flowchart LR
A[Pairwise meta-analysis] --> B[Meta-regression and IPD-MA]
A --> C[Anchored ITC]
C --> D[MAIC]
C --> E[STC and G-computation]
A --> F[Network meta-analysis]
F --> G[Network meta-regression]
G --> H[ML-NMR]
F --> I[NMI]
C --> J[Unanchored ITC]
J --> D
J --> E
H --> K[ML-UMR]
J --> K
About MetaPlots
What this gallery is, how it was built, and how to read the labels
MetaPlots is a catalog of the graphics used, recommended, or proposed for evidence synthesis: pairwise meta-analysis, network meta-analysis (NMA), and indirect treatment comparisons, including population-adjusted methods such as matching-adjusted indirect comparison (MAIC), simulated treatment comparison (STC), multilevel network meta-regression (ML-NMR), network meta-interpolation (NMI), and multilevel unanchored meta-regression (ML-UMR). It is modeled on galleries such as the Plotly and R Graph galleries, but written for methodologists, analysts, reviewers, and health technology assessment (HTA) submissions.
Every entry has the same structure: the figure, what it shows, how to read its encodings, what the example says, how the plot is commonly misread, reproducible code, and key references.
How the figures were made
Every figure was rendered from the R script shown on its page, using published example datasets bundled with R packages (metadat, meta, netmeta, multinma, maicplus, mlumr, and others) wherever possible. A shared theme and color palette are applied by the build script before each figure script runs, so the scripts shown contain only analysis and plotting code. Stata and Python snippets are provided where an equivalent exists; they are not executed.
Some plots have no software implementation, or are only described in methodological papers and conference material. For those, the figure is a custom implementation of the published design, drawn with simulated data where no real data are available, and the caption says so. Figures from papers are never copied; proof-of-concept plots are redrawn and cited.
Bayesian examples use default sampler settings from multinma unless stated. They are demonstrations of the plots, not analyses to be cited.
Status labels
| Label | Meaning |
|---|---|
| Established | In routine use and implemented in mainstream software |
| Specialized | Tied to the machinery of one method, such as MAIC weights or ML-NMR integration |
| Adapted | A general diagnostic adapted to a method, useful but not formally validated for it |
| Proposed | Published as a new display, with software, but not yet in routine use |
| Proof of concept | Mainly used in methods papers and simulation studies |
| Contested | Published, but with credible methodological criticism that users should know |
| Emerging | Supports a method that is itself new, with conventions still forming |
Several attractive displays, such as Love plots for MAIC, convex-hull support plots, and posterior predictive panels for newer methods, are good diagnostic adaptations rather than formally validated tests. Their pages say so.
A map of the methods
Each step along this map adds assumptions: from randomized comparisons within trials, to connected networks that assume transitivity, to population-adjusted comparisons that assume every effect modifier is measured, to unanchored comparisons that also require every prognostic factor to be modeled. The graphics needed to defend a result grow accordingly. See the reporting guide for the minimum set per method.
Sources
The catalog started from a structured review of the methodological literature and current software documentation (September 2026), which covered official documentation for meta, metafor, netmeta 3.7, multinma, orchaRd, metasens, mlumr, and Stata’s meta-analysis and network suites, together with the methods papers cited on each page. Additional displays were added from the wider literature on graphics for evidence synthesis, including the review by Kossmeier, Tran, and Voracek (2020).
Contributing
Corrections and new plots are welcome. Each page links to its source on GitHub (“Edit this page”). A new entry needs a .qmd page under plots/<family>/ and an R script under R/plots/ with the same name; Rscript R/render.R <name> renders the figure and quarto render builds the site.
License
Code and text are released under the MIT License. Datasets belong to their respective packages and original authors.