Which rare source levels reached the synthetic output

Censuses every categorical axis – strata and each non-numeric covariate – on both sides, and marks the levels too few source patients held to be safely copied out. That floor is min_pattern_share, the same rule the generator applies to visit sets and the scorecard applies to copied vectors, so no new threshold is introduced.

Usage

compare_pmx_rare_levels(source, synthetic, roles, floor = NULL)

Arguments

source

Source PMX data.

synthetic

Generated synthetic PMX data. When it carries a "pmx_settings" attribute, its min_pattern_share is used as the floor.

roles

Explicit roles from pmx_roles().

floor

Levels held by fewer than this many source patients are exposed. Left NULL it is taken from the run, and 2 otherwise – the lowest value that means "more than one real patient".

Value

A pmx_rare_levels data frame, one row per categorical column and level, with source_patients, synthetic_patients, exposed and reached. Zero rows when the roles declare no categorical axis.

Details

Read the exposed rows, and among them the ones that reached the output. A level held by two real patients, appearing in a released table, says that someone with that attribute was in this study; for a named trial with public inclusion criteria that can be close to identifying on its own, and no cohort size helps. The remedies are upstream of generation: drop the covariate from covariates, or collapse its rare levels before generating.

This reads real patient data on both sides and is marked "restricted_not_releasable".

What it cannot see. Rarity in the world. If every living carrier of a mutation is in this study, the source count is the whole population and looks unremarkable. And it censuses each column on its own: with d covariates there are 2^d combinations that could single a patient out, and enumerating them is not something this does.

See also

synpmx_scorecard(), which reports this as row B5, compare_pmx_distributions(), vignette("avatar-scorecard").

Examples

data <- pmx_simulated_fixture(20)
roles <- pmx_roles(
  id = "ID", time = "TIME", dv = "DV", amt = "AMT", evid = "EVID",
  cmt = "CMT", dvid = "DVID", covariates = c("WT", "SEX")
)
synthetic <- suppressWarnings(synpmx_avatar(data, roles, seed = 1))
#> synpmx_avatar(): dropped 8 undeclared column(s): NTIME, TAD, OCC, RATE, MDV, CENS, LIMIT, AGE.
#>   Declare a column in `keep` to carry it through verbatim.
compare_pmx_rare_levels(data, synthetic, roles)
#> PMX rare-level census (source against synthetic)
#> 
#> 0 level(s) held by fewer than 2 source patients; 0 of them reached the output.
#> 
#> Every level in the output is one that at least 2 source patients held.
#> 
#> Source-derived; not releasable.