MetaNB is an R package for Bayesian trivariate random-effects meta-analysis of prevalence, sensitivity, and specificity to indirectly synthesize net benefit across validation settings. The package also provides value-of-information (VOI) measures to quantify uncertainty around implementation decisions.
A worked introduction to the package is available here:
https://zhipeiwang.github.io/MetaNB/Introduction_to_MetaNB.html
The vignette demonstrates a typical workflow, including:
- fitting the Bayesian trivariate meta-analysis model;
- assessing convergence;
- summarizing posterior draws;
- visualizing results using forest plots; and
- obtaining value-of-information metrics.
remotes::install_github("zhipeiwang/MetaNB")The key outputs currently provided by MetaNB include:
- Posterior distributions obtained from MCMC sampling, including (but not limited to) per-setting, pooled, and predictive net benefit for model, treat-all and treat-none strategies;
- a posterior probability that a prediction model is clinically useful in a new setting;
- forest plots for net benefit, relative utility, sensitivity, and specificity;
- value-of-information metrics, including:
- expected value of perfect information (EVPI) for the overall implementation decision;
- cluster-level EVPI, where each unobserved setting is allowed to choose its own optimal strategy;
- expected value of perfect partial information (EVPPI) when the decision is to use the optimal strategy for a cluster with a given prevalence; and
- EVPI in a specific observed cluster.
MetaNB is currently under active development. Function names, arguments, and outputs may change as functionality is expanded and refined.
Feedback, bug reports, and suggestions are welcome.