daoh 0.2.6

Changes (default units)

DAOH is now reported in days by default across the package, for a single consistent and interpretable unit. bland_altman_daoh(), plot_daoh_dist(), daoh_reclassify(), daoh_reclassify_centile(), and daoh_icc() now default to use_pc = FALSE (days); pass use_pc = TRUE for percentage of the period.
daoh_summary() gains a use_pc argument (default FALSE, days).
Behaviour change: code relying on the previous percentage defaults must now pass use_pc = TRUE. For daoh_icc() and daoh_reclassify() the result is unchanged (the statistic is invariant to the per-period rescaling); for bland_altman_daoh(), plot_daoh_dist(), and daoh_summary() the reported values switch from percentage to days.

plot_daoh_ba() gains a show_loa argument. When TRUE, the mean-bias and 95% limits-of-agreement values are printed directly on the plot beside their reference lines (in addition to the caption). Defaults to FALSE, so existing plots are unchanged.

plot_daoh_dist() now labels the y-axis "Count" (the absolute number of observations per bin) rather than the ambiguous "Frequency", and gains a ylab argument to override it.

plot_daoh_dist() gains a use_pc argument: set use_pc = FALSE to plot DAOH in days rather than as a percentage of the period (axis label follows).
bland_altman_daoh() now records units ("%" or "days") in its result, and plot_daoh_ba() labels its axes and caption from that unit, so a Bland-Altman computed with use_pc = FALSE is plotted in days. Results from earlier versions (without units) are still treated as percentages.

bland_altman_daoh(), daoh_reclassify(), daoh_reclassify_centile(), and daoh_icc() now join their inputs with data.table instead of base merge(). On multi-million-row inputs each call is roughly an order of magnitude faster; results are unchanged (row order within the join may differ, which the computed statistics do not depend on).

Timezone handling for POSIXct inputs. Two related defects affected the exact method (and any POSIXct input) on machines whose local timezone is not UTC:
- as.Date() on POSIXct converts via UTC, so local times earlier than the UTC offset (e.g. mornings in New Zealand) were assigned to the previous calendar day. This shifted index dates and admission/discharge dates.
- hospital_time() measured exact intervals from the origin at 00:00 UTC, offsetting every value by the local UTC offset (about half a day in NZ) relative to the whole-day numbers used for the index-date period windows, which misaligned all interval clipping at period boundaries.
Both now use the input’s local calendar date (plus local clock-time fraction for exact). Results from Date inputs are unchanged. Results for POSIXct inputs in non-UTC timezones will change — they were previously wrong.

Corrected two test expectations: calc_daoh() returns one row per index date (with daoh = period) when there are no admissions, and the nights-vs-days systematic difference equals the days episode count (same-day stays contribute zero nights, so the nights method does not count them as episodes).
Added a regression test for local-timezone handling of POSIXct inputs.

calc_daoh() now uses a fully vectorised data.table pipeline instead of an R-level loop over patients. The key changes:
- foverlaps() matches admissions to index periods in a single C-level pass, correctly handling patients with multiple index dates.
- Interval merging uses a one-pass cumulative-max sweep (ave()) rather than a sequential R loop per patient, giving O(n log n) overall complexity.
- Dead time and DAOH are computed by vectorised arithmetic over all patients simultaneously.
- All column assignments use data.table::set() (no :=), which avoids data.table’s cedta() namespace check and works correctly under both devtools::load_all() and a standard package install.
data.table (>= 1.14.0) added to Imports.

Fixed a latent issue where min(dod) on an empty group (all patients alive) could emit a spurious “no non-missing arguments” warning.

Initial CRAN release.
calc_daoh(): calculates DAOH for one or more patients given a data.frame of admissions and a data.frame of index dates. Supports three hospital-time algorithms (nights, days, exact) and three death-handling methods (midday, midnight, zero). A configurable gap tolerance (default 12 h) merges near-adjacent admissions before summing hospital time.
hospital_time(): converts admission/discharge date pairs to numeric intervals under the chosen algorithm.
dead_time(): calculates dead days within a follow-up period.
merge_intervals(): merges overlapping or near-adjacent intervals with a configurable gap tolerance. Exported for direct use.
daoh_summary(): summary statistics (mean, SD, percentiles) across a list of calc_daoh() results.
bland_altman_daoh() / plot_daoh_ba(): Bland–Altman agreement analysis and plot comparing two sets of DAOH values.
daoh_icc(): intraclass correlation coefficient between two DAOH vectors.
daoh_reclassify() / daoh_reclassify_centile() / plot_daoh_reclassify(): centile-based reclassification analysis and plot.
plot_daoh_dist(): distribution plot for a single set of DAOH values.
load_example(): loads the bundled example dataset.