DARTER - Project 708421

Project-specific guide to the BS & Dementia cohort study

Published

July 2, 2026

DARTER - Diabetes And inteRgenerational Transmission of hEalth determinants over the life couRse (project 708421).

This section is only for those working on the DARTER project. The content here builds on the general DST guide and adds the project-specific material.

Tip

Searchable variable and register overview for DARTER All variables and registers applied for in the project are collected in a searchable table: steno-aarhus.github.io/darter-project →

New to the project? Start with the general guide and return here: → Phase 1 - Plan your study → Phase 2 - R: the bare essentials → Phase 3 - Log in to DST

In this section

Page	Contents
This page	Setup (`fastreg` + `duckplyr`) and a reusable LPR extraction function
Register paths and datastores	Confirmed paths and access methods for all registers on 708421
DARTER-specific pitfalls	Quirks specific to this project

Initial setup for DARTER

Register data on DARTER is loaded with fastreg::read_register("name") - exactly as in the general guide. You just point fastreg at DARTER’s parquet folder once per script; see Loading templates.

Important

Install the latest duckplyr at the start of each session

DST has duckplyr pre-installed, but in an old version that is missing functionality - several of the patterns in this guide will not work with it. So get the latest version from CRAN, and repeat after every log-out / server reset (the reset reverts the package to the old, pre-installed version). duckplyr provides the DuckDB engine behind compute() (see Large registers below).

install.packages("duckplyr")   # get the latest CRAN version - run before library()
packageVersion("duckplyr")     # check the version in your session; >= 1.1 is fine

Before running a script: verify that path_output at the top of each script points to your workspace folder (defined under Base paths in Register paths and datastores).

Recommendation: create a helper function for LPR extractions

LPR extractions require combining LPR2 somatic, LPR2 psychiatric and LPR3 - and doing the same for each new outcome in the project. It pays off to encapsulate this in one reusable function rather than copying the code repeatedly.

Advantages: - One place to fix if something changes (e.g. a new register or a new column) - The code block for each outcome is reduced from ~40 lines to one function call - Errors are introduced in one place instead of in each copy

How to create the function - define it at the top of your script or in a separate functions.R file:

See the full get_lpr_diagnoses() function

library(fastreg)
library(dplyr)

# icd_codes: 3-character ICD codes WITHOUT the D-prefix, e.g. c("F03", "G30"). REQUIRED (see warning).
# diagtypes: diagnosis TYPE - "A"=action, "B"=secondary, "G"=underlying/grundmorbus (LPR2 only). Default c("A", "B").
get_lpr_diagnoses <- function(pnr_vector, icd_codes, diagtypes = c("A", "B")) {
  #-------------------------------------------------------------
  # Open registers (LPR2 somatic + psychiatric + LPR3)
  #-------------------------------------------------------------
  lpr_adm   <- read_register("lpr_adm")   %>% rename_with(tolower)   # LPR2 somatic contacts
  lpr_diag  <- read_register("lpr_diag")  %>% rename_with(tolower)   # LPR2 somatic diagnoses
  psyk_adm  <- read_register("t_psyk_adm")  %>% rename_with(tolower) %>%
    rename(pnr = v_cpr, recnum = k_recnum)                            # LPR2 psychiatric contacts
  psyk_diag <- read_register("t_psyk_diag") %>% rename_with(tolower) %>%
    rename(recnum = v_recnum)                                          # LPR2 psychiatric diagnoses
  lpr3_k    <- read_register("lpr_a_kontakt")  %>% rename_with(tolower) %>%
    filter(lprindberetningssystem == "LPR3")                               # CRITICAL: keep only rows from the LPR3 system - avoid overlapping rows
  lpr3_d    <- read_register("lpr_a_diagnose") %>% rename_with(tolower)  # LPR3 diagnoses

  #-------------------------------------------------------------
  # LPR2 somatic
  #-------------------------------------------------------------
  lpr2_dx <- lpr_adm %>%
    semi_join(tibble(pnr = pnr_vector), by = "pnr") %>%   # only the cohort - the pattern is explained in Phase 9 (Hospital contacts)
    select(pnr, recnum, date_contact = d_inddto) %>%
    inner_join(
      lpr_diag %>% filter(c_diagtype %in% !!diagtypes) %>% select(recnum, c_diag, c_diagtype),
      by = "recnum"
    ) %>%
    filter(substr(c_diag, 2, 4) %in% !!icd_codes) %>%        # filter on codes BEFORE collect() - only relevant rows are pulled into R
    collect() %>%
    mutate(icd3 = substr(c_diag, 2, 4))                       # "DF03" -> "F03": strip the D-prefix

