New features
- new
predsum() summarises binary prediction model performance (sensitivity, specificity, PPV, NPV, accuracy, AUC, AUC-PR, F1, Brier score) with Clopper-Pearson or bootstrap CIs; output presets “epi”, “data-science”, “all” or a custom set of statistics
- support for Firth-penalised logistic regression (
logistf) models in rm_mvsum(), with penalised likelihood ratio global p-values and profile penalised likelihood CIs
-
rm_uvsum() and rm_mvsum() now support clustered coxph models, reporting robust Wald tests
-
rm_survsum() and rm_cifsum() now accept factor or character status variables; rm_cifsum() works when there are no censored observations
-
rm_covsum() and rm_compactsum() Monte Carlo p-values and predsum() bootstrap CIs are now reproducible (B and seed arguments; RNG state is restored)
- VIF calculation now uses
type = "predictor" in the presence of interactions
Minor improvements and fixes
-
rm_mvsum(include_unadjusted = TRUE): fixed failures with coxph models, non-95% CIwidth, formulas stored in variables, dot formulas and transformed terms; continuous covariates and continuous:continuous interactions are no longer dropped from the unadjusted column; digits, CIwidth, whichp and showEvent are now passed to the univariate models
-
rm_uvsum(): excluded covariates (dates, the response, single-value variables) are no longer fit; reflevel now works with ordered factors; removeInf is now applied
-
rm_covsum() and rm_compactsum() now use expected rather than observed cell counts to choose the exact test, so the two functions agree; the grouping variable is dropped from covariates; non-numeric p-values (e.g. “excl”) are no longer adjusted
-
rm_covsum(): fixed error with date variables on Windows
-
ggkmcif2(): fixed curves being cut off when times was shorter than the data range; accepts US spelling of arguments; works with ggplot2 >= 3.4.0 (now the stated minimum)
-
psthr() displays “Not Estimable” only when all values are NA
- fixed display of special characters in column headers
- fixed unescaped variable names in
getVarLevels(), Global p-values being adjusted and double-formatted in bolded output, and a tidyselect deprecation warning for character column names
- consistent tidy evaluation (
enquo) for variable selection
- vignette: corrected
include_unadjusted section; new Prediction Metrics and Reproducibility sections
- expanded test suite
New features
- new more compact and flexible reporting function (rm_compactsum) that allows more control over each variable summarised and displays text summarising which tests and effect sizes were computed
- new function to apply variable labels to ggplots (replace_plot_labels)
- more comprehensive model summary function to allow plot extensions
- incorporated tidyselect into main summary functions
- support for tidycmprsk models
- ability to output to a scrolling table in HTML formats
Minor improvements and fixes
- fixes to special characters in HTML output
- removed ability to force Wald confidence intervals, confidence intervals now computed by the updated confint in base R
- rm_mvsum and rm_uvsum have been updated in the backend to enable easier extendibility to different model types
- bug fix in nestTable so variables order properly with repeat level names
- numerous bug fixes in formatting of tables, especially with duplicated variable and level names
- added automated testing of the proportional hazards assumption when reporting coxph models
- bug fix for computing global p-values in models with offset terms
- documentation updates
New features
- cumulative incidence can now be summarised by event type and time
- revised survival curves are now properly aligned and return ggplots
- variable labels are automatically displayed in summary tables
- forest plots now support poisson and negative binomial models
- options to globally set the number of digits for a report
Minor improvements and fixes
- updated excelCol to handle column ranges (ie A-Z, instead of A,B,C,…)
- updated mvsum to accept negative binomial models from MASS
- updated uvsum to fit negative binomial models
- differentiated univariate and multivariable estimates in forest plots with shapes
- added functionality to export tables in html and latex formats
- added functionality to summarise date variables
- previously the digit argument did not work for variables with more than one class, this has been fixed
- long data calls (as may be created with tidy statements) are now supported in summary functions
- table bolding for interaction terms is now consistent
- Total N, Number of Events now properly reported when interactions are present
- nestTable now works properly with two-column tables
- covTitle argument restored in rm_covsum
- Added functionality to include effect sizes in summary tables
- forestplot2 now works with geeglm objects
Bug fixes: - fixed bug where some p-values didn’t output after update to R 4.2.2 - special characters now better shown in tables - fixed bug so that NA are always shown as blank - fixed bug in nestTable so that variable indenting is consistent
- 5 October 2022 first CRAN release