mtc.run {gemtc}

Running an mtc.model using an MCMC sampler

Description

The function mtc.run is used to generate samples from a object of type mtc.model using a MCMC sampler. The resulting mtc.result object can be coerced to an mcmc.list for further analysis of the dataset using the coda package.

Usage

mtc.run(model, sampler = NA, n.adapt = 5000, n.iter = 20000, thin = 1)

## S3 method for class 'mtc.result'
summary(object, ...)
## S3 method for class 'mtc.result'
plot(x, ...)
## S3 method for class 'mtc.result'
forest(x, use.description=FALSE, ...)
## S3 method for class 'mtc.result'
print(x, ...)
## S3 method for class 'mtc.result'
as.mcmc.list(x, ...)

Arguments

model

An object of S3 class mtc.model describing a network meta-analysis model.

sampler

Deprecated: gemtc now only supports the JAGS sampler. Specifying a sampler will result in a warning or error. This argument will be removed in future versions.

n.adapt

Amount of adaptation (or tuning) iterations.

n.iter

Amount of simulation iterations.

thin

Thinning factor.

object

Object of S3 class mtc.result.

x

Object of S3 class mtc.result.

use.description

Display treatment descriptions instead of treatment IDs.

...

Additional arguments.

Value

An object of class mtc.result. This is a list with the following elements:

samples

The samples resulting from running the MCMC model, in mcmc.list format.

model

The mtc.model used to produce the samples.

deviance

Residual deviance statistics, a list with the following elements. DIC: deviance information criterion at residual (Dbar + pD). Dbar: mean sum of residual deviance. pD: sum of leverage, also known as the effective number of parameters. dev.ab: mean posterior residual deviance of each arm (for arm-based data). fit.ab: deviance at the posterior mean of te fitted values (for arm-based data). dev.re: mean posterior residual deviance of each study (for relative-effect data). fit.re: deviance at the posterior mean of the fitted values (for relative-effect data).

The object can be coerced to an mcmc.list from the coda package by the generic S3 method as.mcmc.list.

Analysis of the results

Convergence of the model can be assessed using methods from the coda package. For example the Brooks-Gelman-Rubin method (coda::gelman.diag, coda::gelman.plot). The summary also provides useful information, such as the MCMC error and the time series and densities given by plot should also be inspected.

The forest function can provide forest plots for mtc.result objects. This is especially useful in combination with the relative.effect function that can be used to calculate relative effects compared to any baseline for consistency models. The rank.probability function calculates rank probabilities for consistency models.

Author(s)

Gert van Valkenhoef, Joël Kuiper

See Also

mtc.model

relative.effect.table, relative.effect, rank.probability

coda::gelman.diag, coda::gelman.plot

Examples

model <- mtc.model(smoking)

## Not run: results <- mtc.run(model, thin=10)
results <- readRDS(system.file("extdata/luades-smoking-samples.rds", package="gemtc"))

# Convergence diagnostics
gelman.plot(results)

# Posterior summaries
summary(results)
## Iterations = 5010:25000
## Thinning interval = 10 
## Number of chains = 4 
## Sample size per chain = 2000 
## 
## 1. Empirical mean and standard deviation for each variable,
##    plus standard error of the mean:
## 
##         Mean     SD Naive SE Time-series SE
## d.A.B 0.4965 0.4081 0.004563       0.004989
## d.A.C 0.8359 0.2433 0.002720       0.003147
## d.A.D 1.1088 0.4355 0.004869       0.005280
## sd.d  0.8465 0.1913 0.002139       0.002965
## 
## 2. Quantiles for each variable:
## 
##          2.5%    25%    50%    75% 97.5%
## d.A.B -0.2985 0.2312 0.4910 0.7530 1.341
## d.A.C  0.3878 0.6720 0.8273 0.9867 1.353
## d.A.D  0.2692 0.8197 1.0983 1.3824 2.006
## sd.d   0.5509 0.7119 0.8180 0.9542 1.283

plot(results) # Shows time-series and density plots of the samples
forest(results) # Shows a forest plot