio estimate was 0.82 (95% CI [0.70, 0.97]) under the stated random-effects model. Study estimates varied, and funnel asymmetry was treated as a nonspecific small-study pattern rather than proof of publication bias.”