Reporting Issues
Please report issues in the package, including its documentation and suggestions for enhancements in the projects github issues.
Extending SimNPH
You can extend the functions of this package, either for your own use or to contribute the enhancements to the package. The most common ways of extending the package will probably be to add your own data generating mechanisms or additional analysis functions for methods you want to evaluate. For both cases the function skeletons below are a good starting point.
Data Generating Models
Functions to generate datasets should be named beginning with generate_ and a descriptive name of the scenario they should simulate.
In order to work with the SimDesign package, the function has to take the arguments condition and fixed_objects. condition is a one row data.frame that contains all parameters for the scenario that is simulated. fixed_objects is a list in which variables, that are the same across all secenarios are passed.
Generate functions should come with an assumptions_... function that outputs code to generate a default dataset with all the parameters relevant for the data generation model, which can be used to construct a Design dataset. The assumptions_... function should also return the dataset created by the printed code invisibly. Implement this function in the same R file as the generate_... function and document it in the documentation of the generate_... function using the roxygen tag @describein.
Here’s boilerplate for a generate and assumptions functions, you can just copy it and fill in your own code:
#' Generate Dataset ...
#'
#' @param condition condition row of Design dataset
#' @param fixed_objects fixed objects of Design dataset
#'
#' @details
#' Condidtion has to contain the following columns:
#'
#' * X ...
#' * Y ...
#' * ...
#'
#' @return
#' For generate_x: A dataset with the columns t (time) and trt
#' (1=treatment, 0=control), evt (event, currently TRUE for all observations)
#'
#' @export
#' @describeIn generate_x simulates a dataset with ...
#'
#' @examples
generate_x <- function(condition, fixed_objects=NULL){
data.frame(t=numeric(0), trt=integer(0), evt=logical(0))
}
#' Create an empty assumtions data.frame for generate_x_...
#'
#' @param print print code to generate parameter set?
#'
#' @return For assumptions_delayed_effect: a design tibble with default values invisibly
#'
#' @details assumptions_x_... generates a default design `data.frame`
#' for use with generate_x_.... If print is `TRUE` code to produce
#' the template is also printed for copying, pasting and editing by the user.
#' (This is the default when run in an interactive session.)
#'
#' @export
#' @describeIn generate_x_... generate default design tibble
#'
#' @examples
#' Design <- assumptions_x_...()
#' Design
assumptions_x_... <- function(print=interactive()){
skel <- "expand.grid(
delay=m2d(seq(0, 10, by=2)), # delay of 0, 1, ..., 10 months
hazard_ctrl=m2r(24), # median survival control of 24 months
hazard_trt=m2r(36), # median survival treatment of 36 months
random_withdrawal=m2r(120) # median time to random withdrawal 10 years
)
"
if(print){
cat(skel)
}
invisible(
skel |>
str2expression() |>
eval()
)
}
#' Calculate true summary statistics for scenarios with delayed treatment effect
#'
#' @param Design Design data.frame for x
#' @param cutoff_stats Cutoff time for rmst and average hazard ratios
#' @param fixed_objects fixed objects not used for now
#'
#' @return For true_summary_statistics_x: the design data.frame
#' passed as argument with the additional columns:
#' * `rmst_trt` rmst in the treatment group
#' * `median_surv_trt` median survival in the treatment group
#' * `rmst_ctrl` rmst in the control group
#' * `median_surv_ctrl` median survial in the control group
#' * `gAHR` geometric average hazard ratio
#' * `AHR` average hazard ratio
#'
#' @export
#'
#' @describeIn generate_x calculate true summary statistics for ...
#'
#' @examples
true_summary_statistics_x <- function(Design, cutoff_stats=10, fixed_objects=NULL){
true_summary_statistics_x_rowwise <- function(condition, cutoff_stats){
res <- data.frame(
rmst_trt = NA_real_,
medial_surv_trt = NA_real_,
rmst_ctrl = NA_real_,
median_surv_ctrl = NA_real_,
gAHR = NA_real_,
AHR = NA_real_
)
res
}
Design <- Design |>
split(1:nrow(Design)) |>
mapply(FUN=true_summary_statistics_x_rowwise, cutoff_stats = cutoff_stats, SIMPLIFY = FALSE)
Design <- do.call(rbind, Design)
Design
}Analysis Functions
Analysis functions should take parameters regarding the analysis method (for example weights of a weighted logrank test) and return a function that can be used in SimDesigns runSimulation. To use a function as analysis function in runSimulation it has to take the arguments condition, dat, fixed_objects. Condition and fixed_objects are the same as in the generate_... functions. dat is the Dataset generated by the generate_... function. dat contains at least the columns t, trt, evt for time, treatment and event.
In order to work with the summarise functions in this package, analysis functions have to return a list. (Unfortunately vectors and data.frames are concatenated by SimDesign in a not very useful way, lists are passed on as is.)
If your function requires summaries that are more complex than the general functions for estimators and tests provide a summarise function in the same file as the analysis method.
#' Create Analyse Functions for ...
#'
#' @param X
#'
#' @return an analyse function that can be used in runSimulation
#' @export
#'
analyse_X <- function(X){
function(condition, dat, fixed_objects = NULL){
result_tmp <- list(
A = NA_real_,
N_pat=nrow(dat),
N_evt=sum(dat$evt)
)
result_tmp
}
}
#' Summarise Output from Analyse Functions for ...
#'
#' @param X
#'
#' @describeIn analyse_X Summarise Output from Analyse X
#'
#' @return
#' Returns a function with the arguments:
#' * condition
#' * results
#' * fixed objects
#'
#' that can be passed to create_summarise_function or to
#' SimDesign::runSimulation and that returns a `data.frame` with the columns
#' * `Y` ...
#' * ...
#'
#' @export
#'
#' @examples
summarise_x <- function(name=NULL){
res <- function(condition, results, fixed_objects=NULL){
data.frame(
"Y"=NA_real_
)
}
attr(res, "name") <- name
res
}Functions for Censoring and Experiment Designs
Different study setups are also implemented via functions that manipulate the simulated dataset. For example to apply administrative censoring after a fixed followup time, first a recruitment time is sampled from a uniform distribution using the function recruitment_uniform then the observations are censored according to their survival time, the recruitment time and the follwup time using the function admin_censoring_time. This can also be combined with random censoring with a fixed rate, …
These functions should take the simulated dataset as the first argument and further arguments at later positions so that they can be used in pipes.
#' Modify a simulated dataset by ...
#'
#' @param dat a simulated dataset
#' @param X ...
#' @param ...
#'
#' @return
#' Returns the dataset with ...
#' @export
#'
#' @examples
censoring_function_x <- function(dat, X){
dat$X <- X
dat
}Contributing Changes to the Package
Project Structure
The folder structure follows the structure of a typical R package. Documentation and the NAMESPACE file are generated with roxygen, so don’t edit those files manually. Unit tests using testthat are in tests/testthat with filenames corresponding to the filenames of the functions being tested.
This package uses the devtools, with, testthat and roxygen2 packages to facilitate package development, all of those packages are conveniently integrated within rstudio.
Coding Style
Please roughly keep to the tidyverse styleguide
When using functions from other packages (nph, cmprsk, …) refer to them by namespace (for example nph::rSurv_fun). In the DESCRIPTION file: don’t add the package under Depends, do add the package under Suggests or Imports.
Documentation and Tests
When adding functions provide documentation for all functions exported in the package namespace and make sure devtools::document() runs without errors or warnings.
Please also add tests that at least test if the functions can be called without error and produce datasets with the right columns (name and datatype). To add a tests file with the correct name in the correct place use usethis::use_test().