Overview of registers

Variable names for the most commonly used DST registers

A starting point, not an authority. This page can be out of date or simply wrong, and your own delivery may differ from it. Verify against your files with colnames() (see Phase 7 - Inspect your data) before you rely on anything here. Found a mistake? Open an issue or use the feedback box at the bottom - corrections are genuinely welcome.

Where this comes from, and why it can still be wrong

The tables on this page are generated from a schema built out of DST’s published documentation: the variable list for each register and DST’s order list, which names every variable in every DST register with its period. The health registers - LPR, LMDB, the death registers, the cancer register - are documented by Sundhedsdatastyrelsen on esundhed.dk instead, which is also the only source that publishes data types.

Each register links to its own source, and every value set says underneath where its codes come from.

Those sources change. Registers are extended, columns are added and retired, and a code can change meaning without changing its name. Names here are shown after rename_with(tolower), and your own delivery may spell them differently or not contain them at all: what DST documents is the register, not what your project ordered.

Key columns - the ones you actually use in the code - are marked in bold. Names apply after rename_with(tolower). See Pitfalls for the quirks of each register.

How to read the code examples: read_register() vs open_dataset()

The code on this page opens registers with read_register("registername") (fastreg, by name - recommended). That requires fastreg set up with the path to your registers, see Parquet and fastreg if you did not convert them from SAS yourself.

Without fastreg, use open_dataset("path/to/register/") instead and replace the path with your project’s parquet folder. Everything else in the examples works the same either way.

Looking for a particular variable? Find a variable searches every column in every register at once - the way to see how the same thing is named and coded differently from one register to the next.

Where to look things up

This page shows the columns most studies use. When you need more than that, the question you are asking decides where to go:

Your question	Where to look
What are all the columns in this register?	The register’s own variable list - follow the register name in the table below
Does this register reach the end of my follow-up?	Datasafari - shows the latest reference date and the next expected update
Which registers exist at all, and which are closed?	Oversigt over registre - over a thousand, with years covered. Luk means closed and never extended, as dodsaars and dodsaasg both are
Can I even order this variable, and for which years?	Bestillingsliste.xlsx - every variable in every DST register, with its period. The one to open while you are still planning
What does code 330 mean?	Klassifikationer - DISCED-15 for education, SOCIO for employment, DISCO for occupation
Why does this variable jump in one particular year?	Højkvalitetsdokumentation - breaks in the series and validity periods, for selected variables
What does DST mean by “beskæftiget”?	Hvad betyder - the words, not the columns. Look here when a variable measures something narrower than the everyday word suggests

All of it hangs off Dokumentation af data →, DST’s own hub, if you would rather start there. The health registers are documented separately by Sundhedsdatastyrelsen on esundhed.dk.

And whichever you use: what DST holds is not what your project was delivered. Confirm against your own files with colnames().

Overview - all registers

Register	Read as	Join key	Period	Often used
AKM	"akm"	pnr	1976 to 2024	socio13, socio02, socio
BEF	"bef"	pnr	1985-12 to 2026-06	koen, foed_dag, familie_id
CANCER	"cancer"	k_cprnr	1943 to 2024	c_icd10, c_morfo03, c_topo3
DOD	"dod"	pnr	1970 to 2025	doddato
DODSAARS	"dodsaars"	pnr	1970 to 2001	d_dodsdto, c_dodsmaade, c_dod1
DODSAARSAGER	"dodsaarsager"	pnr	2022 to 2024	doedsdato, doedsaarsag_tilgrundliggende, doedsaarsag_kode_1
DODSAASG	"dodsaasg"	pnr	2002 to 2022	d_dodsdato, c_dodtilgrundl_acme, c_dod_1a
FAIK	"faik"	familie_id	1987 to 2024	pnr, famaekvivadisp_13, year
LAB_DM_FORSKER	"lab_dm_forsker"	patient_cpr	2008 to 2025	samplingdate, analysiscode, value
LMDB	"lmdb"	pnr	1995-12 to 2025-12	eksd, atc, atc1
LPR_A_DIAGNOSE	"lpr_a_diagnose"	dw_ek_kontakt	2019 to 2025	diag_kode, diag_kode_type, senere_afkraeftet
LPR_A_KONTAKT	"lpr_a_kontakt"	dw_ek_kontakt	2017 to 2025	dw_ek_forloeb, pnr, kont_starttidspunkt
LPR_A_PROCREGISTRERING	"lpr_a_procregistrering"	dw_ek_kontakt	2019 to 2025	proc_kode, proc_starttidspunkt, proc_kode_type
LPR_ADM	"lpr_adm"	recnum	1977 to 2019-03	pnr, d_inddto, d_uddto
LPR_DIAG	"lpr_diag"	recnum	1977 to 2019-03	c_diag, c_diagtype, c_tildiag
LPR_SKSOPR	"lpr_sksopr"	recnum	1996 to 2019	c_opr, c_oprart, c_osgh
LPR_SKSUBE	"lpr_sksube"	recnum	1999 to 2019	c_opr, d_odto, year
MFR	"mfr"	cpr_barn	1997 to 2018	alder_moder, bmi_moder, cpr_moder
SSSY	"sssy"	pnr	2005 to 2025	ydernr, speciale, ydlant
SYSI	"sysi"	pnr	1990 to 2005	ydernr, speciale, ydlant
T_PSYK_ADM	"t_psyk_adm"	recnum	1995 to 2019	pnr, c_pattype, c_adiag
T_PSYK_DIAG	"t_psyk_diag"	recnum	1995 to 2019	c_diag, c_diagtype, c_tildiag
UDDA	"udda"	pnr	1980-12 to 2025-09	hfaudd, udd, hf_vfra
VNDS	"vnds"	pnr	1973 to 2024	indud_kode, haend_dato, indud_land
VNDS_HIST	"vnds_hist"	pnr	1973 to 2004	indud_kode, haend_dato, indud_land
VNDS_IND	"vnds_ind"	pnr	2005 to 2025	haend_dato, indv_land, indvmd
VNDS_UD	"vnds_ud"	pnr	2005 to 2025	haend_dato, udv_land, udvmd

The Often used column is the handful of columns most studies reach for, not the full list: follow the register link for every column DST documents, or the section below for the ones this guide explains. lab_dm_forsker is in the schema but not in this table, because it is reached through Sundhedsdatastyrelsen’s Forskerservice rather than DST.

