NEWS

NBvarsel 0.1.1 (2026-08-30)

Improved package documentation and roxygen comments for the main selection function and plotting helpers, making the exported API clearer and easier to use.
Refined the documentation for all_subset_plot() and VIF_plot() to better describe their inputs, outputs, and examples.
Extended plotting utilities with optional predictor relabelling via data_dict, making it easier to present clinically meaningful variable names in figures.
Improved the tutorial vignette layout and narrative around the fabricated clinical example and case-study sections, with clearer descriptions of model comparisons and interpretation.
Cleaned up the vignette workflow to avoid brittle global-environment cleanup during execution.

Initial release.
Added adnex_results dataset: pre-computed exhaustive variable selection results from the IOTA/ADNEX ovarian tumour case study (top 20 models per predictor count from a 65,535-model search). Original patient data are not disclosed.
Added a tutorial vignette (nb-varsel-tutorial) demonstrating the full workflow with fabricated clinical data, comparison with backward elimination and LASSO, and a real-data case study using the shipped ADNEX results.
nb_varsel() performs exhaustive or groupwise variable selection for binary outcome models using cross-validated Net Benefit, with support for predictor costs (per-variable or grouped), restricted cubic splines, interaction terms, permutation importance, and parallel computation.
all_subset_plot() creates a two-panel figure showing model performance and predictor inclusion across all evaluated models. Supports customisable colors via highlight_color and tile_color arguments.
VIF_plot() creates a horizontal bar chart of average permutation importance (delta Net Benefit) per predictor. Supports a customisable color argument.