changelog.Rmdestimate_ipwhr. This implements the model described in
Hernán et al. (2000) Marginal Structural Models to Estimate the
Causal Effect of Zidovudine on the Survival of HIV-Positive Men for
a point treatment.identify_interval
object documentation and in the manual about using time-varying
covariates.identify_censoring all have values between 0
and 1, inclusive..Start column name has been added as a reserved
name for the input data.make_table1:
make_table1 that resulted in rows
that had entries with 0 weights sometimes being dropped.make_table1 support for SMDs when there are more
than two treatment groups. make_table1 now provides an SMD
column for each pairwise combination of treatment groups.compute_cumcount_effect_ci now throws an error if an
invalid boot_method input was provided.make_table2, plot, and
forest_plot methods now explicitly check that cumulative
count effect measures aren’t specified for cumulative risk methods and
vice versa, and throw an error in such a case.compute_cumrisk_effect_ci and
compute_cumcount_effect_ci:
effect_measure_type and
boot_method are each stringseffect_measure_type and boot_method when the
input results object performed bootstrapping.smd function now checks that the stratifying
variable has exactly two levels in it.compute_cumcount_effect_ci,
compute_cumrisk_effect_ci, make_table2,
named_list, plot_boot_cumrisk,
table2_helper_cumcount, table2_helper_cumrisk
to include details about the various available effect measures.plot and moved
the content from the plot.cumcount,
plot.cumrisk, and plot.hr in the entry.effect_measure_select entry.actg,
example4, leukemia, and wihs
datasets.Plotting counterfactual cumulative risk and effect measures
and
Table 2s for counterfactual cumulative risk and effect measures.estimate_ipwhr correctly handles function inputs with
scopes outside of the lexical scope of the function definition.estimate_ipwrisk, estimate_gcomprisk,
estimate_aipwrisk,
estimate_aipwrisk_efficient,
estimate_ipwcount, and estimate_ipwhr return
objects now include an element called data rather than
called dataq containing a value of the input data rather
than a quosure.Added a new function estimate_ipwhr which enables
estimation of counterfactual hazard ratios with inverse probability of
treatment weighting and inverse probability of missingness weighting.
Objects produced by this function have class hr.
Added a new vignette called Hazard ratios describing the
theory and usage of estimate_ipwhr.
Provided a new is.hr function to check whether an
object inherits from the hr class.
Added make_table2, forest_plot and
plot methods for the hr class.
Provided a new hr_data function, which takes results
of estimate_ipwhr and returns a data frame with information
backing the make_table2.hr and plot.hr
methods.
Added a new inspect_ipw_weights function that
produces the per-subject weights for ipw objects.
Exposed the boot_method formal argument to the
existing make_table2, plot, and
forest_plot methods.
estimate_ipwrisk now produces per-subject weights
(with the intention that they will be accessed by the user via
inspect_ipw_weights).
compute_cumrisk_effect_ci throws an error if an
input of "log-normal" for boot_method is used
with any input for effect_measure_type other than
"RR" or "CR" since otherwise it can result in
the logarithm function being evaluated for a non-positive
values.
compute_cumrisk_effect_ci throws an error if any
input for boot_method is provided other than
"normal" or "log-normal" rather than returning
a data frame without variance information.
Fixed a bug in compute_cumcount_effect_ci that would
result in an unintended error when using a "log-normal"
input for boot_method.
make_table2.cumrisk now throws an error if any input
for boot_method is provided other than
"normal" or "log-normal".
knitr and forcats are each removed from the list of
Imports. knitr and rmarkdown are each added to the list of
Suggests. The package lower bound for gt was changed from
v0.5 to v0.6.
Added testing for all new routines.
Added testing of estimate_ipwhr per-subject
weights.
Added regression tests for estimate_ipwrisk to
ensure that newer versions of the function produce the same cumulative
risk results and variances as for v0.39.06.
Added testing of boot_method input options for
make_table2.cumrisk, make_table2.cumcount,
plot.cumrisk, plot.cumcount, and
forest_plot.cumrisk.
Added an assertion that subject IDs are unique unless
time-varying covariates specified. That is, across all estimators, prior
to computing the estimates there is a check that there are no duplicated
subject IDs. Duplicated subject IDs in the source data are only allowed
if identify_interval is used to indicate time-varying
data.
make_table1 rewritten to utilize the gt
package, thus allowing far more customization and exporting
ability.compare_fits and allow
user to choose alpha level for confidence intervals.estimate_aipwrisk when
fully_efficient = TRUE, by rewriting the internal function
integrator and helpers.compare_fits function, which allows for the
comparison of multiple fit objectsstart>stop time warnings when
using time-varying covariates and times argument is
provided.estimate_aipwrisk when competing risks are present. Bug was
introduced in version 0.36.17.NaN risk estimates when no
censoring model is specified, there are tied event times, the last
observed event/censoring times are tied, and everyone either has the
event or censoring.forest_plot.cumrisk so that
when no risk time is passed, it uses the maximum time in each input
object.make_table1 that
causes SMDs lower than 0.01 to be displayed as “<0.01”.plot.cumrisk and
plot.cumcount to allow any number of treatment levels.scale_pt_display argument in
make_table2.calc_rate_ci argument in make_table2.make_table2 that prevented selecting
correct treatment reference level when multiple objects were
passed.right_join fuction in dplyr.Added overlap weights as an option.
Added discussion of different weighting methods to manual.
Bug fix. SMR weights were incorrect with more than 2 levels of treatment.
Changed censoring behavior. Now, censoring by end of follow-up is
not treated as a censoring event to be modelled by the censoring models.
To replicate analyses conducted with older versions of the package,
specify tau_censor = TRUE when calling the estimation
function.
Bug fix. When subgroup_ps is FALSE, the wrong n was being returned by the estimating functions.
Bug fix. When user supplied weights or a subgroup, efficient AIPW threw an error.
Bug fix. When missing data are present, the output prediction vector was the wrong length for AIPW.
Bug fix. make_table1 was inconsistent when
identify_subject() was used.
Bug fix. When using R versions greater than 4, error was thrown when using efficient AIPW with null models.
Fix warning related to RNG in future package.