TODOs

[bblodfon]

• [ ] Import code and measures from mlr3proba
• [ ] Discuss design
  • Prediction types? Inputs?
  • Additional arguments?
  • AUC(t) scores should return for multiple time points
• [ ] Include time-dependent C-index from riskRegression (see https://github.com/bcjaeger/aorsf-bench/issues/5)
• [ ] Implement time-dependent Antolini's C-index and proper C-index (paper)
  • Antolini's R code taken from pycox
• [ ] Maybe reimplement survAUC measures (low priority as these are only for Cox model and linear predictions lp risk scores)

Similar software (maybe for a small benchmark later):

• discSurv R package - lots of functions for compet.risk models as well
• riskRegression::Score() R package
  • Successor of pec
  • Score() function has Blanchet’s time-dependent AUC(t) and IBS with IPCW with different censoring models
• SurvivalROC R package => time-dependent AUC (CRAN doc from 2013)
• timeROC R package => time-dependent AUC (from paper in 2013, RiskRegression::Score is more updated, speed-up version with some bugs corrected)
• survcomp R package, github (a list of functions for performance assessment [paper]) - VERY OLD
• survAUC R package (used in mlr3proba, includes a lot, buggy)
• survMetrics R package (IBS, C-index, simulate datasets)
• survival::concordance() => lots of C-indexes supported

[adibender]

Re design, should keep different types of censoring + multiple events in minde