Forest Plot Builder

Paste study-level effect sizes and standard errors; get a fixed-effect and random-effects meta-analysis with forest, funnel, and influence diagnostics

Purpose

Meta-analysis software is full of sensible defaults that obscure the modelling choices a reviewer will scrutinise: fixed vs. random, which tau-squared estimator, which confidence interval adjustment, which publication-bias assessment. The Forest Plot Builder exposes each choice and shows how the summary estimate and its uncertainty move in response.

User inputs

• Effect-size metric (mean difference, SMD, log-RR, log-OR, log-HR, Pearson \(r\))
• Data entry: paste a table of study-level estimates with standard errors (or 2x2 counts for binary outcomes)
• Model: fixed-effect, random-effects with choice of tau-squared estimator (DL, REML, PM, SJ)
• Adjustment: Hartung-Knapp, Knapp-Hartung, Sidik-Jonkman
• Sensitivity: leave-one-out, influence diagnostics

Outputs

• Forest plot with study-level and pooled estimates, customisable ordering and columns
• Heterogeneity statistics: \(Q\), \(I^2\), \(\tau^2\), \(H^2\), 95% prediction interval
• Funnel plot with Egger’s and Begg’s tests
• Leave-one-out summary: how much each study moves the pooled estimate
• Cumulative meta-analysis ordered by year

Didactic value

Switching from fixed-effect to random-effects on a highly heterogeneous dataset shows the confidence interval widen dramatically – a concrete demonstration of why the “wrong” model underestimates uncertainty. The leave-one-out view teaches sensitivity analysis as a practice, not an abstract nod in the methods section.

Embedded in

• meta-analysis/fixed-vs-random-effects.md
• meta-analysis/forest-plots.md
• meta-analysis/heterogeneity.md

Source code

Local: apps/15-forest-plot-builder/

Run with:

shiny::runApp("apps/15-forest-plot-builder")