Build 30-Day Readmission Cohort — build_readmission

Constructs a leak-free 30-day readmission cohort directly in DuckDB SQL. Enforces adult inpatient stay, LOS >= 1 day, discharged alive, verified follow-up window (filtering right-censored lost-to-follow-up stays), reproducible index admission sampling, and disjoint lookback vs post-discharge outcome scoping.

Usage

build_readmission_cohort(
  con,
  cohort_id = 1,
  outcome_cohort_id = NULL,
  cohort_name = "Inpatient 30-Day Readmission Cohort",
  cohort_description = NULL,
  target_visit_concept_ids = 9201,
  outcome_visit_concept_ids = 9201,
  followup_window_days = 30,
  grace_days = 0,
  washin_days = 365,
  index_selection_rule = "random",
  random_state = 42,
  require_verified_followup = TRUE,
  table_name = NULL
)

Arguments

con: Active DuckDB DBI connection.
cohort_id: Integer ID for target index cohort in OMOP cohort table (default 1).
outcome_cohort_id: Optional integer ID for materialized readmission outcome cohort.
cohort_name: Display name for cohort_definition table.
cohort_description: Description for cohort_definition table.
target_visit_concept_ids: Integer vector of visit concepts defining index stay (default 9201).
outcome_visit_concept_ids: Integer vector of visit concepts defining readmission (default 9201).
followup_window_days: Follow-up window in days (default 30).
grace_days: Grace days immediately post-discharge before outcome window begins (default 0).
washin_days: Prior observation lookback requirement in days (default 365).
index_selection_rule: "random" (reproducible seed), "first", or "last".
random_state: Integer seed for reproducible random index sampling (default 42).
require_verified_followup: Logical. Whether to filter out lost-to-follow-up patients (default TRUE).
table_name: Optional character string to materialize the index cohort table with metadata.

Value

A data.frame containing qualifying index cohort records with metadata.

Details

This is a thin wrapper over the shared study-cohort engine (see define_study_cohort(), which generalises it): the legacy conventions (death from the death table and the index stay's own discharge disposition 4216643; follow-up evidenced by later visits, measurements, conditions or drug exposures; age as calendar-year difference) are pinned here and covered by golden-file tests.