Skip to contents

Module 2 - Cohort Attrition Audit

Source: R/module_attrition.R
run_attrition.Rd

Profiles an instantiated cohort: size, index-date distribution, prior observation coverage, and demographics at index. Maps to HARPER section 5 and RECORD-PE item 6. All thresholds are arguments.

Usage

run_attrition(
  con,
  cdm_schema,
  cohort_table,
  cohort_id,
  obs_window = c(-365, 0),
  vocab_schema = "vocab",
  obs_warn_pct = 70,
  obs_fail_pct = 50,
  attrition_warn_threshold = 0.5
)

Arguments

con

A live DBI connection (see cdm_connect()).

cdm_schema

Schema holding the clinical CDM tables (e.g. "mimic_cdm").

cohort_table

Cohort table, schema-qualified (e.g. "results.rwevalidate_test_cohort"). Must have columns subject_id, cohort_definition_id, cohort_start_date, cohort_end_date.

cohort_id

Integer cohort definition id to profile.

obs_window

Length-2 numeric c(pre, post) in days relative to index. The required prior-observation window is abs(obs_window[1]) days (default 365).

vocab_schema

Schema holding the vocabulary tables (concept). In this split-schema environment the clinical schema's concept table is empty, so the gender label is resolved from vocab_schema. Default "vocab".

obs_warn_pct, obs_fail_pct

Prior-observation coverage thresholds (percent of cohort with sufficient prior observation). Below obs_warn_pct flags WARN, below obs_fail_pct flags FAIL. Defaults 70 and 50.

attrition_warn_threshold

Reserved for stepwise inclusion-criteria attrition (single-criterion drop fraction). Not exercised in v0.1 because criteria lists are not yet a supported input. Default 0.5.

Value

A named list:

cohort_size

integer - distinct subjects in the cohort

index_date_summary

data.frame of index_year, n

obs_coverage

data.frame: cohort_n, n_sufficient, pct_sufficient, min_prior_days. cohort_n counts every cohort subject. A subject whose index date falls outside any observation period counts as insufficient prior observation rather than being dropped, so cohort_n matches cohort_size.

demographics

data.frame: sex, mean_age, min_age, max_age, n

flags

character vector of WARN:/FAIL: messages (empty if all pass)

Details

This profiles the cohort as it was handed in. It reports the cohort-level counts that item 6 asks for, but it does not yet rebuild a step-by-step attrition flowchart (how many subjects each inclusion or exclusion criterion removed), because the cohort table does not carry the criteria that built it. Per-criterion attrition is planned for a later version.

Examples

if (requireNamespace("duckdb", quietly = TRUE)) {
  con <- example_cdm()
  res <- run_attrition(con, cdm_schema = "main", cohort_table = "cohort",
                       cohort_id = 1, vocab_schema = "main")
  res$cohort_size
  res$demographics
  cdm_disconnect(con)
}