Forest Plots

What is a Forest Plot?

A forest plot is a graphical display of effect sizes from individual studies along with the meta-analytic summary. It's the standard way to present meta-analysis results and is required for most systematic review publications.

Anatomy of a Forest Plot

Left Side: Study List

  • Author and year for each study
  • Sorted by year or effect size
  • Grouped by subgroups if applicable
  • Total at bottom with diamond

Middle: Data Columns

  • Sample sizes (n) for intervention and control
  • Event counts for binary outcomes
  • Means and SDs for continuous outcomes
  • Customizable columns

Right: Effect Estimates

  • Squares represent point estimates (size ∝ weight)
  • Horizontal lines show 95% confidence intervals
  • Vertical line at null effect (0 for MD/SMD, 1 for RR/OR)
  • Diamond at bottom shows pooled effect and CI

Far Right: Numeric Summary

  • Effect size with 95% CI
  • Study weights (%)
  • Total effect with p-value

Reading a Forest Plot

Direction of Effect

For continuous outcomes (MD, SMD):

  • Negative values: Favor intervention (if lower is better)
  • Positive values: Favor control (if lower is better)
  • Customize labels: 'Favors intervention' vs 'Favors control'

For binary outcomes (RR, OR):

  • RR < 1 means lower event risk in the intervention group; RR > 1 means higher event risk
  • OR < 1 means lower event odds in the intervention group; OR > 1 means higher event odds
  • Which side favors an intervention depends on whether the event is desirable or adverse and on the comparison order

Statistical Significance

  • If CI crosses the null line: NOT statistically significant
  • If CI doesn't cross null: Statistically significant
  • Width of CI indicates precision (narrower = more precise)

Study Weights

  • Larger squares = more weight in meta-analysis
  • Weight based on sample size and variance
  • Fixed-effect: Weight ∝ 1/variance
  • Random-effects: Also accounts for between-study variance

Creating Forest Plots in Scholara

  1. 1Ask the Analysis Chat to run a meta-analysis and generate a forest plot
  2. 2Review the forest plot that appears as an interactive card in the chat
  3. 3Ask for customizations in follow-up messages (e.g., 'Sort by year', 'Add subgroups', 'Use grayscale')
  4. 4Click the expand icon on the plot card to view full-screen
  5. 5Export when satisfied using the export menu on the card

Customization Options

Layout

  • Sort studies: By year, effect size, alphabetically, custom
  • Include/exclude individual studies
  • Add study quality indicators (traffic lights)
  • Show/hide data columns
  • Adjust plot width and height

Labels and Text

  • Title and subtitle
  • X-axis label (e.g., 'Standardized Mean Difference')
  • Direction labels ('Favors A' / 'Favors B')
  • Font sizes for readability
  • Study ID format (first author, full citation, custom)

Statistical Display

  • Show/hide heterogeneity statistics (I², Q, p-value)
  • Display overall effect p-value
  • Include/exclude weights column
  • Precision: Decimal places for estimates

Subgroups

  • Group by study characteristic (design, setting, etc.)
  • Show subgroup totals with separate diamonds
  • Test for subgroup differences
  • Collapse/expand subgroups

Subgroup Analysis

Explore whether effects differ by study characteristics:

  1. 1Select grouping variable (e.g., risk of bias, age group)
  2. 2Scholara creates separate summary effects per subgroup
  3. 3Tests for differences between subgroups (Q-test)
  4. 4Displays in forest plot with separate sections
  5. 5Interpret cautiously: Observational, not experimental comparison

Common Subgroup Variables

  • Risk of bias (low vs unclear/high)
  • Study design (RCT vs quasi-experimental)
  • Population characteristics (age, severity)
  • Intervention details (dose, duration, format)
  • Setting (hospital, community, online)
  • Geographic region

Subgroup analyses should be pre-specified in your protocol to avoid data dredging. Interpret with caution, especially with few studies per subgroup.

Sensitivity Analysis

Assess robustness of results by re-running analyses excluding certain studies:

  • Exclude high risk of bias studies
  • Exclude outliers (studies with extreme effects)
  • Exclude studies with imputed data
  • Use different effect size metric (e.g., OR vs RR)
  • Use different model (fixed vs random effects)
  • Compare results to primary analysis

Exporting Forest Plots

Image Formats

  • PNG: High resolution for manuscripts (300+ DPI)
  • PDF: Vector format, scalable without quality loss
  • SVG: Editable in vector graphics software
  • TIFF: Required by some journals

Publication Quality

  • 300 DPI minimum for print journals
  • Black and white or grayscale preferred (check journal)
  • Ensure text is readable at published size
  • Follow journal-specific formatting requirements
  • Include figure caption separately

Interpreting Special Cases

Large Confidence Intervals

Very wide CIs indicate:

  • Small sample size in that study
  • Large variance in outcomes
  • Imprecise effect estimate
  • Study carries less weight in meta-analysis

Outliers

Studies with effects far from others:

  • Check for data entry errors
  • Investigate study methods for differences
  • Consider excluding in sensitivity analysis
  • Discuss in limitations if heterogeneity high

Overlapping CIs but Significant Overall

Individual study CIs can overlap null while pooled effect is significant:

  • Meta-analysis increases power through pooling
  • Consistent direction across studies strengthens evidence
  • Not a contradiction, reflects increased precision

Common Mistakes to Avoid

  • Comparing CI overlap instead of using statistical tests
  • Over-interpreting subgroup analyses (especially post-hoc)
  • Ignoring high heterogeneity without explanation
  • Not checking for publication bias
  • Using fixed-effect model when heterogeneity is present
  • Including same data twice (multiple publications from one study)

A good forest plot tells the story of your meta-analysis at a glance. Invest time in clear labeling and formatting. Always report heterogeneity statistics alongside the plot and discuss clinical as well as statistical significance.

References