Extract Machine Learning Feature Matrix from OMOP CDM

Builds vectorized patient feature matrices anchored to cohort index date T0 (cohort_start_date). Computes baseline demographics and prior observation lookback window event counts (e.g. 30d, 180d, 365d), directly connecting OMOP DuckDB cohorts to predictive modeling workflows.

Usage

extract_patient_features(
  con,
  cohort_id = NULL,
  outcome_cohort_id = NULL,
  lookback_days = c(30, 180, 365),
  include_demographics = TRUE,
  include_conditions = TRUE,
  include_drugs = TRUE,
  include_procedures = TRUE,
  include_measurements = TRUE,
  format = "df",
  offset_anchor = c("visit_start_date", "visit_end_date", "index_date"),
  anchor = NULL,
  stream_to_parquet = NULL,
  partition_by = NULL,
  cohort_table = "cohort",
  measurement_concept_ids = NULL,
  lookahead_days = 0,
  ...
)

extract_features(
  con,
  cohort_id = NULL,
  outcome_cohort_id = NULL,
  lookback_days = c(30, 180, 365),
  include_demographics = TRUE,
  include_conditions = TRUE,
  include_drugs = TRUE,
  include_procedures = TRUE,
  include_measurements = TRUE,
  format = "df",
  offset_anchor = c("visit_start_date", "visit_end_date", "index_date"),
  anchor = NULL,
  stream_to_parquet = NULL,
  partition_by = NULL,
  cohort_table = "cohort",
  measurement_concept_ids = NULL,
  lookahead_days = 0,
  ...
)

Arguments

con

Active DuckDB DBI connection.

cohort_id

Integer cohort definition ID defining the target population.

outcome_cohort_id

Optional integer cohort definition ID defining binary outcome Y.

lookback_days

Integer vector of days for prior observation windows (default c(30, 180, 365)).

include_demographics

Logical. Include age at index, gender, race, ethnicity, and index year.

include_conditions

Logical. Include condition occurrence counts across lookback windows.

include_drugs

Logical. Include drug exposure counts across lookback windows.

include_procedures

Logical. Include procedure occurrence counts across lookback windows.

include_measurements

Logical. Include measurement counts across lookback windows.

format

Output format: "df" (data.frame), "arrow", or "sparse".

offset_anchor

Milestone date to anchor lookback windows ('visit_start_date' | 'visit_end_date' | 'index_date').

anchor

Alias for offset_anchor ('start' | 'end' | 'index').

stream_to_parquet

Optional file path to stream feature matrix directly to Parquet out-of-core.

partition_by

Optional column(s) to partition streamed Parquet files by (e.g. 'site').

cohort_table

Table name containing cohort entries (default 'cohort').

measurement_concept_ids

Specific measurement concept IDs to extract aggregated features for.

lookahead_days

Days after anchor date to include in measurement extraction (default 0).

...

Additional arguments.

Value

A data.frame, Arrow Table, or sparse covariate data.frame.