Feature: Class effects models are now available, thanks to @sjperren (PR #37). The
new class_effects argument in nma() allows
models with independent, exchangeable, or common class effects to be
fitted. Class standard deviations can also be shared between classes or
subsets of classes, controlled by the class_sd argument.
These features are demonstrated in a new vignette, analysing a network
of interventions for social anxiety.
Feature: Leverage plots can now be produced by
plot.nma_dic(), with the option
type = "leverage".
Feature: Networks with integration points can now be combined with
combine_network(), where previously these were discarded.
One potential use case is to specify different types of marginal
distributions or correlation structures for different AgD studies in the
network, by setting these up separately with
add_integration() before combining with
combine_network().
Feature: Added as.data.frame.nma_dic() and
as_tibble.nma_dic() methods that return data frames of the
pointwise contributions to the DIC, and as.matrix.nma_dic()
and as.array.nma_dic() methods that return posterior
samples of the residual deviances as a matrix or 3D array.
Feature: The softmax() and inv_softmax()
transforms are now exported.
Improvement: Removed suggested dependency on the logitnorm package.
The logit Normal distribution functions are now implemented
internally.
Fix: Resolved a bug where trying to fit meta-regression models with
discrete covariates would sometimes result in a misspecified and
inestimable model, due to the inclusion of additional columns in the
design matrix for the reference level of the covariates.
Fix: IPD Poisson models were broken due to an incorrect offset for
log time at risk (thanks to @n8thangreen for spotting this).
Fix: In a Binomial model, studies where everyone experienced the
outcome (r = n on all arms) no longer give NaN
residual deviance.
multinma 0.7.2
Fix: Predictions for non-proportional hazards IPD NMA or ML-NMR
survival models using aux_regression = ~.trt were
incorrectly omitting the treatment effects on the auxiliary parameter(s)
in some cases (#43).
Fix: Calling marginal_effects() for survival outcomes
with a single target population previously gave an error.
Fix: Predictions from exponential models where
aux_regression had been specified were giving an error
(#44). aux_regression and aux_by have no
effect for exponential models since there are no auxiliary (shape)
parameters and are ignored, now with a warning.
Fix: Avoid an error when trying to fit M-spline models combining IPD
and AgD in R versions prior to 4.1.0, due to integer coercion of factors
by c().
multinma 0.7.1
Fix: Producing survival/hazard/cumulative hazard predictions for
survival models with predict() outside of a
plot() call no longer gives an error (#40).
Fix: Increased StanHeaders version requirement to version 2.32.9 or
later, to avoid CRAN sanitizer warnings (caused by
stan-dev/rstan#1111).
multinma 0.7.0
Feature: The new marginal_effects() function produces
marginal treatment effects, as a wrapper around absolute predictions
from predict(). For example, for an analysis with a binary
outcome marginal odds ratios, risk ratios, or risk differences may be
produced. For survival outcomes, marginal effects may be based on the
full range of predictions produced by predict(), such as
marginal differences in restricted mean survival times, or time-varying
marginal hazard ratios.
Feature: Progress bars are now displayed when running interactively
for calculations with predict() or
marginal_effects() from ML-NMR models that may take longer
to run. These can be controlled with the new progress
argument.
Deprecation: The trt_ref argument to
predict() has been renamed to baseline_ref;
using trt_ref is now soft-deprecated. Renaming this
argument baseline_ref follows the naming convention for the
other arguments (baseline_type,
baseline_level) that specify the details of a provided
baseline distribution. This also makes way for the new
marginal_effects() functionality.
Fix: Fallback formatting used by print methods when the crayon
package is not installed now works properly, rather than giving
errors.
Fix: Small bug caused predict() for AgD meta-regression
models with new data and baseline_type = "response" to fail
with an error.