1. Demographics and deaths

BEF - Population Register

Status register - one snapshot per person per reference time point. Delivered quarterly since 2008 (March, June, September, December); before 2008 December only. Whether year == 2020 corresponds to a particular reference time point depends on the project convention - confirm in your project guide. A person who dies during 2020 still appears in the 2020 snapshot - use DOD to determine whether a person was alive on a specific date.

BEF cannot on its own answer whether a person was resident in Denmark on a given date: between two snapshots people can both leave and come back. To decide residence on an index date, combine BEF with VNDS - see Phase 10 - the source population.

Column	Type	Role	Label	Years
pnr	character	join key	Personal identifier	1985 to 2026
koen	numeric	code	Sex	1985 to 2026
foed_dag	date	date	Date of birth	1985 to 2026
familie_id	character	join key	Household key	1985 to 2026
reg	character	code	Region	1985 to 2026
civst	character	code	Marital status	1985 to 2026
kom	character	code	Municipality code	1985 to 2026
year	integer	date	Register year
alder	numeric	value	Age at the reference time point	1985 to 2026
opr_land	numeric	code	Country of origin	1985 to 2026
referencetid	date	date	Reference time point	1985 to 2026

All other columns (30)

Column	Type	Role	Label	Years
mor_id	character	identifier	Mother’s person id	1985 to 2026
far_id	character	identifier	Father’s person id	1985 to 2026
aegte_id	character	identifier	Spouse id	1985 to 2026
e_faelle_id	character	identifier	Cohabiting partner id	1985 to 2026
fdato	date	date	Date of birth, CPR form
antboernf	numeric	value	Number of children in the family	1985 to 2026
antboernh	numeric	value	Number of children in the household	1985 to 2026
antpersf	numeric	value	Number of people in the family	1985 to 2026
antpersh	numeric	value	Number of people in the household	1985 to 2026
antefam	numeric	value	Number of E-families in the household	1985 to 2026
familie_type	numeric	code	Family type	1985 to 2026
fam_koen	numeric	code	Sex of the family’s reference person	1985 to 2026
plads	numeric	code	Position in the family	1985 to 2026
hustype	numeric	code	Household type	1985 to 2026
fm_mark	numeric	code	Parent marker	1985 to 2026
civ_vfra	date	date	Date the marital status took effect	1985 to 2026
bop_vfra	date	date	Date of moving in or immigrating	1985 to 2026
ie_type	numeric	code	Immigrant, descendant or Danish origin	1985 to 2026
foedreg_kode	numeric	code	Place of birth registration	1985 to 2026
statsb	numeric	code	Citizenship	1985 to 2026
opholdmd_dk	numeric	value	Months of residence in Denmark	1985 to 2026
van_vtil	date	date	Immigration date	1985-12 to 2003-12
foerste_indvandring	date	date	First immigration date	2004-12 to 2026-06
seneste_indvandring	date	date	Most recent immigration date	2004-12 to 2026-06
adresse_id	character	identifier	Address id	1985 to 2026
fkirk	character	code	Membership of the Danish National Church	2004-12 to 2026-06
cprtjek	character	code	CPR check	2004-12 to 2026-06
cprtype	character	code	CPR type	2004-12 to 2026-06
version	numeric	code	Module data version	2004-12 to 2026-06
betalingskom	character	code	Betalingskommune	1985 to 2026

• mor_id: A pnr-like identifier for the mother, so BEF can be turned into a family structure without a separate register. It is only filled where the link is registered, which is not the case for everyone born before CPR.
• fdato: Not on DST’s variable list for BEF, which documents foed_dag instead. Present in this delivery. Prefer foed_dag unless you have checked what yours contains.
• antefam: A household can hold several families. That is why the family counts and the household counts differ, and why FAIK’s household income cannot be read as one family’s income without checking this.
• van_vtil: Ends December 2003 and is replaced by foerste_indvandring and seneste_indvandring. A study spanning 2003 has to read both, or it silently loses immigration dates on one side of the break.
• foerste_indvandring: Begins December 2004. Before that the information is in van_vtil.
• adresse_id: Identifies a dwelling, so two people with the same value live at the same address. It is not a geographic coordinate and cannot be decoded into one.

No published source gives a data type for 1 of these 41 columns, so the Type column is our own assumption. Check with sapply(class) on a row of your own data before relying on it, especially for code columns, which lose their leading zeros if they arrive as numbers.

Join key: pnr.

Joins to other registers:

• familie_id joins to FAIK (many-to-one).

