- Save analysis ready now uses
foreign::write.dta()so that output can be read directly into Stata.
- Add an observation warning that returns a message if the number of observations provided to the model differs from the number of observations used by the model.
- The r image has been reverted to use the old version of the readr package. Hence, the
pathargument is now used again in the calls toreadr::write_csv(). - Add a cron job to the GitHub Action workflow which runs the cox-ipw GitHub Actions test workflow once per week.
- Ran
opensafely codelists updateto update the codelists in this repo.
- Due to the upgrading of the readr package (from version 1.3.1 to 2.1.4) in the r image on 22/09/2023, update calls to
readr::write_csv()to use thefileargument instead of the deprecatedpathargument. - In the GitHub Actions workflows update the actions/checkout action from the v3 to the v4 sliding tag as per the README
- Refine the workflows in the test action
- Shorten analysis ready output file name further
- Shorten analysis ready output file name
- Update follow-up end criteria to include outcome
- Save the anaylysis ready dataset in csv.gz format instead of rds
- Specify analysis ready data set as output in YAML
- Add option to not run analysis (i.e., just save analysis ready dataset and stop)
- Fix bug that occurred with episode labels if only one post-exposure time point was specified
- Various improvements to the code to fix linting warnings
- Fix weighting bug for models
- Tidy naming in ipw sample script
- Add origin for data converted to date format
- Further fix for the sampling bug for cases where the number of controls to be sampled is greater than the number available
- Collapse region from 9 into 3 regions when covariate thresholds are not met
- Fix sampling bug where exposed controls were being incorrectly handled
- Fix person time bug where Cox weights were not being appropriately applied
- Update
mathieudutour/github-tag-actionGitHub Action to v6.1 tag - Set cox weights to one to calculate person-time in each episode if cox weights not otherwise provided
- Stop code removing strata variables and add warning for potentially low counts
- Fix typo that caused fatal error for fn-check_covariates.R
- In GitHub Actions workflows bump
actions/checkoutGitHub Action to v3 sliding tag
- Add option to sample exposed individuals. If not sampled, all exposed individuals are included regardless of case/control status.
- Print output to log at each stage of
fn-check_covariates.Rfor better debugging - For structured variable names, impose class (e.g., make variables containing '_date' into dates)
- Print output to log at each stage for better debugging
- Add
cox-ipwversion number to output - Use Breslow method for ties in Cox model
- Add option to save analysis ready dataset
- Print
survival::tmerge()inputs to log - Make
outcome_time_medianinclude time from previous episodes
- Fix robust SEs so they are only applied when sampling is on
- Clarify ln(hr) versus hr
- Add code to calculate median time to outcome for each time period
- Add support for '.rds' as an input file
- Remove protected covariate feature due to lack of generalizability
- Fixed outcomes per episode calculation
- Added median person-time per episode to output
- Added CHANGELOG.md
- Added reporting of protected covariates that stop model running
- Updated defaults
- Updated how args are recorded
- Added error capturing so jobs always complete
- Increased population size so model runs on test data
- Tidied formatting
- Updated test action
- Removed automatic cut points
- Improved documentation