NEWS

rifttable 0.7.3 (2026-10-31)

New functionality:
- The timepoint for time-to-event analyses (e.g., "cuminc", "cumincdiff") can be directly provided as a variable in the design. The previous way of providing it via arguments = list(timepoint = 365.25) remains functional as a second option.
- The spelling strata instead of stratum in the design is also accepted as a fallback.
Bug fixes:
- table1_design() omits an empty row for categorical variables that are completely NA (thanks @tsurudak for the bug report).
- survdiff_ci() now reports two-sided p-values (thanks @eribul for the bug report #12).
- Confidence limits for ratios and differences of cumulative incidence (type = "cumincdiff" and "cumincratio") in settings without competing events have been fixed. Previously, upper and lower limits were swapped in intermediary steps when using the default MOVER method, leading to shifted intervals (thanks @tgerke for the bug report #13 and pull request).
- All data are now kept as a tibble internally to address edge cases of failing input checks on data provided as a data.frame.
- The exposure variable name or label can be now be identical to a category of this variable, and empty labels are better supported.

Bug fixes:
- survdiff_ci() now uses the conf.level argument correctly in confidence interval estimation using the MOVER approach (#8).
Internal:
- Anticipate changes in {dplyr} 1.2.0 in label handling in unit tests (thanks to @DavisVaughan, #10).

Breaking changes:
- Require base R pipe |> and thus R >= 4.1.
- Make the id variable identifying clustered observations within the same individual a global rifttable() option for the entire data set, not only for specific estimators.
New functionality:
- Expand input checks to missing values in time/event variables, to missing effect modifiers for joint models and their levels, to nonexistent custom estimators, and to empty input data sets.
- Add type = "sum" estimator.
Internal and bug fixes:
- Cover entire package by unit tests.
- Return exposure consistently as a character.
- Let the design accept weight in addition to weights.
- Require {risks} >= 0.4.3.
- Use modern tidyselect, anonymous functions, and code style.

New functionality:
- Provide easier interface to code competing events by directly providing event = "event_variable@Event_Type_One" in the design. If multiple event types are present, estimate cumulative incidence and differences/ratios of cumulative incidence in a competing-event setting.
- Allow for clustered observations in survival data, e.g., multiple rows per person.
- Estimate ratios of survival and cumulative incidence, i.e., x-year risk ratios. Use MOVER estimation for confidence intervals of both ratios and differences in survival and cumulative incidence by default.
- Support weights, e.g., inverse-probability weights, directly via a weights argument in the design. Weighted estimates are currently available for many survival estimators: type = "cuminc", "surv", their differences and ratios (e.g., "cumincdiff"), and "hr". This is a breaking change for Cox models (type = "hr"), where providing weights in the arguments list now generates an error.
Bug fixes:
- Allow for @ in factor levels for a table1_design().
- rt_gt(): Output knitr-formatted tables for GitHub-flavored markdown also in Quarto .qmd, similar to .Rmd.
- Better handling of edge cases, e.g., ratios of 0, when rounding estimates.
Expanded documentation
- Restructure site.
- Separate documentation of estimators by outcome type.
- New FAQs on confidence levels, reference levels, custom functions, and joint models.

New functionality:
- Add overall argument exposure_levels to let user control handling of missing exposure levels (NA) or factors with empty levels as the exposure.
- type = "geomean" for geometric means.
Documentation: Expand FAQs.
Internal and bug fixes:
- Consider exposure or trend of "" as missing, and stratum = "" as no subsetting by the effect_modifier, instead of subsetting to effect modifier being an empty string. Input check that a stratum must be provided for joint models and strata are not empty.
- Consider missing type as "blank".
- Do not add empty rows/columns if type2 has empty results for some cells or if only a trend variable and no exposure is given.
- Rounding works even if result vector contains strings (e.g., no estimate).
- More safeguards for all-NA outcome variables. More input checks.
- Do not warn about non-0/1 outcomes in log-linear models for ratios of continuous variables.
- Add initial set of unit tests.

New functionality:
- Cox models (type = "hr") allow for weights, clustering, and robust standard errors.
- Argument ratio_digits_decrease: By default, decrease number of decimal digits shown for ratios by 1 digit for ratios > 3 and by 2 digits for ratios
  1. Leads to rounded ratios and confidence intervals of 1.23, 3.4, and 11.
- rt_gt() now indents the first column and applies markdown formatting to it by default.
New FAQ vignette.
Bug fixes:
- Binary outcomes returned NA instead of 0 in unstratified tables with all-null outcome.
- type = "maxfu" ignored digits and diff_digits.
- Allow for different exposure (strata labels) and arguments in one table.
- Show unstratified estimate if exposure is "", not just for NA.
Internal:
- rt_gt(): suppress random id of gt tables to keep git diff slim.
- Keep variables .event, .outcome, etc. available under their original names.
- Require {risks} >= 0.4.2.
- Examples load the breastcancer dataset from the risks package.

Breaking changes:
- to is set to ", " by default, instead of "-" for ratio variables and " to " for difference variables
- Custom functions are now directly called via the type variable, following a restructuring of all estimation functions with greater flexibility.
- design$type no longer accepts additional arguments, such as time points. Supply list instead via design$arguments.
- Suppression of strata with sparse or re-identifiable data with design$nmin now differentiates between counts of total observations or outcomes, depending on estimator.
New function table1_design(): Generate design of a descriptive "Table 1."
New outcome option "variable@level" for categorical variables that displays level as a binary outcome. Used by table1_design().
Support unstratified tables displaying the trend/linear slope (trend variable in the design) without an exposure.
More customization:
- rifttable(reference = ...): Label for the reference category.
- design$ci: Width of confidence intervals.
- design$na_rm: Omitting observations with missing outcome data.
- design$arguments: Flexibly passing along any argument to estimation functions.
New vignette describing all estimators.
Internal:
- Drop dependency on R >= 4.1 and native pipe.
- Require {risks} >= 0.4.0.
- Remove dependency on {labelled} package.
- The {gt} and {quantreg} packages are now optional as soft dependencies.
- Compatible with {dplyr} 1.1.0, {tidyselect} 1.2.0

khsmisc::table2() "graduated" into its own package. See {khsmisc} Changelog for earlier versions.
Add breastcancer() dataset
Use R >= 4.1 native pipe, |>
Remove RMTL estimators