The workflow uses a frequentist Network Meta-Analysis framework implemented through the netmeta R package.

Data structure

The current pipeline expects pairwise contrast data with the following columns:

study, treat1, treat2, logHR, selogHR

Here, logHR is the log hazard ratio and selogHR is its standard error.

Models

The original Shiny app supports three analysis modes:

  1. Simple treatment-level NMA using netmeta().
  2. Additive Component Network Meta-Analysis using netcomb() or discomb().
  3. Interaction CNMA using a component interaction C-matrix when the network is connected.
Code 
source("R/nma_functions.R")
source("R/node_merging.R")
dat <- load_pairwise_data()
components <- extract_components(dat)
knitr::kable(data.frame(components = components))
components
Acal
BR
Clb
FCR
I
Ibr
O
R
Ven
Zanubrutinib

Connectivity logic

For CNMA, the app checks whether the treatment network is connected using igraph. If the network is connected, netcomb() is used. If it is disconnected, discomb() is used. Interaction CNMA is only attempted on connected networks.