Value sets for the coded columns (9)

Code system	Values
koen	1 Mand, 2 Kvinde, 9 Uoplyst
reg	0 Uoplyst, 81 Nordjylland, 82 Midtjylland, 83 Syddanmark, 84 Hovedstaden, 85 Sjælland
civst	U Ugift, G Gift (+ separeret), F Skilt, E Enke/Enkemand, P Registreret partnerskab, O Ophævet partnerskab, L Længstlevende af 2 partnere, D Død, 9 Uoplyst civilstand
kom	Not listed here - see DST’s classification
familie_type	1 Ægtepar, 2 Registreret partnerskab, 3 Samlevende par, 4 Samboende par, 5 Enlig (herunder også ikke hjemmeboende børn), 7 Ægtepar forskellig køn, 8 Ægtepar samme køn, 9 Enlig, 10 Ikke hjemmeboende børn
plads	1 Hovedperson, 2 Ægtefælle/partner, 3 Hjemmeboende barn
hustype	1 Enlig mand, 2 Enlig kvinde, 3 Ægtepar, 4 Par i øvrigt, 5 Ikke hjemmeboende børn (under 18 år), 6 Andre husstande bestående af flere familier
fm_mark	1 Bor sammen med begge forældrene, 2 For børn: Bor hos mor, der er i nyt par. For voksne: Bor sammen med mor, 3 For børn: Bor hos enlig mor. For voksne: Værdien findes ikke, 4 For børn: Bor hos far, der er i nyt par. For voksne: Bor sammen med far, 5 For børn: Bor hos enlig far. For voksne: Værdien findes ikke, 6 Bor ikke hos forældrene
herkomst	1 Personer med dansk oprindelse, 2 Indvandrere, 3 Efterkommere, 9 Uoplyst

• koen: DST’s classification KOEN_V1_1980 also defines 9 for not stated, which a delivery may not contain but a value set should. Sex is taken from the tenth digit of the CPR number: even is female, odd is male.
• reg: Do not confuse these with AMT, the pre-2007 counties, which has 16 codes in the ranges 11-14, 21-24, 31-37 and 88. Different geography, different era.
• civst: Codes P, O and L came in with the registered-partnership act of 1 October 1989; before that the set was smaller. Registered partnerships could no longer be entered into from 15 June 2012.
• kom: These codes are valid from 1 January 2007. A study reaching further back needs the pre-reform classification, where the same number can mean a different municipality - confirmed for two reused codes against a current-only DST source: 707 is Norddjurs today, not its pre-2007 meaning, and likewise 849 is Jammerbugt. lookup: below covers only this post-2007 set (99 entries), not the full 278-code values_from file. Christiansø (411) is included in lookup: even though it is not a municipality (see description above): it is a real value a kom column can hold, and DST’s own current-only classification lists it as its own area code alongside the 98 municipalities. Excluding it would just move the “unhandled code” problem this fix is meant to solve onto that one value.
• familie_type: There is no code 6, and the set changed in December 2015. Codes 7 and 8 split the old “Ægtepar” by sex, and codes 9 and 10 split the old “Enlig”, which had included children not living at home. A series that crosses 2015 therefore changes composition without any code going missing: 1 and 5 stop being used and four new codes appear. The exact switch-over dates are on DST’s page and should be read there before a study is dated around them.
• fm_mark: Codes 3 and 5 occur for children only. For an adult the value does not exist, so an adult cohort holding them means the row is not what you think.
• herkomst: A descendant is born in Denmark: neither parent is both a Danish citizen and born in Denmark. So the category says something about the parents, not about where the person was born, and it does not change over a lifetime the way citizenship does.

Where these values come from:

• koen: DST’s classification KOEN_V1_1980 (the code list as CSV).
• reg: DST’s regional classification.
• civst: DST’s variable list for BEF.
• kom: DST’s municipality classification (the code list as CSV).
• familie_type: DST high-quality variable documentation: familie-type.
• plads: DST high-quality variable documentation: plads.
• hustype: DST high-quality variable documentation: hustype.
• fm_mark: DST TIMES documentation: fm_mark.
• herkomst: DST’s herkomst classification (the code list as CSV).

Worth knowing:

• pnr: A person appears once per snapshot, not once in total. Taking a single year loses people who were resident but not in that particular snapshot, so a population is built from the union of all snapshots in the window.
• year: Not a DST variable. It comes from the parquet conversion, which concatenates the yearly deliveries, so it exists in the data you read but not in DST’s own documentation of BEF. Because it is made rather than delivered, the name is not guaranteed: check colnames() rather than assuming.
• alder: Age at the snapshot, not at any date you choose. Recompute from foed_dag and your own index date rather than reusing it.
• referencetid: The date the snapshot describes. Every other column in the row is a status as of this moment, which is what makes BEF a status register rather than an event register.

BEF does not contain date of death. Use DOD (doddato) for censoring, not DODSAARS, which stops in 2001 (see below). DST’s own documentation: statistikdokumentation/befolkningen →.

DOD - Death Register

Døde i Danmark. One row per deceased person, covering 1970-2025 and still updated (DST register overview, last reference date 31 December 2025). This is the register you use to censor at death. It is a short one:

Column	Type	Role	Label
pnr	character	join key	Personal identifier
doddato	date	date	Date of death

All other columns (3)

Column	Type	Role	Label
alder_haend	integer	value	Age at the time of death
cprtjek	character	value	CPR check
cprtype	character	value	CPR type

No published source gives a data type for 5 of these 5 columns, so the Type column is our own assumption. Check with sapply(class) on a row of your own data before relying on it, especially for code columns, which lose their leading zeros if they arrive as numbers.

Join key: pnr.

Joins to other registers:

• pnr joins to BEF (many-to-one).

Worth knowing:

• doddato: This is the column to censor on. Not d_dodsdto, which belongs to DODSAARS and stops in 2001.

DOD may not be in parquet on your project

DOD is sometimes delivered only as a raw SAS file, in which case read_register("dod") will not find it and you read it with haven::read_sas() instead. Ask your data manager for the path. And check the span of your own delivery, which is the years your project ordered rather than everything DST holds:

dod %>% summarise(min(doddato), max(doddato)) %>% collect()

DODSAARS - Cause of Death Register (ends 2001)

Dødsårssagsregistret. One row per deceased person, with the cause of death. It carries a death date too, in d_dodsdto, which is why it is so often used for censoring - but its coverage stops on 31 December 2001. Use it for causes of death up to 2001, and use DOD for the date.

