Skip to main content

Table 1 Overview of the methods applied to synthesise the evidence from network meta-analyses

From: An investigation of the impact of using different methods for network meta-analysis: a protocol for an empirical evaluation

Method label

Package used in R

Contrast-level or arm-level input data

Frequentist or Bayesian framework

Likelihood and link functions

Heterogeneity

Prior distributions

Treatment-specific fixed effects

Mean effect of treatment k relative to baseline

Heterogeneity or random effects parameter

Contrast-synthesis model 1

gemtc (version 0.8.1)

Arm-level

Bayesian

Binomial likelihood and logit link

Homogeneous/common

N/A

δ k  ~ N(0, (15*5)2)a

τ bk  ~ U(0,10)b

Contrast-synthesis model 2

gemtc (version 0.8.1)

Arm-level

Bayesian

Binomial likelihood and logit link

Homogeneous/common

N/A

δ k  ~ N(0, (15*5)2)a

Informativec

Contrast-synthesis model 3

netmeta (version 0.9-2)

Contrast-level

Frequentist

N/A

Homogeneous/common

N/A

N/A

N/A

Arm-synthesis model 1d

pcnetmeta (version 2.4)

Arm-level

Bayesian

Binomial likelihood and probit link

Homogeneous/common

μ k  ~ N(0, 1000)

N/A

σ k  ~ U(0,10)

Arm-synthesis model 2e

pcnetmeta (version 2.4)

Arm-level

Bayesian

Binomial likelihood and probit link

Heterogeneous

μ k ~ N(0, 1000)

N/A

σ k ~ U(0,10)

  1. N/A not applicable
  2. aSource: documentation for gemtc package https://cran.r-project.org/package=gemtc
  3. b Ï„ bk represents the between-trial heterogeneity standard deviation in treatment effects
  4. cEach network was categorised according to the type of its included treatment comparisons and outcomes [21]. Specifically, in the presence of placebo in the network, the network was categorised as pharmacological vs placebo. If only pharmacological treatments were available, then the network was categorised as pharmacological vs pharmacological, whereas if a non-pharmacological treatment was included in the network, then the network was categorised as non-pharmacological vs any category. Outcomes were categorised as all-cause mortality, subjective, or semi-objective. The predictive distributions for between-trial heterogeneity variance for each of the treatment comparison by outcome type categories, estimated in Turner et al. [16], were used as informative priors
  5. dModel assumes homogeneity of the variances (i.e. common variance) of the random effects and assumes that the off-diagonal elements of the correlation matrix are equal (specified by the hom_eqcor option)
  6. eModel assumes an unstructured covariance matrix of the random effects and assumes that the off-diagonal elements of the correlation matrix are equal (specified by the het_eqcor option)