Changelog
Source:NEWS.md
Version 0.8.15 (2024/06/29)
 Random walk prior for the coefficients changed to a firstorder weighted model, following Phillippo et al., so that the expected smoothness accounts for the knot spacing.
Version 0.8.13 (2024/03/05)
 Bug fix for models with external data and either nonproportional hazards or covariates on cure probability.
Version 0.8.12 (2024/01/06)

irmst
,hazard_ratio
andhrtime
tidied. Fix of bug inirmst
introduced in v0.8.9, and all now have asample
argument.
Version 0.8.11 (2023/12/28)
 Outputs standardised over the covariate values of a reference population can now be produced using the new
standardise_to()
function.
Version 0.8.10 (2023/12/24)
loo
no longer ignores covariates on the cure probability in cure models.More internal refactoring of code to produce outputs.
Version 0.8.9 (2023/12/22)
Cleaned and simplified internal code for model outputs (
rmst
,survival
etc.).rmst
now outputs a tibble, and credible limits in the default output are renamed tolower
andupper
for consistency withsurvival
andhazard
.Arrays now allowed for
coef
argument indsurvspline
and related functions.
Version 0.8.8 (2023/12/19)
Background hazards in additive hazards models can now be stratified, using the new
backhaz_strata
argument tosurvextrap()
. This argument names stratifying variables (e.g. age group) that should be in both the background hazards and the individual and/or external data.Bug fix for
hsurvmspline_wane
anddsurvmspline_wane
where offsets were not implemented correctly. This affected hazard predictions for models that included both background hazards and treatment effect waning.
Version 0.8.7 (2023/10/29)
 A random walk prior can now be specified for the spline coefficients, by
smooth_model = "random_walk"
.
Version 0.8.6 (2023/10/24)
Nonproportional hazards models can now be applied to a subset of the covariates, by supplying a formula as the
nonprop
argument tosurvextrap
.Prior simulation functions such as
prior_sample
now require an explicit covariate model, specified throughformula
andnewdata
, rather than a design matrixX
. These functions now fully support nonproportional hazards models and cure probabilities with associated regression models.
Version 0.8.5 (2023/09/30)
 Code for constructing
mspline
specification objects tidied, with newmspline_init
andmspline_list_init
functions.
Version 0.8.4 (2023/09/09)
 Deprecated Stan array syntax updated. The package now requires
rstan
version 2.26.
Version 0.8.2 (2023/06/14)
 Individual data can now be excluded, so that the model is fitted to external data alone.
Version 0.8 (2023/05/06)
The GitHub commit on 6 May 2023 implemented several changes that are listed here. survextrap
is now in “beta” status. All major features are now implemented, but there may still be some lack of polish.
New features
A new vignette
cetuximab
, giving an indepth case study of a realistic use ofsurvextrap
. To be included in a forthcoming paper.
Improvements to usability and stability of Msplines:
iknots
andbknots
are replaced by a singleknots
, and the lower boundary knot is always fixed at zero.add_knots
can be used to add userdefined knots to the default ones. See the case study.New function
mspline_spec
to define Mspline knots based on data, in advance of fitting any models.Msplines can now be smoother at the final knot, through an option
bsmooth
, which is on by default. Thanks to Iain Timmins for the suggestion.
Incremental restricted mean survival time (
irmst
) added.Any summary of the posterior distribution can be produced by supplying a list of summary functions (e.g.
mean
,quantiles
) to output functions (such asrmst
andsurvival
).Leaveoneout cross validation for external data.
Bug fixes and improvements
Bug fix for
hsurvmspline()
(hencehazard()
) which was using the wrong tail end ofpsurvmspline()
.Waning models no longer ignore nonproportional hazards and cure. Waning vignette merged into the methods vignette.
Leaveoneout cross validation now supports models with background hazards.
More consistent naming of arguments and variables (e.g.
hscale
for hazard scale, andhsd
for hazard smoothing parameter).More information given when printing fitted model objects.
Thorough documentation polishing and code tidying.