  #-------------------------------------------------------------
  # LPR2 psychiatric
  #-------------------------------------------------------------
  lpr2_psyk_dx <- psyk_adm %>%
    semi_join(tibble(pnr = pnr_vector), by = "pnr") %>%   # only the cohort - see Phase 9
    select(pnr, recnum, date_contact = d_inddto) %>%
    inner_join(
      psyk_diag %>% filter(c_diagtype %in% !!diagtypes) %>% select(recnum, c_diag, c_diagtype),
      by = "recnum"
    ) %>%
    filter(substr(c_diag, 2, 4) %in% !!icd_codes) %>%
    collect() %>%
    mutate(icd3 = substr(c_diag, 2, 4))

  #-------------------------------------------------------------
  # LPR3
  #-------------------------------------------------------------
  lpr3_dx <- lpr3_k %>%
    semi_join(tibble(pnr = pnr_vector), by = "pnr") %>%   # only the cohort - see Phase 9
    select(pnr, dw_ek_kontakt, date_contact = kont_starttidspunkt) %>%
    inner_join(
      lpr3_d %>%
        filter(diag_kode_type %in% !!diagtypes,               # NB: "G" (grundmorbus) exists only in LPR2, not LPR3
               is.na(senere_afkraeftet) | senere_afkraeftet != "Ja") %>%
        select(dw_ek_kontakt, c_diag = diag_kode, c_diagtype = diag_kode_type),
      by = "dw_ek_kontakt"
    ) %>%
    filter(substr(c_diag, 2, 4) %in% !!icd_codes) %>%
    collect() %>%
    mutate(date_contact = as.Date(date_contact),               # datetime → date
           icd3 = substr(c_diag, 2, 4))

  bind_rows(lpr2_dx, lpr2_psyk_dx, lpr3_dx)                   # return combined table
}

Note

semi_join or filter(... %in% ...)? Both select rows, but each has its use:

semi_join(tibble(pnr = pnr_vector), by = "pnr"): for your cohort. A large R vector of pnrs pushes down into Arrow/DuckDB more reliably as a small table than a filter(pnr %in% ...), which can be slow or rejected outright (especially with older duckplyr).
filter(substr(c_diag, 2, 4) %in% !!icd_codes): for a short code list. Here filter is fine - but remember !!, which injects the local R vector into the query (without !! DuckDB looks for a column of that name). The background is in Extract from LPR.

Use the function - one call per extraction, only change icd_codes

kohort     <- readRDS("datasets/full_cohort.rds")
pnr_list   <- unique(kohort$pnr)

# Specify the ICD codes you want (3 chars, no D). The function filters BEFORE data is pulled into R.
dementia_dx <- get_lpr_diagnoses(
  pnr_vector = pnr_list,
  icd_codes  = c("F00", "F01", "F02", "F03", "G30", "G31"),  # dementia (G30/G31 are ICD G-codes)
  diagtypes  = c("A", "B")         # "A"=action, "B"=secondary, "G"=underlying/grundmorbus (LPR2 only). Extend e.g. to c("A","B","G")
)
# Returns one row per diagnosis: pnr | date_contact | c_diag | icd3 | c_diagtype (A/B/G)
# c_diagtype enables sensitivity analyses, e.g. action diagnoses only: filter(c_diagtype == "A")
# Multiple outcomes? Pass the union of all codes here and split by icd3 in R afterwards - one register scan only.

# First occurrence after index date per person
dementia <- dementia_dx %>%
  inner_join(kohort %>% select(pnr, index_date), by = "pnr") %>%
  filter(date_contact > index_date) %>%
  group_by(pnr) %>% arrange(date_contact) %>% slice(1) %>% ungroup() %>%
  select(pnr, dementia_date = date_contact)

result <- kohort %>% select(pnr) %>% left_join(dementia, by = "pnr")
saveRDS(result, "datasets/extract_dementia.rds")

Warning

Always specify icd_codes - never fetch all diagnoses. The function filters on your codes before collect(), so only the relevant rows are pulled into R’s RAM. That is why icd_codes is a required argument.

Do not remove the code filter to “grab everything”: every A/B/G diagnosis for a whole cohort across LPR2 and LPR3 can be millions of rows. It fills R’s memory, and such a heavy extraction can get you kicked off the DST server.

Note

This is the DARTER variant (using read_register() and the confirmed register names for 708421, as of June 2026). The general open_dataset() version and the explanation behind the pattern are in Extract from LPR.

Large registers - `compute()` vs `collect()`

get_lpr_diagnoses() ends each extraction with collect() (pulls into R). On a very large register (e.g. LMDB or laboratory results) you can pipe to compute() instead, which materialises the result in DuckDB without filling R’s RAM - always reduce with semi_join()/filter()/select() before compute(). The technique and the difference between compute() and collect() are explained generally in Phase 5 - Extracting data step by step.

In this section

Initial setup for DARTER

Recommendation: create a helper function for LPR extractions

Large registers - compute() vs collect()

See also

Large registers - `compute()` vs `collect()`