Fix: The number of studies on a contrast in a network plot
plot.nma_data() with weight_edges = TRUE was
incorrect when a study had multiple arms of the same treatment. This now
correctly counts the number of studies making a comparison, rather than
the number of arms.
multinma 0.6.1
Fix: Piecewise exponential hazard models no longer give errors
during set-up. Calculation of RW1 prior weights needed to be handled as
a special case.
multinma 0.6.0
Feature:
Survival/time-to-event models are now supported
set_ipd() now has a Surv argument for
specifying survival outcomes using survival::Surv(), and a
new function set_agd_surv() sets up aggregate data in the
form of event/censoring times (e.g. from digitized Kaplan-Meier curves)
and overall covariate summaries.
Left, right, and interval censoring as well as left truncation
(delayed entry) are all supported.
The available likelihoods are Exponential (PH and AFT forms),
Weibull (PH and AFT forms), Gompertz, log-Normal, log-Logistic, Gamma,
Generalised Gamma, flexible M-splines on the baseline hazard, and
piecewise exponential hazards.
Auxiliary parameters (e.g. shapes, spline coefficients) are always
stratified by study to respect randomisation, and may be further
stratified by treatment (e.g. to relax the proportional hazards
assumption) and/or by additional factors using the aux_by
argument to nma().
A regression model may be defined for the auxiliary parameters using
the aux_regression argument to nma(), allowing
non-proportionality to be modelled by treatment and/or covariate effects
on the shapes or spline coefficients.
The predict() method produces estimates of survival
probabilities, hazards, cumulative hazards, mean survival times,
restricted mean survival times, quantiles of the survival time
distribution, and median survival times. All of these predictions can be
plotted using the plot() method.
The geom_km() function assists in plotting Kaplan-Meier
curves from a network object, for example to overlay these on estimated
survival curves. The transform argument can be used to
produce log-log plots for assessing the proportional hazards assumption,
along with cumulative hazards or log survival curves.
A new vignette demonstrates ML-NMR survival analysis with an example
of progression-free survival after autologous stem cell transplant for
newly diagnosed multiple myeloma, with corresponding datasets
ndmm_ipd, ndmm_agd, and
ndmm_agd_covs.
Feature:
Automatic checking of numerical integration for ML-NMR models
The accuracy of numerical integration for ML-NMR models can now be
checked automatically, and is by default. To do so, half of the chains
are run with n_int and half with n_int/2
integration points. Any Rhat or effective sample size warnings can then
be ascribed to either: non-convergence of the MCMC chains, requiring
increased number of iterations iter in nma(),
or; insufficient accuracy of numerical integration, requiring increased
number of integration points n_int in
add_integration(). Descriptive warning messages indicate
which is the case.
This feature is controlled by a new int_check argument
to nma(), which is enabled (TRUE) by
default.
Saving thinned cumulative integration points can now be disabled
with int_thin = 0, and is now disabled by default. The
previous default was int_thin = max(n_int %/% 10, 1).
Because we can now check sufficient accuracy automatically, the
default number of integration points n_int in
add_integration() has been lowered to 64. This is still a
conservative choice, and will be sufficient in many cases; the previous
default of 1000 was excessive.
As a result, ML-NMR models are now much faster to run by default,
both due to lower n_int and disabling saving cumulative
integration points.
Other updates
Feature: dic() now includes an option to use the pV
penalty instead of pD.
Feature: The baseline and aux arguments to
predict() can now be specified as the name of a study in
the network, to use the parameter estimates from that study for
prediction.
Improvement: predict() will now produce aggregate-level
predictions over a sample of individuals in newdata for
ML-NMR models (previously newdata had to include
integration points).
Improvement: Compatibility with future rstan versions (PR #25).
Improvement: Added a plot.mcmc_array() method, as a
shortcut for plot(summary(x), ...).
Fix: In plot.nma_data(), using a custom
layout that is not a string (e.g. a data frame of layout
coordinates) now works as expected when nudge > 0.
Fix: Documentation corrections (PR #24).
Fix: Added missing as.tibble.stan_nma() and
as_tibble.stan_nma() methods, to complement the existing
as.data.frame.stan_nma().
