Skip to contents

Add prescription drug data to skeleton

Source: R/add_rx.R
add_rx.Rd

Searches for specific drug codes (ATC or product names) in Swedish prescription registry data and adds corresponding boolean variables to the skeleton based on prescription periods and duration of treatment.

Usage

add_rx(
  skeleton,
  lmed,
  id_name = "lopnr",
  codes = list(rx_hormones_pubblock = c("L02AE", "H01CA")),
  source = "atc",
  rxs = NULL
)

Arguments

skeleton

A data.table containing the main skeleton structure created by create_skeleton

lmed

A data.table containing prescription registry data (LMED). Must have columns for person ID, prescription date (edatum), treatment duration (fddd), and drug codes (atc) or product names (produkt)

id_name

Character string specifying the name of the ID variable (default: "lopnr")

codes

Named list of drug code patterns. Names become column names in the skeleton; values are character vectors. Matching semantics depend on source (see below).

Prefixing a pattern with "!" turns it into a row-level veto: any prescription whose code matches the (un-prefixed) pattern is masked and does not contribute to the named output column. Final rule: a prescription row contributes to the named column iff at least one un-prefixed pattern matches AND no "!" pattern matches.

Behaviour notes worth knowing:

  • Vetoes are independent per named code. A "!" entry inside one list element does not leak into any other element of the same codes list. Two named codes can produce two completely different views of the same prescription rows.

  • Veto match style follows source. For source = "atc" the veto is prefix-based via startsWith(): "!C10AA" masks C10AA01, C10AA02, ... For source = "produkt" the veto is exact-match via %chin%: "!Sertralin" does NOT mask "Sertralin Sandoz" because product names are exact, not prefixes.

  • All-negative pattern set produces an empty column. c("!C10AA") on its own gives an all-FALSE result – without any positive pattern there is no set to carve from. Use a wider include + the negative, e.g. c("C10A", "!C10AA").

  • Per-(id, isoyearweek) aggregation respects the veto on a per-source-row basis. The veto removes specific prescription rows from the matched set before the per-week aggregation runs. If a person has both a vetoed Rx and a non-vetoed Rx whose coverage windows overlap in the same skeleton week, the non-vetoed Rx still drives that week to TRUE – the veto only kills its own row's contribution, not the whole week.

Examples:

  • c("N06A") – any antidepressant.

  • c("C10A", "!C10AA", "!C10AB") – any lipid-modifying agent except statins and fibrates.

Default includes hormone therapy codes for puberty blockers (L02AE, H01CA). Common patterns include:

  • Antidepressants: "N06A"

  • Hormone therapy: "G03", "L02AE", "H01CA"

  • Cardiovascular drugs: "C07", "C08", "C09"

source

Character string specifying search field and matching strategy:

  • "atc" (default) - Prefix matching in ATC codes (e.g., "N06A" matches "N06AB06"). Uses startsWith() for fast C-level matching.

  • "produkt" - Exact matching on product names (e.g., "Delestrogen" matches only "Delestrogen", not "Delestrogen Extra"). Uses %chin% for fast lookup.

rxs

Deprecated. Use codes instead.

Value

The skeleton data.table is modified by reference with prescription variables added. Variables are TRUE during periods when the prescription is active based on start/stop dates calculated from prescription date + treatment duration

The function also returns the skeleton, invisibly. The return is the object the caller passed, while that object has a free column slot for every new column. Past that point data.table cannot grow the column list in place, so the new columns land on the returned table alone. A caller that passes an expression rather than a variable MUST use the return value.

Any other name that pointed at the same table before the call is stale afterwards. R cannot grow a list in place. A growth therefore leaves the caller's binding on a NEW object, and every other name on the old one. Take an alias after the call, never before it.

Coverage interval

Each endpoint of the coverage interval is resolved once, independently:

  • start: supplied start_isoyearweek, else the ISO week of the supplied start_date, else the ISO week of edatum.

  • stop: supplied stop_isoyearweek, else the ISO week of the supplied stop_date, else the ISO week of edatum + round(fddd) - 1.

The - 1 is because foverlaps(type = "any") matches inclusively at both endpoints: without it a duration of N days would cover N + 1 days.

Rows whose round(fddd) is missing, non-finite or not positive are dropped, with one warning naming the count. This applies if and only if the stop endpoint is actually resolved from fddd, that is when the caller supplied neither stop_isoyearweek nor stop_date.

Interval validation

The resolved pair is then validated, on one rule that every combination of supplied columns reaches. A row is dropped, with one warning naming the count, when any of the following holds:

  • either endpoint is missing;

  • either endpoint is not a well-formed ISO week ("YYYY-WW" with week 01 to 53, or the annual "YYYY-**");

  • the start week is later than the stop week;

  • both endpoints came from dates and the start date is later than the stop date. This catches an interval inverted by days but contained in one ISO week, which compares equal as week strings.

An endpoint that is well formed but outside the skeleton's weeks is kept, not dropped: "2020-53" is a real week that a skeleton ending in "2020-52" does not carry, and the interval still covers every week before it.

Rows before the weekly spine

create_skeleton builds an annual spine ("<year>-**", is_isoyear == TRUE) for every ISO year before the weekly period. After validation, any endpoint that falls before the first weekly row is remapped onto the annual row of its ISO year, the same rule add_diagnoses uses. A prescription that starts before the weekly period and ends inside it therefore sets both the annual rows of the pre-weekly portion and the weekly rows it covers.

The remap applies to every endpoint, including one the caller supplied. A supplied start_isoyearweek of "2019-51", on a skeleton whose weekly period starts in 2020, marks the 2019 annual row.

The prescription table is not modified

add_rx() computes start_date, stop_date, start_isoyearweek and stop_isoyearweek on a local working copy. lmed is read, never written. Earlier versions wrote these helper columns back into lmed by reference; because the ISO week columns depend on the skeleton, reusing one lmed across two skeletons then silently reused the first skeleton's values.

See also

create_skeleton for creating the skeleton structure, add_diagnoses for diagnosis codes, add_operations for surgical procedures, make_lowercase_names for data preprocessing

Other data_integration: add_annual(), add_cancer_without_morphology(), add_cods(), add_diagnoses(), add_onetime(), add_operations(), add_quality_registry()

Examples

# Load fake data
data("fake_person_ids", package = "swereg")
data("fake_prescriptions", package = "swereg")
swereg::make_lowercase_names(fake_prescriptions, date_columns = "edatum")

# Create skeleton
skeleton <- create_skeleton(fake_person_ids[1:10], 2020, "2020-53")

# Add prescription data
rx_patterns <- list(
  "antidepressants" = c("N06A"),
  "hormones" = c("G03", "L02AE")
)
add_rx(skeleton, fake_prescriptions, "p444_lopnr_personnr", rx_patterns, "atc")