long2jags: Transform data from long arm-based to JAGS format

In guido-s/nmajags: R functions to run Bayesian network meta-analysis with JAGS

Description

This function transforms data from long arm-based format, i.e., two rows for a pairwise comparison, to a list format for Bayesian network meta-analysis using JAGS.

Usage

long2jags(
  studlab,
  treat,
  event,
  n,
  mean,
  sd,
  data = NULL,
  reference.group = 1,
  covar,
  func = "max"
)

Arguments

studlab	A vector with study labels or an object created with longarm.
treat	A vector with treatment labels (mandatory if studlab is a vector).
event	Number of events.
n	Number of observations.
mean	Estimated mean.
sd	Standard deviation.
data	An optional data frame containing the study information.
reference.group	Reference treatment.
covar	Covariable to use in network meta-regression.
func	A character string specifying function to summarize data for additional covariate. Either, "min", "max", "mean", or ""; can be abbreviated.

Details

This function transforms data from long arm-based format, i.e., two rows for a pairwise comparison, to a list format for Bayesian network meta-analysis using JAGS.

At the moment, the function can be used to transform data with a binary or continuous outcome. The following arguments are mandatory if argument studlab is not an R object created with longarm. Arguments studlab and treat must be provided to identify studies and treatments and, depending on the outcome, the following additional arguments:

• event, n (binary outcome);

• n, mean, sd (continuous outcome).

Value

A list of class jagsdata.

Author(s)

Georgia Salanti [e-mail], Guido Schwarzer [e-mail]

See Also

netjags, longarm, netmeta

Examples

## Not run: 
library(netmeta)
data(Dogliotti2014)
Dogliotti2014$trt <- treats(Dogliotti2014$treatment, nchar = 4)
#
p1 <- pairwise(trt, stroke, total, studlab = study,
               data = Dogliotti2014, sm = "OR")

# Standard frequentist random effects NMA model
#
n1.iv <- netmeta(p1)

# Standard Bayesian random effects NMA model
#
library(R2jags)

# Transform long-arm based data into a list suitable for JAGS
# analysis, with placebo as reference treatment
#
dat.jags <- long2jags(study, trt,
                      event = stroke, n = total,
                      data = Dogliotti2014, 
                      reference = "pl")

# Run JAGS and create JAGS object
#
n1.jags <- netjags(dat.jags)

# Extract lnOR from BUGS output
#
trts <- attr(n1.jags, "trts")
TE.random <- n1.jags$BUGSoutput$mean$LOR.full
rownames(TE.random) <- trts
colnames(TE.random) <- trts

# Extract se(lnOR) from BUGS output
#
seTE.random <- n1.jags$BUGSoutput$sd$LOR.full
rownames(seTE.random) <- trts
colnames(seTE.random) <- trts

# Conduct frequentist NMA with tau from Bayesian NMA
#
tau.jags <- as.numeric(n1.jags$BUGSoutput$mean$tau)
n1.iv.jags <-
  suppressWarnings(netmeta(p1, tau.preset = tau.jags))

# Print treatment matrix for frequentist NMA
#
round(n1.iv$TE.random, 3)[trts, trts]

# Print treatment matrix for Bayesian NMA
#
round(TE.random, 3)

# Print treatment matrix for frequentist NMA
# (using tau from Bayesian NMA)
#
round(n1.iv.jags$TE.random, 3)[trts, trts]

# Print standard errors from frequentist NMA
#
round(n1.iv$seTE.random, 3)[trts, trts]

# Print standard errors from Bayesian NMA
#
round(seTE.random, 3)

# Print standard errors from frequentist NMA
# (using tau from Bayesian NMA)
#
round(n1.iv.jags$seTE.random, 3)[trts, trts]

# Print square root of between-study heterogeneity tau2
#
tau.jags
n1.iv$tau

## End(Not run)

guido-s/nmajags documentation built on Jan. 15, 2022, 12:35 a.m.