Disclosure-safe two-way cross-tabulation

Returns a small-cell-suppressed contingency table cross-tabulating two categorical columns of an OMOP CDM table. The server applies primary small-count suppression PLUS iterative complementary (secondary) suppression to a fixpoint and never returns exact margins, so suppressed cells cannot be recovered by row/column subtraction. Cells below the disclosure threshold render as NA; structural zeros render as 0. Concept-id axes are decorated with concept names.

When stratify_by is supplied, the result is a named list of independent protected 2-way slices (one per stratum level) rather than a single table; the unstratified total is never returned.

Usage

ds.omop.crosstab(
  table,
  row,
  col,
  by = "persons",
  row_concept_id = NULL,
  col_concept_id = NULL,
  cohort_table = NULL,
  stratify_by = NULL,
  cohort = NULL,
  scope = c("per_site", "pooled"),
  pooling_policy = "strict",
  symbol = "omop",
  conns = NULL,
  execute = TRUE
)

Arguments

table

Character; the CDM table name (e.g., "person").

row

Character; the column for table rows (e.g., "gender_concept_id").

col

Character; the column for table columns (e.g., "race_concept_id").

by

Character; "persons" (default, distinct person counts) or "records" (row counts).

row_concept_id

Integer/vector or NULL; optional concept ID(s) to restrict the row axis. Default: NULL.

col_concept_id

Integer/vector or NULL; optional concept ID(s) to restrict the column axis. Default: NULL.

cohort_table

Character or NULL; optional cohort table to scope the population (inner join on subject_id). Default: NULL.

stratify_by

Character or NULL; optional third categorical column to produce a named list of independent stratified 2-way slices. Default: NULL.

cohort

Cohort reference (a dsomop_cohort_handle, a cohort_definition_id, or a server-side cohort table name), or NULL. Takes precedence over cohort_table.

scope

Character; "per_site" (default) or "pooled".

pooling_policy

Character; "strict" (default) or "pooled_only_ok". Under strict, a cell absent or suppressed on any site is suppressed in the pooled table.

symbol

Character; the session symbol (default: "omop").

conns

DSI connection object(s) or NULL to use the session default.

execute

Logical; if FALSE, return a dry-run result containing only the generated call code (default: TRUE).

Value

A dsomop_result object with $per_site (named list of server cross-tab objects), $pooled (cell-wise summed table when pooled), and $meta.

Cross-tab is descriptive, not inferential

A cross-tab answers "how do these two variables co-occur?". It is NOT a substitute for a multivariable model. For genuine multivariable questions (three or more interacting variables, continuous adjustment, or estimating an association while controlling for confounders), route to ds.glm instead of building higher-dimensional cell tables, which are disclosure-unsafe and lose utility on small data.

Examples

if (FALSE) { # \dontrun{
ct <- ds.omop.crosstab("person", "gender_concept_id", "race_concept_id",
                       scope = "pooled")
ct$pooled$counts
} # }