predict.stan_nma: Predictions of absolute effects from NMA models

In multinma: Bayesian Network Meta-Analysis of Individual and Aggregate Data

Predictions of absolute effects from NMA models

Description

Obtain predictions of absolute effects from NMA models fitted with nma(). For example, if a model is fitted to binary data with a logit link, predicted outcome probabilities or log odds can be produced.

Usage

## S3 method for class 'stan_nma'
predict(
  object,
  ...,
  baseline = NULL,
  newdata = NULL,
  study = NULL,
  trt_ref = NULL,
  type = c("link", "response"),
  level = c("aggregate", "individual"),
  baseline_type = c("link", "response"),
  baseline_level = c("individual", "aggregate"),
  probs = c(0.025, 0.25, 0.5, 0.75, 0.975),
  predictive_distribution = FALSE,
  summary = TRUE
)

Arguments

object	A stan_nma object created by nma().
...	Additional arguments, passed to uniroot() for regression models if baseline_level = "aggregate".
baseline	An optional distr() distribution for the baseline response (i.e. intercept), about which to produce absolute effects. If NULL, predictions are produced using the baseline response for each study in the network with IPD or arm-based AgD. For regression models, this may be a list of distr() distributions of the same length as the number of studies in newdata (possibly named by the study names, or otherwise in order of appearance in newdata). Use the baseline_type and baseline_level arguments to specify whether this distribution is on the response or linear predictor scale, and (for ML-NMR or models including IPD) whether this applies to an individual at the reference level of the covariates or over the entire newdata population, respectively. For example, in a model with a logit link with baseline_type = "link", this would be a distribution for the baseline log odds of an event. Use the trt_ref argument to specify which treatment this distribution applies to.
newdata	Only required if a regression model is fitted and baseline is specified. A data frame of covariate details, for which to produce predictions. Column names must match variables in the regression model. If level = "aggregate" this should either be a data frame with integration points as produced by add_integration() (one row per study), or a data frame with individual covariate values (one row per individual) which are summarised over. If level = "individual" this should be a data frame of individual covariate values, one row per individual. If NULL, predictions are produced for all studies with IPD and/or arm-based AgD in the network, depending on the value of level.
study	Column of newdata which specifies study names or IDs. When not specified: if newdata contains integration points produced by add_integration(), studies will be labelled sequentially by row; otherwise data will be assumed to come from a single study.
trt_ref	Treatment to which the baseline response distribution refers, if baseline is specified. By default, the baseline response distribution will refer to the network reference treatment. Coerced to character string.
type	Whether to produce predictions on the "link" scale (the default, e.g. log odds) or "response" scale (e.g. probabilities).
level	The level at which predictions are produced, either "aggregate" (the default), or "individual". If baseline is not specified, predictions are produced for all IPD studies in the network if level is "individual" or "aggregate", and for all arm-based AgD studies in the network if level is "aggregate".
baseline_type	When a baseline distribution is given, specifies whether this corresponds to the "link" scale (the default, e.g. log odds) or "response" scale (e.g. probabilities).
baseline_level	When a baseline distribution is given, specifies whether this corresponds to an individual at the reference level of the covariates ("individual", the default), or from an (unadjusted) average outcome on the reference treatment in the newdata population ("aggregate"). Ignored for AgD NMA, since the only option is "aggregate" in this instance.
probs	Numeric vector of quantiles of interest to present in computed summary, default c(0.025, 0.25, 0.5, 0.75, 0.975)
predictive_distribution	Logical, when a random effects model has been fitted, should the predictive distribution for absolute effects in a new study be returned? Default FALSE.
summary	Logical, calculate posterior summaries? Default TRUE.

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 predictions.

Examples

## Smoking cessation

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


# Predicted log odds of success in each study in the network
predict(smk_fit_RE)

# Predicted probabilities of success in each study in the network
predict(smk_fit_RE, type = "response")

# Predicted probabilities in a population with 67 observed events out of 566
# individuals on No Intervention, corresponding to a Beta(67, 566 - 67)
# distribution on the baseline probability of response, using
# `baseline_type = "response"`
(smk_pred_RE <- predict(smk_fit_RE,
                        baseline = distr(qbeta, 67, 566 - 67),
                        baseline_type = "response",
                        type = "response"))
plot(smk_pred_RE, ref_line = c(0, 1))

# Predicted probabilities in a population with a baseline log odds of
# response on No Intervention given a Normal distribution with mean -2
# and SD 0.13, using `baseline_type = "link"` (the default)
# Note: this is approximately equivalent to the above Beta distribution on
# the baseline probability
(smk_pred_RE2 <- predict(smk_fit_RE,
                         baseline = distr(qnorm, mean = -2, sd = 0.13),
                         type = "response"))
plot(smk_pred_RE2, ref_line = c(0, 1))


## 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)


# Predicted probabilities of response in each study in the network
(pso_pred <- predict(pso_fit, type = "response"))
plot(pso_pred, ref_line = c(0, 1))

# Predicted probabilites of response in a new target population, with means
# and SDs or proportions given by
new_agd_int <- data.frame(
  bsa_mean = 0.6,
  bsa_sd = 0.3,
  prevsys = 0.1,
  psa = 0.2,
  weight_mean = 10,
  weight_sd = 1,
  durnpso_mean = 3,
  durnpso_sd = 1
)

# We need to add integration points to this data frame of new data
# We use the weighted mean correlation matrix computed from the IPD studies
new_agd_int <- add_integration(new_agd_int,
                               durnpso = distr(qgamma, mean = durnpso_mean, sd = durnpso_sd),
                               prevsys = distr(qbern, prob = prevsys),
                               bsa = distr(qlogitnorm, mean = bsa_mean, sd = bsa_sd),
                               weight = distr(qgamma, mean = weight_mean, sd = weight_sd),
                               psa = distr(qbern, prob = psa),
                               cor = pso_net$int_cor,
                               n_int = 1000)

# Predicted probabilities of achieving PASI 75 in this target population, given
# a Normal(-1.75, 0.08^2) distribution on the baseline probit-probability of
# response on Placebo (at the reference levels of the covariates), are given by
(pso_pred_new <- predict(pso_fit,
                         type = "response",
                         newdata = new_agd_int,
                         baseline = distr(qnorm, -1.75, 0.08)))
plot(pso_pred_new, ref_line = c(0, 1))

multinma documentation built on May 31, 2023, 5:46 p.m.