Fix: Bug in ordered multinomial models where data in studies with
missing categories could be assigned the wrong category (#28).
multinma 0.5.1
Fix: Now compatible with latest StanHeaders v2.26.25 (fixes
#23)
Fix: Dealt with various tidyverse deprecations
Fix: Updated TSD URLs again (thanks to @ndunnewind)
multinma 0.5.0
Feature: Treatment labels in network plots can now be nudged away
from the nodes when weight_nodes = TRUE, using the new
nudge argument to plot.nma_data() (#15).
Feature: The data frame returned by calling as_tibble()
or as.data.frame() on an nma_summary object
(such as relative effects or predictions) now includes columns for the
corresponding treatment (.trt) or contrast
(.trta and .trtb), and a
.category column may be included for multinomial models.
Previously these details were only present as part of the
parameter column
Feature: Added log t prior distribution
log_student_t(), which can be used for positive-valued
parameters (e.g. heterogeneity variance).
Improvement: set_agd_contrast() now produces an
informative error message when the covariance matrix implied by the
se column is not positive definite. Previously this was
only checked by Stan after calling the nma() function.
Improvement: Updated plaque psoriasis ML-NMR vignette to include new
analyses, including assessing the assumptions of population adjustment
and synthesising multinomial outcomes.
Improvement: Improved behaviour of the .trtclass
special in regression formulas, now main effects of
.trtclass are always removed since these are collinear with
.trt. This allows expansion of interactions with
* to work properly, e.g. ~variable*.trtclass,
whereas previously this resulted in an over-parametrised model.
Fix: CRAN check note for manual HTML5 compatibility.
Fix: Residual deviance and log likelihood parameters are now named
correctly when only contrast-based aggregate data is present (PR
#19).
multinma 0.4.2
Fix: Error in get_nodesplits() when studies have
multiple arms of the same treatment.
Fix: print.nma_data() now prints the repeated arms when
studies have multiple arms of the same treatment.
Fix: CRAN warning regarding invalid img tag height attribute in
documentation.
multinma 0.4.1
Fix: tidyr v1.2.0 breaks ordered multinomial models when some
studies do not report all categories (i.e. some multinomial category
outcomes are NA in multi()) (PR #11)
multinma 0.4.0
Feature: Node-splitting models for assessing inconsistency are now
available with consistency = "nodesplit" in
nma(). Comparisons to split can be chosen using the
nodesplit argument, by default all possibly inconsistent
comparisons are chosen using get_nodesplits().
Node-splitting results can be summarised with
summary.nma_nodesplit() and plotted with
plot.nodesplit_summary().
Feature: The correlation matrix for generating integration points
with add_integration() for ML-NMR models is now adjusted to
the underlying Gaussian copula, so that the output correlations of the
integration points better match the requested input correlations. A new
argument cor_adjust controls this behaviour, with options
"spearman", "pearson", or "none".
Although these correlations typically have little impact on the results,
for strict reproducibility the old behaviour from version 0.3.0 and
below is available with cor_adjust = "legacy".
Feature: For random effects models, the predictive distribution of
relative/absolute effects in a new study can now be obtained in
relative_effects() and predict.stan_nma()
respectively, using the new argument
predictive_distribution = TRUE.
Feature: Added option to calculate SUCRA values when summarising the
posterior treatment ranks with posterior_ranks() or
posterior_rank_probs(), when argument
sucra = TRUE.
Improvement: Factor order is now respected when trt,
study, or trt_class are factors, previously
the order of levels was reset into natural sort order.
Improvement: Update package website to Bootstrap 5 with release of
pkgdown 2.0.0
Fix: Model fitting is now robust to non-default settings of
options("contrasts").
Fix: plot.nma_data() no longer gives a ggplot
deprecation warning (PR #6).
Fix: Bug in predict.stan_nma() with a single covariate
when newdata is a data.frame (PR #7).
Fix: Attempting to call predict.stan_nma() on a
regression model with only contrast data and no newdata or
baseline specified now throws a descriptive error
message.
multinma 0.3.0
Feature: Added baseline_type and
baseline_level arguments to
predict.stan_nma(), which allow baseline distributions to
be specified on the response or linear predictor scale, and at the
individual or aggregate level.
