Direct access supplementary material II: https://survival-lumc.github.io/CauseSpecCovarMI/
In studies analyzing competing time-to-event outcomes, interest often lies in both estimating the effects of baseline covariates on the cause-specific hazards, and predicting cumulative incidence functions. When missing values occur in these baseline covariates, they may be discarded as part of a complete case analysis (CCA) or multiply imputed. In the latter case, the imputations may be performed either compatibly with a substantive model pre-specified as a cause-specific Cox model (SMC-FCS), or approximately so (MICE). In a large simulation study, we assessed the performance of these three different methods in terms of estimating cause-specific regression coefficients and predicting cumulative incidence functions. Concerning regression coefficients, results provide further support for use of SMC-FCS over MICE, particularly when covariate effects are large and the baseline hazards of the competing events are substantially different. CCA also shows adequate performance in settings where missingness is not outcome-dependent. With regard to cumulative incidence prediction, SMC-FCS and MICE showed more similar performance, as also evidenced in the illustrative analysis of competing outcomes following a hematopoietic stem cell transplantation. The findings are discussed alongside recommendations for practising statisticians.
This project is set-up as an R-package compendium, with the following structure:
.
├── R # User-made functions
├── analysis # Scripts for illustrative analysis
│ ├── figures # Manuscript figures
│ └── simulations # Scripts to run simulation study
├── data # Full and processed simulation data/synthetic MDS data
│ └── sim-reps_indiv # Individual simulation replications
│ ├── preds
│ └── regr
├── data-raw # Scripts for processing raw data
├── man # Documentation user-made functions
└── vignettes # Supplementary manuscript resultsThe vignettes directory contains the supplementary material from the manuscript, and is rendered here (add link) using pkgdown.
To reproduce results from the manuscript, first install the package and its dependencies using
remotes::install_github("survival-lumc/CauseSpecCovarMI", dependencies = TRUE)You can then clone the directory using git clone https://github.com/survival-lumc/CauseSpecCovarMI.git , and set your RStudio working directory to the home of this repository (you can make use the included R project file).
The simulation study was run on a computer cluster using a SLURM scheduler. If you have access to such a scheduler, you should install the package and clone the directory on the cluster first, and then run (from the home directory)
sbatch analysis/simulations/all-simulations.sh The above will run each replication of each scenario, and for each replication will save two .rds files (one for regression coefficients, another for the predictions) containing the results to the data/sim-reps_indiv/. To summarise all of these, you can run
sbatch analysis/simulations/summarise-simulations.shThis will produce four files, which represent the full and summarised results for both the regression coefficient results and those for the predictions:
data/sims_regr_full.fstdata/sims_regr_summary.fstdata/sims_preds_full.fstdata/sims_preds_summary.fstThe summarised data is available directly in the package and is accessed using CauseSpecCovarMI::regr_results and CauseSpecCovarMI::preds_results. Due to the size of the full result datafiles, there are available upon request from the corresponding author (such that one does not need to run the entire simulation study again, which would take in the order of days).
Should you want to replicate one replication of one scenario locally (for example the 10th replication of scenario 3), you can run
Rscript analysis/simulations/run-simulation.R 3 10The scenarios (and their numbers/parameters) are contained in data/scenarios.rds.
The figures can be reproduced locally by running
Rscript analysis/simulations/manuscript-figures-simulations.RThe original dataset on which the illustrative analysis was performed could not be shared, however a synthetic version is provided in the package as CauseSpecCovarMI::da. This will yield similar results after the imputation procedure, but not comparable for complete-case analysis. On a cluster using SLURM, you can run
sbatch analysis/run-illustrative-analysis.shThis will make use of 10 cores to run the 100 imputations (with 25 iterations) from both mice() and smcfcs() in parallel.
It can also be run locally by first changing the number of cores in analysis/illustrative-analysis.R , and then
Rscript analysis/illustrative-analysis.R