This repository holds the simulation and analysis code for the research project "The Case–Crossover Design Under Changing Baseline Outcome Risk: A Simulation of Ambient Temperature and Preterm Birth," published in Epidemiology: https://doi.org/10.1097/ede.0000000000001477 by Carrión et. al, 2022.
...
Make sure you have R version 4.x installed, then download or clone this repository to a directory on your computer.
R package dependencies are managed by renv. If you don't have the renv package installed, it will be automatically installed when you open the RStudio project or an R session in this project directory.
Run renv::restore(), and it will install the same package versions to your project directory that we used. This will not impact the versions of R packages you use outside of this project.
The publication based on this project used 1000 simulated datasets per simulated relative risk. This is controlled by the value of repeats in _targets.R.
We have set this to 10 so that users can run a quick simulation by default, but you may change this back to 1000 or any number. However, memory (RAM) usage scales with the values of repeats (see below).
To run the simulation, start an R or RStudio session in the directory where you've downloaded this repository. In the R console, run:
library(targets)
tar_make()
And it will begin the workflow provided in the _targets.R file. You can cancel the run at any time, and any completed targets will be skipped the next time you run tar_make().
When the workflow is finished, it will render a summary report to code/report.html that you can open in your web browser.
Alternatively, you can run the workflow much faster by running multiple workers at once. Instead of tar_make(), run:
tar_make_future(workers = 4L)
replacing the number 4 with however many simultaneous workers you would like to run, up to the number of CPU cores on your system. You must also have enough system RAM to support all of the workers running at once.
Megabytes of RAM used per worker increases with repeats. Systems with 8 GB of RAM will likely be able to run the workflow with the default setting of 10 repeats.
Use these rough estimates of memory usage per worker to decide how many simultaneous workers your system can support based on your selection for repeats:
| repeats | estimated RAM per worker |
|---|---|
| 10 | 300 MB |
| 40 | 800 MB |
| 100 | 2500 MB |
| 1000 | 3600 MB |
The last section of the _targets.R file lists output tables and figures you may want to view after you have completed running the simulation. Each of these are called a "target," and you can display them with the tar_read function, or load them into your R environment with tar_load.
For example:
tar_load(table_coverage_2018) will load table_coverage_2018 into your environment.
tar_read(vis_2018) will display the coverage and bias plots for the 2018 simulations.
These tables and plots have also been pulled into the code/report.html generated at the end of the workflow.
The workflow has been made reproducible using the targets package by Will Landau and contributors. Random seeds are set, so you should receive the same results if you run this workflow multiple times while providing the same parameters.
The Targets package gives many ways to examine your run of the workflow. For a useful flowchart showing status and execution time, try: tar_visnetwork(targets_only = TRUE, label = 'time').