[version 1; peer review: 1 approved with reservations, 1 not approved]
No competing interests were disclosed.
Bayesian methods provide a flexible framework for meta-analysis, particularly for network meta-analysis (NMA), which enables simultaneous comparison of multiple interventions and robust modeling of heterogeneity. However, implementing Bayesian meta-analysis often requires advanced programming skills.
We developed BayesMetaNMA, an open-source R/Shiny application for Bayesian pairwise and network meta-analyses. The application uses rjags for Markov Chain Monte Carlo (MCMC) estimation, netmeta for network structure visualization and frequentist NMA outputs, and meta for conventional pairwise analyses. Users can select from various effect measures (standardized mean difference, mean difference, odds ratio, risk ratio, hazard ratio), set MCMC parameters, and define prior distributions. Outputs include MCMC diagnostics, posterior summaries, study-level estimates, and network-specific analyses such as ranking tables and inconsistency checks.
BayesMetaNMA produces comprehensive outputs for both pairwise and network models, including convergence diagnostics (trace, density, autocorrelation, Gelman–Rubin plots), pooled and treatment-specific effects, heterogeneity estimates, and optional meta-regression. All plots and summaries are downloadable.
BayesMetaNMA provides a user-friendly interface for applying Bayesian methods to evidence synthesis without extensive coding. By integrating established R packages in an interactive workflow, it facilitates robust Bayesian analyses for a wide range of research applications.
Meta-analysis combines results from multiple studies to produce a pooled estimate of an effect, aiding evidence-based decision-making. While frequentist approaches remain common, Bayesian methods offer advantages such as incorporating prior knowledge, explicit probabilistic interpretation, and flexible modeling of complex evidence structures (
Network meta-analysis (NMA) extends pairwise meta-analysis by synthesizing evidence from multiple interventions, even in the absence of direct head-to-head trials. Despite their benefits, Bayesian NMAs typically require coding in specialized environments such as JAGS or Stan, which can be a barrier for applied researchers.
BayesMetaNMA addresses this challenge by providing a graphical interface for Bayesian pairwise and network meta-analyses. Built with R/Shiny, it integrates Bayesian computation, network visualization, and diagnostic tools into a single interactive platform.
BayesMetaNMA is implemented in R (
Pairwise Random-Effects Model:
Study-specific effects θ_i are modeled as: y_i ~ N(θ_i, σ_i 2), θ_i ~ N(μ, τ 2) with priors μ ~ N(μ 0, σ_μ 2), τ ~ Uniform(0, τ_max).
Network Random-Effects Model:
Treatment effects μ_j are estimated relative to a reference treatment: y_i ~ N(θ_i, σ_i
2), θ_i = (μ_t1[i] − μ_t2[i]) + δ_i, δ_i ~ N(0, τ
2), μ_j ~ N(μ
0, σ_μ
2), τ ~ Uniform(0, τ_max) (
Data are uploaded as CSV:
Pairwise: Study, Effect, SE Network: Study, Treatment1, Treatment2, Effect, SE (log transformation required for OR, RR, HR)
Example datasets are provided for all supported effect measures.
Data & Settings: Upload dataset or load an example; choose analysis type and summary measure. Priors & MCMC: Specify iterations, burn-in, and prior parameters. Run Analysis: Execute Bayesian estimation via JAGS. Explore Outputs: Convergence plots, forest plots, network diagrams, posterior summaries, heterogeneity estimates.
– Pairwise Meta-Analysis of SMD: Data loaded, priors specified, results examined via convergence diagnostics and posterior summaries. – Network Meta-Analysis of logOR: Outputs include Bayesian treatment effect estimates, heterogeneity parameters, network rankings, and probability calculations.
Unified Bayesian platform for both pairwise and network analyses; comprehensive diagnostics; rich visual outputs; configurable priors and MCMC settings; open-source.
Requires JAGS for local use; computation time for large models; limited prior distribution options; meta-regression requires user-prepared covariates; no Bayesian node-splitting.
Expanded prior specification; effect size calculation from raw data; enhanced meta-regression; Bayesian inconsistency evaluation methods.
The datasets supporting the findings of this study are openly available in Zenodo at
Source code available from:
Archived software available from:
License: MIT License
We thank the developers of R, JAGS, and the R packages shiny, bs4Dash, rjags, coda, ggplot2, igraph, netmeta, dmetar, grid, and meta.