Censoring on dodsaars silently treats everyone who died after 2001 as alive. No error, no warning: they simply stay in your risk set to the end of follow-up, so survival looks better than it is and rates come out too low. The full version of this is pitfall 1 - three death registers.

Column	Type	Role	Label
pnr	character	join key	Personal identifier
d_dodsdto	date	date	Date of death
c_dodsmaade	character	code	Manner of death
c_dod1	character	code	Underlying cause of death
c_dod2	character	code	Contributing cause of death 2
c_dod3	character	code	Contributing cause of death 3
c_dod4	character	code	Contributing cause of death 4
year	integer	date	Register year

All other columns (27)

Column	Type	Role	Label
c_dodskom	character	code	Municipality of death
c_attart	character	code	Type of certificate
c_sex	character	code	Sex
v_alder	numeric	value	Age at death
daar	integer	date	Year of death
c_bopkom	character	code	Municipality of residence at death
c_handsted	character	code	Place of the event
c_liste_14	character	code	Cause group, 14-item list
c_liste_49	character	code	Cause group, 49-item list
c_liste_65	character	code	Cause group, 65-item list
cprtjek	character	code	CPR-tjek
cprtype	character	code	CPR-type
c_aldertim	numeric	value	Dødsalder i timer
c_atckode1	character	code	C_ATCKODE1
c_atckode2	character	code	C_ATCKODE2
c_atckode3	character	code	C_ATCKODE3
c_atckode4	character	code	C_ATCKODE4
c_civstd	character	code	Civilstand
c_institut	character	code	Institution for dødsfald
c_obduktio	character	code	C_OBDUKTIO
c_operatio	character	code	Operationstilkendegivelse
c_u28dg	numeric	value	Angivelse af dødsalder under 28 dage
c_ulyktype	character	code	C_ULYKTYPE
v_aldermdr	numeric	value	Alder i måneder
v_bopamt	numeric	value	V_BOPAMT
v_dodsamt	numeric	value	Dødsstedsamt
v_klok	numeric	value	Tiden for dødens indtræffelse

DST publishes no labels for 4 of these columns. Where the Label column is filled in anyway, it is this guide’s reading of the column name, not an official description.

No published source gives a data type for 29 of these 35 columns, so the Type column is our own assumption. Check with sapply(class) on a row of your own data before relying on it, especially for code columns, which lose their leading zeros if they arrive as numbers.

Join key: pnr.

Joins to other registers:

• pnr joins to BEF (many-to-one).

Value sets for the coded columns (5)

Code system	Values
c_dodsmaade	1 Naturlig død, 2 Ulykke, 3 Selvmord, 4 Drab/vold, 5 Uoplyst
kom	Not listed here - see DST’s classification
icd10	Not listed here - see DST’s classification
atc	Not listed here - see DST’s classification
icd8	Not listed here - see DST’s classification

• c_dodsmaade: Two separate traps. First, this code set is NOT the one used from 2002: in dodsaasg the same column holds c_dodsmaade_2002, where 1 is Voldshandling rather than Naturlig død. Second, code 4 changed meaning. It was Selvmord until 31 December 1990 and Drab/vold from 1 January 1991. dodsaars runs from 1970 to 2001 and so contains both, in one column, with nothing to tell them apart except the date of death. Reading code 4 as one thing across the whole register counts homicides as suicides or the reverse. Codes 6 (Drab/vold) and 9 (Uoplyst) belong only to the old set and stop at the end of 1990; codes 3 and 5 only start in 1991.
• kom: These codes are valid from 1 January 2007. A study reaching further back needs the pre-reform classification, where the same number can mean a different municipality - confirmed for two reused codes against a current-only DST source: 707 is Norddjurs today, not its pre-2007 meaning, and likewise 849 is Jammerbugt. lookup: below covers only this post-2007 set (99 entries), not the full 278-code values_from file. Christiansø (411) is included in lookup: even though it is not a municipality (see description above): it is a real value a kom column can hold, and DST’s own current-only classification lists it as its own area code alongside the 98 municipalities. Excluding it would just move the “unhandled code” problem this fix is meant to solve onto that one value.
• icd10: Do not strip a leading D from these codes. The habit comes from LPR, where the D is really there, and applying it here removes the first character of a real code: E119 becomes 119, which matches nothing and raises no error. The danger is worst where a code genuinely begins with D. ICD-10 chapter D covers in-situ and benign neoplasms, so D46 is myelodysplastic syndrome, a whole code. Strip its “prefix” and you get 46, which looks like a code and is not one.
• atc: As a rule, filter on the full 7-character code rather than on the level columns: atc2 holds three characters, so a longer pattern matched against it can never match, and it returns nothing at all with no error. The level columns are well suited to grouping, and to filtering when every code you want is the same length as the column.
• icd8: A study whose period starts before 1994 is reading two classifications out of one column. ICD-10 codes match nothing in the early years, and the usual substr(c_diag, 2, 4) returns a meaningless fragment of an ICD-8 code rather than failing, so nothing tells you it went wrong.

Where these values come from:

• c_dodsmaade: Kodeark for Doedsaarsagsregisteret, published on www.esundhed.dk.
• kom: DST’s municipality classification (the code list as CSV).
• icd10: WHO ICD-10 browser.
• atc: WHO ATC/DDD Index.
• icd8: Retired classification, no current DST page.

Worth knowing:

• d_dodsdto: A date of death exists here, but the register stops in 2001. Censor on DOD instead, which covers the whole period.
• c_dod1: The underlying cause. Coded in the ICD revision in force at the time of death, so the code system changes inside the register’s own lifetime.
• year: Not a DST variable. It is the partition the yearly deliveries were written into, so filtering on it stops the other years being read at all. Use it to limit how much is read, not to decide when something happened: for that, use the register’s own date column.