Feature: The baseline argument to
predict.stan_nma() can now accept a (named) list of
baseline distributions if newdata contains multiple
studies.
Improvement: Misspecified newdata arguments to
functions like relative_effects() and
predict.stan_nma() now give more informative error
messages.
Fix: Constructing models with contrast-based data previously gave
errors in some scenarios (ML-NMR models, UME models, and in some cases
AgD meta-regression models).
Fix: Ensure CRAN additional checks with --run-donttest
run correctly.
multinma 0.2.1
Fix: Producing relative effect estimates for all contrasts using
relative_effects() with all_contrasts = TRUE
no longer gives an error for regression models.
Fix: Specifying the covariate correlation matrix cor in
add_integration() is not required when only one covariate
is present.
Improvement: Added more detailed documentation on the likelihoods
and link functions available for each data type (likelihood
and link arguments in nma()).
multinma 0.2.0
Feature: The set_*() functions now accept
dplyr::mutate() style semantics, allowing inline variable
transformations.
Feature: Added ordered multinomial models, with helper function
multi() for specifying the outcomes. Accompanied by a new
data set hta_psoriasis and vignette.
Feature: Implicit flat priors can now be specified, on any
parameter, using flat().
Improvement: as.array.stan_nma() is now much more
efficient, meaning that many post-estimation functions are also now much
more efficient.
Improvement: plot.nma_dic() is now more efficient,
particularly with large numbers of data points.
Improvement: The layering of points when producing “dev-dev” plots
using plot.nma_dic() with multiple data types has been
reversed for improved clarity (now AgD over the top of IPD).
Improvement: Aggregate-level predictions with predict()
from ML-NMR / IPD regression models are now calculated in a much more
memory-efficient manner.
Improvement: Added an overview of examples given in the
vignettes.
Improvement: Network plots with weight_edges = TRUE no
longer produce legends with non-integer values for the number of
studies.
Fix: plot.nma_dic() no longer gives an error when
attempting to specify .width argument when producing
“dev-dev” plots.
multinma 0.1.3
Format DESCRIPTION to CRAN requirements
multinma 0.1.2
Wrapped long-running examples in \donttest{} instead of
\dontrun{}
multinma 0.1.1
Reduced size of vignettes
Added methods paper reference to DESCRIPTION
Added zenodo DOI
multinma 0.1.0
Feature: Network plots, using a plot() method for
nma_data objects.
Feature: as.igraph(), as_tbl_graph()
methods for nma_data objects.
Feature: Produce relative effect estimates with
relative_effects(), posterior ranks with
posterior_ranks(), and posterior rank probabilities with
posterior_rank_probs(). These will be study-specific when a
regression model is given.
Feature: Produce predictions of absolute effects with a
predict() method for stan_nma objects.
Feature: Plots of relative effects, ranks, predictions, and
parameter estimates via plot.nma_summary().
Feature: Optional sample_size argument for
set_agd_*() that:
Enables centering of predictors (center = TRUE) in
nma() when a regression model is given, replacing the
agd_sample_size argument of nma()
Enables production of study-specific relative effects, rank
probabilities, etc. for studies in the network when a regression model
is given
Allows nodes in network plots to be weighted by sample size
Feature: Plots of residual deviance contributions for a model and
“dev-dev” plots comparing residual deviance contributions between two
models, using a plot() method for nma_dic
objects produced by dic().
Feature: Complementary log-log (cloglog) link function
link = "cloglog" for binomial likelihoods.
Feature: Option to specify priors for heterogeneity on the standard
deviation, variance, or precision, with argument
prior_het_type.
Feature: Added log-Normal prior distribution.
Feature: Plots of prior distributions vs. posterior distributions
with plot_prior_posterior().
Feature: Pairs plot method pairs().
Feature: Added vignettes with example analyses from the NICE TSDs
and more.
Fix: Random effects models with even moderate numbers of studies
could be very slow. These now run much more quickly, using a sparse
representation of the RE correlation matrix which is automatically
enabled for sparsity above 90% (roughly equivalent to 10 or more
studies).