BayesMetaNMA is an open-source R/Shiny application that enables Bayesian pairwise and network meta-analysis through an interactive GUI. It integrates JAGS for MCMC estimation and supports various effect measures, diagnostics, and visualizations. While user-friendly, the tool lacks flexibility in priors, rare-events support, and comprehensive documentation. Peer review found it underdeveloped compared to existing platforms.
1. The current implementation relies on large-sample normal approximations, which are inappropriate for binary outcomes with rare events. Please incorporate exact likelihood models (e.g., binomial with logit or cloglog links) to handle sparse data more accurately. This would improve the tool's applicability in safety or adverse-event meta-analyses.
2. The software currently offers limited prior choices, restricting users to uniform or vague normal priors. Please expand the prior library to include informed, hierarchical, or mixture priors, and allow users to define custom priors interactively. Provide guidance or warnings when priors are poorly matched to data scale or structure.
3. The literature review part can be improved by including more recent publications on machine learning methods, e.g., Bayesian method or Gaussian method, which can refer to Uncertainty-awarded, high-precision multi-step prediction of structural health monitoring sensor streams under extreme typhoon events: an enhanced Bayesian dynamic linear model leveraging the kernel regression basis function for severe environmental adaptation, Data interpretation and forecasting of SHM heteroscedastic measurements under typhoon conditions enabled by an enhanced Hierarchical sparse Bayesian Learning model with high robustness, Towards high-accuracy data modelling, uncertainty quantification and correlation analysis for SHM measurements during typhoon events using an improved most likely heteroscedastic Gaussian process, Bayesian Network in Structural Health Monitoring: Theoretical Background and Applications Review.
4. The manuscript and tool lack detailed explanations of model assumptions, output interpretation, and diagnostic thresholds. Add embedded help modules, tooltips, and interpretive summaries within the Shiny interface. Include a user manual with worked examples covering common pitfalls, such as non-convergence or inconsistency detection.
5. The NMA model indexes only two-arm trials and ignores correlation between multi-arm study contrasts. Refactor the underlying code to support full multivariate random-effects modeling, ensuring consistency and coherence in treatment rankings. Additionally, implement Bayesian inconsistency detection methods (e.g., node-splitting or design-by-treatment interaction) to enhance model validity.
Are the conclusions about the tool and its performance adequately supported by the findings presented in the article?
Yes
Is the rationale for developing the new software tool clearly explained?
Partly
Is the description of the software tool technically sound?
Partly
Are sufficient details of the code, methods and analysis (if applicable) provided to allow replication of the software development and its use by others?
Partly
Is sufficient information provided to allow interpretation of the expected output datasets and any results generated using the tool?
Yes
Reviewer Expertise:
Machine learning
I confirm that I have read this submission and believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard, however I have significant reservations, as outlined above.
Authors provide an R Shiny application for Bayesian pairwise and network meta-analyses. The provided github link is closed, and the archived code linked on Zenodo gave errors when trying to run the example data so I was not able to confirm whether the code works or produces sensible results.
The authors do not describe what gap this application is intended to fill, as freely available GUIs are already available (eg, https://crsu-metainsight.le.ac.uk/MetaInsight/) for those who can't program, and these solutions offer more capability, stronger validation, and better documentation.
The included options for analyses are all based on large sample normal approximations which will not be appropriate for binary rare events. No tools are provided to create the required summary measures from raw data.
NMA code indexes over studies K defined as rows in the underlying data frame only comparing a max of two treatments within trials. This simplifies code since they are not required to account for correlation between contrasts with the same control arm or through random effects between arms, but limits how useful the tool can be.
The manuscript itself does not provide summary of the underlying logic, outputs, or their interpretations and so it is unlikely to tool would be useful to someone who does not already have some expertise in the area the vast majority of whom would have sufficient coding ability to use existing simple packages like multinma to conduct a wider variety of analysis with superior documentation and rigour of implementation.
Are the conclusions about the tool and its performance adequately supported by the findings presented in the article?
No
Is the rationale for developing the new software tool clearly explained?
No
Is the description of the software tool technically sound?
No
Are sufficient details of the code, methods and analysis (if applicable) provided to allow replication of the software development and its use by others?
Partly
Is sufficient information provided to allow interpretation of the expected output datasets and any results generated using the tool?
No
Reviewer Expertise:
NMA methods, health economics and outcomes research
I confirm that I have read this submission and believe that I have an appropriate level of expertise to state that I do not consider it to be of an acceptable scientific standard, for reasons outlined above.