Fixed misalignment of group-sequential rejection boundaries in update_p_thresholds_cpp: thresholds are now placed at each hypothesis’s analyses_analysed columns (matching the observed p-value placement) instead of starting at analysis 1. Previously, any multi-look hypothesis whose first tested analysis was not analysis 1 (e.g. OS tested at analyses 2 & 3) had its interim boundary silently discarded and its interim tested against the final-look boundary, inflating simulated unconditional power (Tables 6a/6b) above the full-alpha local power and over-spending alpha. Design boundaries and local power (Tables 1/4/5) were unaffected.
appendMCP 0.3.0
New Features
Added lintr static code analysis setup and report (lintr-reports/)
Bug Fixes
Fixed alpha bug that didn’t propagate from config before
Fixed alpha bug with reporting local power
Fixed zero-power for propagated-weight hypotheses in update_p_thresholds_cpp
Fixed bug in enroll_rate processing for complex piecewise constant enrollment rates
Fixed differential dropout: updated get_dist_tite and plot_distribution_tte_ahr to use ahr_dd
Fixed process_analyses_2 to handle cases where hypothesis_analysed is a subset (e.g., c(2,3))
Fixed vignettes and templates after timeline_figure refactor