- Reference SDs can now be used when modelling using SMDs to avoid using study-specific SDs which can lead to imprecision and heterogeneity.
`rank.mbnma()`

can now only rank a single parameter (e.g.`param`

argument must be length 1). This facilitates differentiation between treatment and class parameters.

- Error in nma.run when using
`link="smd"`

- a log link function was previously used but has now been fixed - Class models did not work properly with character class labels - now fixed

- Truncated priors are used as the default priors for time-course
parameters that can only take positive values (e.g.
`et50`

) for all functions `texp()`

has been removed -`titp()`

is a more stable parameterisation of this function

- Error fixed preventing natural splines from functioning, causing removal from CRAN

- Return values added to documentation for functions in which they were missing

- Can now specify numeric values for time-course parameters in the
`method`

argument. Can be useful for discrete values that cannot be estimated (e.g. fractional polynomial powers, Hill parameter). - Fractional polynomial powers in
`tfpoly()`

can only take numeric values from set defined in Jansen 2015. - Integrated Two-Component Prediction (ITP) function
(
`titp()`

) added `get.relative()`

can be used to combine two MBNMA models to allow different time-course functions to be fitted to a different set of treatments (see examples in the vignette)- New priors that restrict posterior to positive values where necessary can be easily incorporated.
`binplot()`

can be used to plot the results of NMAs conducted at multiple time bins. This can be particularly useful to explore which time-course functions might be appropriate, and to check the validity of MBNMA predictions.`mb.nodesplit()`

can be performed at specific time-points, in addition to by time-course parameter`corparam`

set to`FALSE`

as default

- Error with
`overlay.nma`

argument in`plot.mb.predict()`

fixed

`get.relative()`

function can be used to calculate relative effects/mean differences between treatments/classes`cumrank()`

added for cumulative ranking plots. Also calculates SUCRA values for each treatment and time-course parameter at specified follow-up times (even those at which treatments have not been compared within any study)- Studies reporting change from baseline or absolute means can now be
specified in
`mb.network()`

, or will be automatically inferred from the data (studies with no time=0 are assumed to report change from baseline) - Model intercept (response at time=0) is now conditional on change
from baseline
*for each study* `texp()`

now implements 2-parameter exponential function (though the simpler 1-parameter model remains the default)

- Error with
`predict()`

not properly incorporating absolute time-course parameters fixed

- Error with
`model.file`

input length fixed for`mb.run()`

- Added variance adjustment (
`covar="varadj"`

) for correlation between time-points - this is now the default in`mb.run()`

- Added log linear time-course function (
`tloglin()`

) - Added spline functions (piecewise linear splines, B-splines, restricted cubic splines, natural splines)
- Added
`overlay.nma`

option to`predict()`

to allow plotting of “lumped” NMA results over MBNMA predictions - Modelling can now incorporate Standardised Mean Differences
(
`link="smd"`

) or Ratios of Means (`link="log"`

) to allow modelling of studies with different scales `lower_better`

argument used instead of`decreasing`

for rankings- Time-course functions given to
`mb.run()`

are now given as`class("timefun")`

and time-course parameters are specified within these functions - Predictions from
`predict()`

can now be ranked - Forest plots now also plot posterior densities using
`ggdist::stat_halfeye()`

- Neater outputs when using
`print()`

or`summary()`

- Wishart prior used to model correlations between within-study baseline effects on different time-course parameters, in addition to relative effects on different time-course parameters.

- Corrected calculation for Bayesian p-value in
`mb.nodesplit()`

- Added citation file
`plot.mb.network()`

now uses a`layout`

argument that takes an igraph layout function instead of`layout_in_circle`

(which was a logical argument). This allows any igraph layout to be plotted rather than just a circle (e.g.`igraph::as_star()`

)- Objects returned from
`plot.mb.network`

now have specific igraph attributes assigned to them, which can be easily changed by the user. `user.fun`

now takes a formula as an argument (for example`~ (beta.1 * dose) + (beta.2 * dose^2)`

) rather than a string.`mb.network`

objects are now stored within lists of most other mb class objects for easy reference of data format

- Exponential function models were not working previously but the dose-response function has been rewritten so that it runs the model correctly.
- Ensured comparisons are cycled through correctly in mb.nodesplit
- Ensured
`timeplot`

raw responses can be plotted by either arm (`plotby="arm"`

) or relative (`plotby="rel"`

) effects.

Welcome to MBNMAtime. Ready for release into the world. I hope it can be of service to you! For dose-response MBNMA, also check out the sister package, MBNMAdose.