DODSAASG - the 2002-2021 link in the chain

Column	Type	Role	Label
pnr	character	join key	Personal identifier
d_dodsdato	date	date	Date of death
c_dodtilgrundl_acme	character	code	Underlying cause of death (ACME)
c_dod_1a	character	code	Cause of death, certificate line 1a
c_dodsmaade	character	code	Manner of death
aar	integer	date	Year

All other columns (35)

Column	Type	Role	Label
c_liste14	character	code	Cause group, 14-item list
c_liste49	character	code	Cause group, 49-item list
c_dodssted	character	code	Place of death
cprtjek	character	code	CPR-tjek
cprtype	character	code	CPR-type
c_bopamtf07	character	code	C_BOPAMTF07
c_bopkom	character	code	C_BOPKOM
c_bopkomf07	character	code	C_BOPKOMF07
c_dodkom	character	code	C_DODKOM
c_dodregion	character	code	C_DODREGION
c_dod_1b	character	code	C_DOD_1B
c_dod_1c	character	code	C_DOD_1C
c_dod_1d	character	code	C_DOD_1D
c_dod_21	character	code	C_DOD_21
c_dod_22	character	code	C_DOD_22
c_dod_23	character	code	C_DOD_23
c_dod_24	character	code	C_DOD_24
c_dod_25	character	code	C_DOD_25
c_dod_26	character	code	C_DOD_26
c_dod_27	character	code	C_DOD_27
c_dod_28	character	code	C_DOD_28
c_findested	character	code	Findested
c_haendelsessted	character	code	C_HAENDELSESSTED
c_laegefunktion	character	code	C_LAEGEFUNKTION
c_listea	character	code	C_LISTEA
c_listeb	character	code	C_LISTEB
c_obduktion	character	code	C_OBDUKTION
c_operation	character	code	C_OPERATION
c_praecis_dodssted	character	code	C_PRAECIS_DODSSTED
c_praecis_findested	character	code	C_PRAECIS_FINDESTED
c_region	character	code	C_REGION
c_sex	character	code	C_SEX
d_findedato	date	date	D_FINDEDATO
d_statdato	date	date	D_STATDATO
v_alder	numeric	value	V_ALDER

DST publishes no labels for 3 of these columns. Where the Label column is filled in anyway, it is this guide’s reading of the column name, not an official description.

No published source gives a data type for 38 of these 41 columns, so the Type column is our own assumption. Check with sapply(class) on a row of your own data before relying on it, especially for code columns, which lose their leading zeros if they arrive as numbers.

Join key: pnr.

Joins to other registers:

• pnr joins to BEF (many-to-one).

Value sets for the coded columns (3)

Code system	Values
icd10	Not listed here - see DST’s classification
c_dodsmaade_2002	0 Ulykke, 1 Voldshandling, 2 Selvmord, 4 Uoplyst, 5 Naturlig død
c_dodssted	-1 Ikke valgt (fordi afdøde er fundet død), 0 Død på sygehus eller hospice, 1 Død på bopælsadressen, 2 Død på kendt adresse, 3 Dødssted uden adresse

• icd10: Do not strip a leading D from these codes. The habit comes from LPR, where the D is really there, and applying it here removes the first character of a real code: E119 becomes 119, which matches nothing and raises no error. The danger is worst where a code genuinely begins with D. ICD-10 chapter D covers in-situ and benign neoplasms, so D46 is myelodysplastic syndrome, a whole code. Strip its “prefix” and you get 46, which looks like a code and is not one.
• c_dodsmaade_2002: This is the set behind the c_dodsmaade column in dodsaasg, and the numbers do not mean what the same numbers mean in dodsaars. There, under c_dodsmaade, 1 is Naturlig død, 2 is Ulykke and 5 is Uoplyst. Here 1 is Voldshandling, 2 is Selvmord and 5 is Naturlig død. A study that spans 2001 and 2002 is reading two different code sets out of two columns with the same name, and every value maps to something plausible in the other set, so nothing looks wrong. A blank is a registration error, not a missing manner of death.
• c_dodssted: All five codes are valid from 1 January 2002. The variable says nothing about deaths before that, which is the whole period dodsaars covers.

Where these values come from:

• icd10: WHO ICD-10 browser.
• c_dodsmaade_2002: Kodeark for Doedsaarsagsregisteret, published on www.esundhed.dk.
• c_dodssted: Kodeark for Doedsaarsagsregisteret, published on www.esundhed.dk.

Worth knowing:

• d_dodsdato: Note the spelling: d_dodsdato here, d_dodsdto in DODSAARS. The two registers do not use the same column names.
• c_dodtilgrundl_acme: The underlying cause, selected by the ACME algorithm. This is usually the one an analysis wants, rather than the individual certificate lines.
• aar: A real DST variable in this register, unlike the year column the parquet conversion adds to most others. If you have seen aar referred to as the partition column elsewhere, this is where the name comes from.

DODSAARSAGER - the current register (2022 onwards)

Column	Type	Role	Label
pnr	character	join key	Personal identifier
doedsdato	date	date	Date of death
doedsaarsag_tilgrundliggende	character	code	Underlying cause of death
doedsaarsag_kode_1	character	code	Cause of death, code 1
doedsmaade_kode	character	code	Manner of death

All other columns (34)

