mtc.hy.prior: Set priors for the heterogeneity parameter

In gemtc: Network Meta-Analysis Using Bayesian Methods

Set priors for the heterogeneity parameter

Description

These functions generate priors for the heterogeneity parameter in mtc.model. Priors can be set explicitly or, for outcomes on the log odds-ratio scale, based on empirical research.

Usage

mtc.hy.prior(type, distr, ...)

mtc.hy.empirical.lor(outcome.type, comparison.type)

Arguments

type	Type of heterogeneity prior: 'std.dev', 'var', or 'prec' for standard deviation, variance, or precision respectively.
distr	Prior distribution name (JAGS syntax). Typical ones would be 'dunif' (uniform), 'dgamma' (Gamma), or 'dlnorm' (log-normal). Use 'dhnorm' for the half-normal. Note that, as in JAGS, the precision (and not the variance or standard deviation) is used for the normal distribution and its derivatives.
...	Arguments to the distr. Can be numerical values or "om.scale" for the estimated outcome measure scale (see mtc.model)
outcome.type	The type of outcome to get an empirical prior for. Can be one of 'mortality' (all-cause mortality), 'semi-objective' (e.g. cause-specific mortality, major morbidity event, drop-outs), or 'subjective' (e.g. pain, mental health, dichotomous biomarkers).
comparison.type	The type of comparison to get an empirical prior for. Can be one of 'pharma-control' (pharmacological interventions versus control), 'pharma-pharma' (pharmacological versus pharmacological interventions) and 'non-pharma' (any other comparisons).

Details

The generated prior is a list, the structure of which may change without notice. It can be converted to JAGS compatible code using as.character.

Empirical priors for the log odds-ratio (LOR) are taken from [Turner et al. 2012].

Value

A value to be passed to mtc.model.

Author(s)

Gert van Valkenhoef

See Also

mtc.model

Examples

# NOTE: the mtc.run commands below are for illustrative purposes, such a small
# number of iterations should obviously not be used in practice.

# set a uniform prior standard deviation
model1 <- mtc.model(smoking, hy.prior=mtc.hy.prior("std.dev", "dunif", 0, 2))
result <- mtc.run(model1, n.adapt=10, n.iter=10)

# set an empirical (log-normal) prior on the variance
model2 <- mtc.model(smoking, hy.prior=mtc.hy.empirical.lor("subjective", "non-pharma"))
result <- mtc.run(model2, n.adapt=10, n.iter=10)

# set a gamma prior on the precision
model3 <- mtc.model(smoking, hy.prior=mtc.hy.prior("prec", "dgamma", 0.01, 0.01))
result <- mtc.run(model3, n.adapt=10, n.iter=10)

gemtc documentation built on July 9, 2023, 5:33 p.m.