Skip to contents

Reads the medical and product dictionaries from a CPRD Aurum code browser release (for example CPRD_CodeBrowser_YYYYMM_Aurum/, containing CPRDAurumMedical.txt and CPRDAurumProduct.txt) and returns lookup tables for medcodes and prodcodes, plus mapping tables to SNOMED CT and Read v2.

Usage

read_cprd_aurum(
  path,
  tables = cprd_aurum_tables(),
  version = NULL,
  source = "CPRD Aurum code browser"
)

Arguments

path

Path to the CPRD Aurum code browser release directory, containing CPRDAurumMedical.txt and CPRDAurumProduct.txt. Only the files needed for tables must be present.

tables

Character vector of tables to read. Any of "cprd_aurum_medcode_lkp", "cprd_aurum_prodcode_lkp", "cprd_aurum_medcode_snomed", "cprd_aurum_medcode_read2" and "cprd_aurum_prodcode_snomed". Defaults to all of them. The medcode tables come from CPRDAurumMedical.txt and the prodcode tables from CPRDAurumProduct.txt.

version

Character string for the version label. If NULL (default), the release date (yyyymm) is taken from a directory name like CPRD_CodeBrowser_YYYYMM_Aurum, with a message saying which version is used. If the name doesn't include a release date, an error asks for version to be supplied.

source

Character string for the data source.

Value

A named list with one element per requested table. Each holds lookup or mapping, a list with table (data.table) and metadata.

Details

CPRD supplies the code browser to researchers with access to its data, and it can't be downloaded publicly, so there is no get_*() function: supply the path to your local copy.

Columns

The lookups keep every column of the dictionary files. Column names are converted to lower case, with spaces replaced by underscores (MedCodeId becomes medcodeid, Term from EMIS becomes term_from_emis).

Column types are declared rather than guessed: observations and drugissues are integer counts, and every other column, including the numeric identifiers, is character. An error is raised if an expected column is missing. A column the reader doesn't know about is read as character, with a warning.

Column filters

The medcode lookup's emiscodecategoryid, and the prodcode lookup's routeofadministration and formulation, are column filters (see lookup_metadata()). Their values are taken from the release, and every value is included by default, so no codes are hidden by default. The EMIS code category IDs have no labels.

Blank values

Empty fields are read as NA, except in the column filter columns, which keep them as empty strings (""). Filters match exact values and drop NA, so the empty string is kept as a filter value in its own right.

Mappings

  • medcode to SNOMED CT (cprd_aurum_medcode_snomed): from medcodeid to snomedctconceptid, keeping snomedctdescriptionid.

  • medcode to Read v2 (cprd_aurum_medcode_read2): from medcodeid to read2_code, the first 5 characters of CleansedReadCode, keeping the 2-character read2_term_code. Rows without a cleansed Read code are dropped. OriginalReadCode is kept in the medcode lookup but not mapped.

  • prodcode to SNOMED CT (cprd_aurum_prodcode_snomed): from prodcodeid to dmdid. dm+d identifiers are SNOMED CT concepts, so this links prodcodes to the dm+d hierarchy in SNOMED CT. Rows without a dmdid are dropped.

Codes are not checked against the target code systems. A code that isn't in the target lookup raises a warning when mapped.

See also

Examples

if (FALSE) { # \dontrun{
result <- read_cprd_aurum("path/to/CPRD_CodeBrowser_YYYYMM_Aurum")
result$cprd_aurum_medcode_lkp$lookup$table
} # }