Column	Type	Role	Label
doedsaarsag_liste_14_kode	character	code	Cause group, 14-item list
doedsaarsag_liste_49_kode	character	code	Cause group, 49-item list
doedssted_kode	character	code	Place of death
dw_ek_borger	character	code	Citizen key
flag_valideret	character	code	Validated flag
borger_alder_doedsstatus	numeric	value
borger_bo_kom_doedsstatus	character	code
borger_bo_reg_doedsstatus	character	code
borger_koen_doedsstatus	character	code
cprtjek	character	code
cprtype	character	code
doedsaarsag_gruppering_a_kode	character	code
doedsaarsag_gruppering_b_kode	character	code
doedsaarsag_kode_2	character	code
doedsaarsag_kode_3	character	code
doedsaarsag_kode_4	character	code
doedsaarsag_kode_a	character	code
doedsaarsag_kode_b	character	code
doedsaarsag_kode_c	character	code
doedsaarsag_kode_d	character	code
doedsstatusdato	date	date
doedssted_praecisering_kode	character	code
doedstidspunkt	date	date
findedato	date	date
findested_kode	character	code
findested_praecisering_kode	character	code
findetidspunkt	date	date
haendelsessted_kode	character	code
hospice	character	code
laegefunktion_kode	character	code
obduktionstype_kode	character	code
sygehus_org_reg	character	code
sygehus_shaksghkode	character	code
sygehus_sorkode	character	code

• dw_ek_borger: An LPR3-style surrogate key alongside pnr. Not present in the two older cause-of-death registers.

DST publishes no labels for 10 of these columns. Where the Label column is filled in anyway, it is this guide’s reading of the column name, not an official description.

No published source gives a data type for 39 of these 39 columns, so the Type column is our own assumption. Check with sapply(class) on a row of your own data before relying on it, especially for code columns, which lose their leading zeros if they arrive as numbers.

Join key: pnr.

Joins to other registers:

• pnr joins to BEF (many-to-one).

Value sets for the coded columns (1)

Code system	Values
icd10	Not listed here - see DST’s classification

• icd10: Do not strip a leading D from these codes. The habit comes from LPR, where the D is really there, and applying it here removes the first character of a real code: E119 becomes 119, which matches nothing and raises no error. The danger is worst where a code genuinely begins with D. ICD-10 chapter D covers in-situ and benign neoplasms, so D46 is myelodysplastic syndrome, a whole code. Strip its “prefix” and you get 46, which looks like a code and is not one.

Where these values come from:

• icd10: WHO ICD-10 browser.

Every register in the chain renames the same thing. The cause columns are c_dod1 to c_dod4 here, c_dod_1a to c_dod_1d in DODSAASG from 2002, and doedsaarsag_kode_1 and so on in DODSAARSAGER from 2022. The date column goes d_dodsdto, d_dodsdato, doedsdato - one letter between the first two.

Code written for one part of the chain will not run against another, and a mistyped column name fails as a missing column rather than a wrong answer.

Cause of death is split across three registers, by period. dodsaars runs to 2001, dodsaasg covers 2002-2022, and dodsaarsager takes over from 2022. Each one stops where the next begins, so a study spanning more than one of those windows needs all the registers that overlap its period, chained together. A single one gives you a silently incomplete picture: causes simply stop appearing after its last year.

None of the three is the source for the date of death. For that, use DOD with the column doddato. The cause coding itself is documented by the Danish Health Data Authority: Dødsårsagsregisteret →.

Do you ever need both DOD and DODSAARS?

Only if your own DOD delivery does not reach back far enough for your study period. Then stack them, renaming so the date columns match, and note that the two registers overlap for 1970-2001 - so without the distinct() every early death is counted twice:

deaths <- bind_rows(
  dod %>% select(pnr, death_date = doddato),
  dodsaars %>% select(pnr, death_date = d_dodsdto)
) %>%
  mutate(death_date = as.Date(death_date)) %>%
  distinct(pnr, .keep_all = TRUE) # keeps the DOD row, since it comes first

On DARTER, see Register paths and datastores.

VNDS - Migration Register

One row per migration event per person.

VNDS was closed in February 2026 and split into three registers (VNDS_ind, VNDS_ud, VNDS_hist, described below). It is still documented here because it is the only migration register many projects have: a delivery made before the split will not contain the replacements.

Check colnames() or your project’s register list to see which you have. If you have both, use the new ones and do not mix them with VNDS, or you will count the same migration twice. For emigration dates used in censoring, combine VNDS_ud (2005+) and VNDS_hist (1973-2004) in place of the old VNDS.

What the three replacement registers are, and why DST split them

• VNDS_ind: immigrations to Denmark from 2005 onwards.
• VNDS_ud: emigrations from Denmark from 2005 onwards.
• VNDS_hist: historical migrations 1973-2004, no longer updated.

The split reduces duplicates that arose because immigration and emigration data came from different sources over the years. The earliest data is now frozen in the historical register. Source: DST notice, 19 February 2026.

Column	Type	Role	Label
pnr	character	join key	Personal identifier
indud_kode	character	code	Immigration or emigration
haend_dato	date	date	Date of the migration event
indud_land	character	code	Country migrated from or to

All other columns (2)

Column	Type	Role	Label
cprtjek	character	value	CPR check
cprtype	character	value	CPR type

No published source gives a data type for 6 of these 6 columns, so the Type column is our own assumption. Check with sapply(class) on a row of your own data before relying on it, especially for code columns, which lose their leading zeros if they arrive as numbers.

Join key: pnr.

Joins to other registers:

• pnr joins to BEF (many-to-one).

Value sets for the coded columns (1)

Code system	Values
indud_kode	I Indvandring, U Udvandring

• indud_kode: Use U for censoring at emigration. People who never emigrated have no U event at all, so they are simply absent rather than carrying a missing date. Two caveats from the guide: CPR only has complete immigration and emigration data from 1971, so someone who immigrated before that has no event and looks resident since birth; and a move to Greenland is a status of its own in CPR rather than an emigration. Confirm the values on your own delivery.

Where these values come from:

• indud_kode: DST’s variable list for VNDS.

Worth knowing:

• indud_land: Not listed in the guide’s own table. The country code set has not been sourced.

