Skip to contents

Dual-strategy medication-class classifier

Source: R/medication_classes.R
extract_medication_classes.Rd

Classifies each cohort member's medication exposure two independent ways, so sites with very different drug coding (legacy CUI / NDC source strings vs standardized RxNorm) can be compared:

Usage

extract_medication_classes(
  db_con,
  cohort_table,
  medication_specs,
  strategy = c("both", "source", "standard"),
  crosswalk = NULL,
  cohort_definition_id = NULL,
  person_col = NULL,
  index_date_col = NULL,
  site_col = NULL,
  lookback_days = NULL,
  lookahead_days = 0L,
  window = NULL,
  end_date_col = NULL,
  on_unresolved = c("raise", "warn", "ignore"),
  exclude_forms = NULL,
  exclude_dose_form_concept_ids = NULL,
  apply_exclude_to_source = FALSE,
  exclude_forms_source = FALSE,
  disaggregate_combinations = FALSE,
  is_combination = FALSE,
  vocab_schema = NULL,
  vocab_db = NULL,
  route_concept_ids = NULL,
  by_type = FALSE,
  offset_anchor = c("visit_start_date", "visit_end_date", "index_date"),
  anchor = NULL,
  drug_type_concept_ids = NULL,
  return_audit = FALSE,
  ...
)

Arguments

db_con

DuckDB DBI connection with drug_exposure (and concept_ancestor / concept for the standard strategy).

cohort_table

Table / view with a person column (subject_id / person_id) and an index date (cohort_start_date / index_date).

medication_specs

Named list class_name -> spec. spec is a list with any of ancestor_concept_ids (ATC class or RxNorm ingredient concept ids), atc_codes, rxnorm_ingredient_ids, source_values (inline strategy-A strings); or a bare vector, taken as ancestor_concept_ids.

strategy

"source", "standard" or "both".

crosswalk

Strategy-A mapping of source strings to classes (data frame, CSV path or table name with columns like source_value / drug_class). Combined with inline source_values.

cohort_definition_id

Restrict a stacked OMOP cohort table to one definition.

person_col, index_date_col, site_col

Override column auto-detection. A site column (site_anon / site_id / site) carries through for per-site auditing.

lookback_days, lookahead_days

Exposure window around the index date. NULL lookback = unbounded.

window

Optional named temporal frame overriding the lookback/lookahead days: "pre_admission" = [index - lookback_days (default 365), index - 1]; "inpatient" = [index, stay end]; "discharge" = [stay end, stay end] (orders written at discharge).

end_date_col

Stay end column for "inpatient" / "discharge" (default cohort_end_date / visit_end_date).

on_unresolved

Action when concepts or ATC codes fail to resolve ("raise", "warn", or "ignore").

exclude_forms

Character vector of dosage/form patterns to exclude (e.g. "topical", "ophthalmic").

exclude_dose_form_concept_ids

Optional concept IDs of dose forms to exclude via 'RxNorm has dose form'.

apply_exclude_to_source

Also apply form/dose-form exclusions to strategy A source values.

exclude_forms_source

Alias for apply_exclude_to_source.

disaggregate_combinations

If TRUE, adds <class>_plain and <class>_combination columns.

is_combination

Alias for disaggregate_combinations.

vocab_schema

Optional schema name containing vocabulary tables.

vocab_db

Optional attached database containing vocabulary tables.

route_concept_ids

Vector of standard OMOP route concept IDs to filter exposures by.

by_type

Logical; if TRUE, also extracts flags separated by source concept type.

offset_anchor

Reference anchor date to evaluate offsets against ("visit_start_date", "visit_end_date", "index_date").

anchor

Alias for offset_anchor.

drug_type_concept_ids

Optional vector of concept IDs to restrict drug exposure type concepts.

return_audit

Also return audit_medication_strategies() output (needs strategy = "both").

...

Forwarded to audit_medication_strategies().

Value

A data frame with person_id, index_date, optional site, and per class <class>_source, <class>_standard (0/1, per the chosen strategy) and <class> (OR of the available flags). With return_audit = TRUE, a list list(flags =, audit =).

Details

  • Strategy A, source / crosswalk ("source"): a user-supplied crosswalk (CSV path, DuckDB table, or data frame) maps source strings to drug classes. Matched against drug_source_value and the concept_code of drug_source_concept_id.

  • Strategy B, OMOP standard hierarchy ("standard"): concept_ancestor traversal from ATC class concepts (levels 3/4/5, by concept id or ATC code) or RxNorm ingredients down to drug_concept_id.

  • strategy = "both" runs both; the combined flag is their OR.

Mirror of omop_etl.medication_classes.extract_medication_classes in Python.