Analysis Guide

AI4Meta provides a comprehensive suite of statistical tools for meta-analysis, all accessible from the Analysis tab of your project.

Forest Plot

The forest plot is the primary visualization for meta-analysis results.

• Displays individual study effects and the pooled summary estimate.
• Choose between fixed-effect (Mantel-Haenszel, inverse variance) and random-effects (DerSimonian-Laird, REML, PM) models.
• Customize: study labels, column display, color scheme, confidence interval style.
• Export as SVG, PNG, or PDF.

Funnel Plot

Assess publication bias visually and statistically.

• Standard funnel plot — effect size vs. standard error.
• Contour-enhanced funnel plot — with significance contours.
• Egger's test — statistical test for funnel plot asymmetry.
• Trim-and-fill — estimate missing studies and adjusted effect.

Subgroup Analysis

Compare effects across predefined subgroups.

• Select a categorical extraction field as the grouping variable.
• View forest plot with subgroup subtotals and overall effect.
• Test for subgroup differences (Q-between, I² for each subgroup).

Sensitivity Analysis

• Leave-one-out — recalculate the pooled effect removing each study in turn.
• Influence diagnostics — Cook's distance, DFBETAS, hat values.
• Cumulative meta-analysis — plot how the pooled effect changes as studies are added chronologically.

Network Meta-Analysis

Compare multiple interventions simultaneously.

• Network graph — visualize direct comparisons between treatments.
• League table — all pairwise comparisons in a matrix.
• SUCRA / P-score — rank treatments by probability of being best.
• Inconsistency checks — node-splitting and design-by-treatment interaction.

Heterogeneity Assessment

• I² — percentage of variability due to heterogeneity (low <25%, moderate 25–75%, high >75%).
• τ² — between-study variance estimate.
• Q statistic — Cochran's Q test for heterogeneity.
• Prediction interval — range of expected effects in a future study.

Meta-Regression

Explore sources of heterogeneity using study-level covariates.

• Select one or more continuous/categorical moderators.
• View regression coefficients, p-values, and R² analog.
• Bubble plot visualization for single-moderator models.

Reliability

Reliability in AI4Meta spans screening, extraction, and model comparisons. For the Feng-based decision tree on choosing the right metric for each variable type, see the Reliability Guide.

Supported Effect Measures