Use: filter(indud_kode == "U") → min(haend_dato) per pnr for first emigration date. Non-emigrants do not appear in VNDS with a “U” event and get emigration_date = NA.

VNDS_hist - historical migrations (1973-2004)

Column	Type	Role	Label
pnr	character	join key	Personal identifier
indud_kode	character	code	Immigration or emigration
haend_dato	date	date	Date of the migration event
indud_land	character	code	Country migrated from or to

All other columns (2)

Column	Type	Role	Label
cprtjek	character	value
cprtype	character	value

No published source gives a data type for 6 of these 6 columns, so the Type column is our own assumption. Check with sapply(class) on a row of your own data before relying on it, especially for code columns, which lose their leading zeros if they arrive as numbers.

Join key: pnr.

Joins to other registers:

• pnr joins to BEF (many-to-one).

Value sets for the coded columns (1)

Code system	Values
indud_kode	I Indvandring, U Udvandring

• indud_kode: Use U for censoring at emigration. People who never emigrated have no U event at all, so they are simply absent rather than carrying a missing date. Two caveats from the guide: CPR only has complete immigration and emigration data from 1971, so someone who immigrated before that has no event and looks resident since birth; and a move to Greenland is a status of its own in CPR rather than an emigration. Confirm the values on your own delivery.

Where these values come from:

• indud_kode: DST’s variable list for VNDS.

VNDS_ind - immigrations (2005 onwards)

The two new registers drop indud_kode: which direction the event went is now carried by which register you are reading, so a combined series has to add the direction back itself.

Column	Type	Role	Label
pnr	character	join key	Personal identifier
haend_dato	date	date	Date of immigration
indv_land	character	code	Country immigration from
indvmd	character	date	Month of immigration

All other columns (18)

Column	Type	Role	Label
adresse_id	character	code
alder_haend	integer	value
alder_ult	integer	value
bank_statsb	character	code
civst	character	code
cprtjek	character	value
cprtype	character	value
foedreg_kode	character	code
foed_dag	date	date
foed_land	character	code
ie_type	character	code
koen	integer	code
kom	character	code
opr_land	character	code
referencetid	date	date
reg	character	code
statsb	character	code
version	character	value

• adresse_id: DST’s variable list gives this column no label, so the schema records the name only. Its meaning has not been sourced.
• alder_haend: DST’s variable list gives this column no label, so the schema records the name only. Its meaning has not been sourced.
• alder_ult: DST’s variable list gives this column no label, so the schema records the name only. Its meaning has not been sourced.
• bank_statsb: DST’s variable list gives this column no label, so the schema records the name only. Its meaning has not been sourced.
• cprtjek: DST’s variable list gives this column no label, so the schema records the name only. Its meaning has not been sourced.
• cprtype: DST’s variable list gives this column no label, so the schema records the name only. Its meaning has not been sourced.
• foedreg_kode: DST’s variable list gives this column no label, so the schema records the name only. Its meaning has not been sourced.
• foed_dag: DST’s variable list gives this column no label, so the schema records the name only. Its meaning has not been sourced.
• foed_land: DST’s variable list gives this column no label, so the schema records the name only. Its meaning has not been sourced.
• ie_type: DST’s variable list gives this column no label, so the schema records the name only. Its meaning has not been sourced.
• opr_land: DST’s variable list gives this column no label, so the schema records the name only. Its meaning has not been sourced.
• referencetid: DST’s variable list gives this column no label, so the schema records the name only. Its meaning has not been sourced.
• statsb: DST’s variable list gives this column no label, so the schema records the name only. Its meaning has not been sourced.
• version: DST’s variable list gives this column no label, so the schema records the name only. Its meaning has not been sourced.

No published source gives a data type for 22 of these 22 columns, so the Type column is our own assumption. Check with sapply(class) on a row of your own data before relying on it, especially for code columns, which lose their leading zeros if they arrive as numbers.

Join key: pnr.

Joins to other registers:

• pnr joins to BEF (many-to-one).

Value sets for the coded columns (5)

Code system	Values
civst	U Ugift, G Gift (+ separeret), F Skilt, E Enke/Enkemand, P Registreret partnerskab, O Ophævet partnerskab, L Længstlevende af 2 partnere, D Død, 9 Uoplyst civilstand
ie_type	1 Dansk, 2 Indvandrere, 3 Efterkommere, 9 Uoplyst
koen	1 Mand, 2 Kvinde, 9 Uoplyst
kom	Not listed here - see DST’s classification
reg	0 Uoplyst, 81 Nordjylland, 82 Midtjylland, 83 Syddanmark, 84 Hovedstaden, 85 Sjælland

• civst: Codes P, O and L came in with the registered-partnership act of 1 October 1989; before that the set was smaller. Registered partnerships could no longer be entered into from 15 June 2012.
• ie_type: DST’s current high-quality page for IE_TYPE states that the variable has no value set. The codes here come from the older TIMES3 archive page, which does publish them.
• koen: DST’s classification KOEN_V1_1980 also defines 9 for not stated, which a delivery may not contain but a value set should. Sex is taken from the tenth digit of the CPR number: even is female, odd is male.
• kom: These codes are valid from 1 January 2007. A study reaching further back needs the pre-reform classification, where the same number can mean a different municipality - confirmed for two reused codes against a current-only DST source: 707 is Norddjurs today, not its pre-2007 meaning, and likewise 849 is Jammerbugt. lookup: below covers only this post-2007 set (99 entries), not the full 278-code values_from file. Christiansø (411) is included in lookup: even though it is not a municipality (see description above): it is a real value a kom column can hold, and DST’s own current-only classification lists it as its own area code alongside the 98 municipalities. Excluding it would just move the “unhandled code” problem this fix is meant to solve onto that one value.
• reg: Do not confuse these with AMT, the pre-2007 counties, which has 16 codes in the ranges 11-14, 21-24, 31-37 and 88. Different geography, different era.

