posterior_ranks {multinma}

Treatment rankings and rank probabilities

Description

Produce posterior treatment rankings and rank probabilities from a fitted NMA model. When a meta-regression is fitted with effect modifier interactions with treatment, these will differ by study population.

Usage

posterior_ranks(
  x,
  newdata = NULL,
  study = NULL,
  lower_better = TRUE,
  probs = c(0.025, 0.25, 0.5, 0.75, 0.975),
  sucra = FALSE,
  summary = TRUE
)

posterior_rank_probs(
  x,
  newdata = NULL,
  study = NULL,
  lower_better = TRUE,
  cumulative = FALSE,
  sucra = FALSE
)

Arguments

x

A stan_nma object created by nma()

newdata

Only used if a regression model is fitted. A data frame of study details, one row per study, giving the covariate values at which to produce relative effects. Column names must match variables in the regression model. If NULL, relative effects are produced for all studies in the network.

study

Column of newdata which specifies study names, otherwise studies will be labelled by row number.

lower_better

Logical, are lower treatment effects better (TRUE; default) or higher better (FALSE)? See details.

probs

Numeric vector of quantiles of interest to present in computed summary, default c(0.025, 0.25, 0.5, 0.75, 0.975)

sucra

Logical, calculate the surface under the cumulative ranking curve (SUCRA) for each treatment? Default FALSE.

summary

Logical, calculate posterior summaries? Default TRUE.

cumulative

Logical, return cumulative rank probabilities? Default is FALSE, return posterior probabilities of each treatment having a given rank. If TRUE, cumulative posterior rank probabilities are returned for each treatment having a given rank or better.

Details

The function posterior_ranks() produces posterior rankings, which have a distribution (e.g. mean/median rank and 95% Credible Interval). The function posterior_rank_probs() produces rank probabilities, which give the posterior probabilities of being ranked first, second, etc. out of all treatments.

The argument lower_better specifies whether lower treatment effects or higher treatment effects are preferred. For example, with a negative binary outcome lower (more negative) log odds ratios are preferred, so lower_better = TRUE. Conversely, for example, if treatments aim to increase the rate of a positive outcome then lower_better = FALSE.

Value

A nma_summary object if summary = TRUE, otherwise a list containing a 3D MCMC array of samples and (for regression models) a data frame of study information.

See Also

plot.nma_summary() for plotting the ranks and rank probabilities.

Examples

## Smoking cessation

# Run smoking RE NMA example if not already available
if (!exists("smk_fit_RE")) example("example_smk_re", run.donttest = TRUE)


# Produce posterior ranks
smk_rank_RE <- posterior_ranks(smk_fit_RE, lower_better = FALSE)
smk_rank_RE
plot(smk_rank_RE)

# Produce rank probabilities
smk_rankprob_RE <- posterior_rank_probs(smk_fit_RE, lower_better = FALSE)
smk_rankprob_RE
plot(smk_rankprob_RE)

# Produce cumulative rank probabilities
smk_cumrankprob_RE <- posterior_rank_probs(smk_fit_RE, lower_better = FALSE,
                                           cumulative = TRUE)
smk_cumrankprob_RE
plot(smk_cumrankprob_RE)

# Further customisation is possible with ggplot commands
plot(smk_cumrankprob_RE) +
  ggplot2::facet_null() +
  ggplot2::aes(colour = Treatment)


## Plaque psoriasis ML-NMR

# Run plaque psoriasis ML-NMR example if not already available
if (!exists("pso_fit")) example("example_pso_mlnmr", run.donttest = TRUE)


# Produce population-adjusted rankings for all study populations in
# the network

# Ranks
pso_rank <- posterior_ranks(pso_fit)
pso_rank
plot(pso_rank)

# Rank probabilities
pso_rankprobs <- posterior_rank_probs(pso_fit)
pso_rankprobs
plot(pso_rankprobs)

# Cumulative rank probabilities
pso_cumrankprobs <- posterior_rank_probs(pso_fit, cumulative = TRUE)
pso_cumrankprobs
plot(pso_cumrankprobs)

# Produce population-adjusted rankings for a different target
# population
new_agd_means <- data.frame(
  bsa = 0.6,
  prevsys = 0.1,
  psa = 0.2,
  weight = 10,
  durnpso = 3)

# Ranks
posterior_ranks(pso_fit, newdata = new_agd_means)

# Rank probabilities
posterior_rank_probs(pso_fit, newdata = new_agd_means)

# Cumulative rank probabilities
posterior_rank_probs(pso_fit, newdata = new_agd_means,
                     cumulative = TRUE)