landmaRk

landmaRk 0.1.2

New Features

• predict_longitudinal() now supports custom “summary measure” functions for method (in the same spirit as "locf"), computed directly from the raw longitudinal data without requiring a prior fit_longitudinal() call. Such functions must have arguments data, id, time, value and landmark.

Bug Fixes

• predict_longitudinal() no longer requires fit_longitudinal() to have been called when method is a summary measure (e.g. "locf" or a custom function); the check for an existing model fit is now only applied to fit-based prediction methods (e.g. "lcmm", "lme4"). `“locf”’ is implemented as a special case of summary measure

Other Changes

• Removed timeROC from Imports as it is no longer used.
• Resolved an R CMD check NOTE about a missing global binding for AUC in performance_metrics().

landmaRk 0.1.1

• Package is made publicly available via CRAN.

New Features

• AUC(t) metric for evaluating landmark models.
• Cross-validation support for model evaluation.
• Subject-specific predictions with LCMM.
• Cluster assignment from LCMM fits can be used in the survival step.
• summary() method for LandmarkAnalysis class.
• prune() method for LandmarkAnalysis class.
• Parallelised longitudinal model fitting.
• Survival plots via plot() method.
• Convergence messages for fitted models.
• Print number of subjects in each risk set for show() method.

Bug Fixes

• Fixed cross-validation not working correctly with LCMM.
• Fixed LOCF (last observation carried forward) issues.
• Fixed bug where static covariates were not used in Cox PH models.
• Fixed bug selecting IDs for survival analysis.
• Fixed bugs in working out risk sets.
• Fixed Windows parallelism issues.
• Better error handling for fit_survival() when predictions are not available.
• Character covariates are now converted to factors.

Other Changes

• Renamed package from landmarkR to landmaRk.
• Renamed main class to LandmarkAnalysis.
• Standardised horizon parameter handling.
• Now depends on release version of lcmm (>= 2.2.2).
• Baseline set to 0 for all survival models.
• Internal functions refactored into smaller functions.
• Code formatted using Air.
• Added CODE_OF_CONDUCT.md.

landmaRk 0.0.0.9000

• Package is made publicly available via r-universe and GitHub.

Features

• Support for lmer, and hlme is added for longitudinal sub-models.
• Support for last observation carried forward (LOCF).
• Support for cox proportional hazards survival sub-model.
• C-index and Brier score metrics are implemented for evaluating landmark models.
• Support for arbitrary longitudinal sub-models.