Where these values come from:

• civst: DST’s variable list for BEF.
• ie_type: DST TIMES3 archive: IE_TYPE.
• koen: DST’s classification KOEN_V1_1980 (the code list as CSV).
• kom: DST’s municipality classification (the code list as CSV).
• reg: DST’s regional classification.

VNDS_ud - emigrations (2005 onwards)

Column	Type	Role	Label
pnr	character	join key	Personal identifier
haend_dato	date	date	Date of emigration
udv_land	character	code	Country emigration to
udvmd	character	date	Month of emigration

All other columns (18)

Column	Type	Role	Label
adresse_id	character	code
alder_haend	integer	value
alder_ult	integer	value
bank_statsb	character	code
civst	character	code
cprtjek	character	value
cprtype	character	value
foedreg_kode	character	code
foed_dag	date	date
foed_land	character	code
ie_type	character	code
koen	integer	code
kom	character	code
opr_land	character	code
referencetid	date	date
reg	character	code
statsb	character	code
version	character	value

• adresse_id: DST’s variable list gives this column no label, so the schema records the name only. Its meaning has not been sourced.
• alder_haend: DST’s variable list gives this column no label, so the schema records the name only. Its meaning has not been sourced.
• alder_ult: DST’s variable list gives this column no label, so the schema records the name only. Its meaning has not been sourced.
• bank_statsb: DST’s variable list gives this column no label, so the schema records the name only. Its meaning has not been sourced.
• cprtjek: DST’s variable list gives this column no label, so the schema records the name only. Its meaning has not been sourced.
• cprtype: DST’s variable list gives this column no label, so the schema records the name only. Its meaning has not been sourced.
• foedreg_kode: DST’s variable list gives this column no label, so the schema records the name only. Its meaning has not been sourced.
• foed_dag: DST’s variable list gives this column no label, so the schema records the name only. Its meaning has not been sourced.
• foed_land: DST’s variable list gives this column no label, so the schema records the name only. Its meaning has not been sourced.
• ie_type: DST’s variable list gives this column no label, so the schema records the name only. Its meaning has not been sourced.
• opr_land: DST’s variable list gives this column no label, so the schema records the name only. Its meaning has not been sourced.
• referencetid: DST’s variable list gives this column no label, so the schema records the name only. Its meaning has not been sourced.
• statsb: DST’s variable list gives this column no label, so the schema records the name only. Its meaning has not been sourced.
• version: DST’s variable list gives this column no label, so the schema records the name only. Its meaning has not been sourced.

No published source gives a data type for 22 of these 22 columns, so the Type column is our own assumption. Check with sapply(class) on a row of your own data before relying on it, especially for code columns, which lose their leading zeros if they arrive as numbers.

Join key: pnr.

Joins to other registers:

• pnr joins to BEF (many-to-one).

Value sets for the coded columns (5)

Code system	Values
civst	U Ugift, G Gift (+ separeret), F Skilt, E Enke/Enkemand, P Registreret partnerskab, O Ophævet partnerskab, L Længstlevende af 2 partnere, D Død, 9 Uoplyst civilstand
ie_type	1 Dansk, 2 Indvandrere, 3 Efterkommere, 9 Uoplyst
koen	1 Mand, 2 Kvinde, 9 Uoplyst
kom	Not listed here - see DST’s classification
reg	0 Uoplyst, 81 Nordjylland, 82 Midtjylland, 83 Syddanmark, 84 Hovedstaden, 85 Sjælland

• civst: Codes P, O and L came in with the registered-partnership act of 1 October 1989; before that the set was smaller. Registered partnerships could no longer be entered into from 15 June 2012.
• ie_type: DST’s current high-quality page for IE_TYPE states that the variable has no value set. The codes here come from the older TIMES3 archive page, which does publish them.
• koen: DST’s classification KOEN_V1_1980 also defines 9 for not stated, which a delivery may not contain but a value set should. Sex is taken from the tenth digit of the CPR number: even is female, odd is male.
• kom: These codes are valid from 1 January 2007. A study reaching further back needs the pre-reform classification, where the same number can mean a different municipality - confirmed for two reused codes against a current-only DST source: 707 is Norddjurs today, not its pre-2007 meaning, and likewise 849 is Jammerbugt. lookup: below covers only this post-2007 set (99 entries), not the full 278-code values_from file. Christiansø (411) is included in lookup: even though it is not a municipality (see description above): it is a real value a kom column can hold, and DST’s own current-only classification lists it as its own area code alongside the 98 municipalities. Excluding it would just move the “unhandled code” problem this fix is meant to solve onto that one value.
• reg: Do not confuse these with AMT, the pre-2007 counties, which has 16 codes in the ranges 11-14, 21-24, 31-37 and 88. Different geography, different era.

Where these values come from:

• civst: DST’s variable list for BEF.
• ie_type: DST TIMES3 archive: IE_TYPE.
• koen: DST’s classification KOEN_V1_1980 (the code list as CSV).
• kom: DST’s municipality classification (the code list as CSV).
• reg: DST’s regional classification.

2. LPR2 - Somatic (up to March 2019)

The register itself is owned by the Danish Health Data Authority, whose documentation covers what is reported and how it changed over time: Landspatientregisteret →.

Join: lpr_adm LEFT JOIN lpr_diag ON recnum.

lpr_adm - Contacts

Column	Type	Role	Label	Years
recnum	character	join key	Contact identifier	1977 to 2019
pnr	character	identifier	Personal identifier	1977 to 2019
d_inddto	date	date	Date of admission	1977 to 2019
d_uddto	date	date	Date of discharge	1977 to 2019
c_pattype	character	code	Patient type	1977 to 2019
c_spec	character	code	Specialty	1977 to 2019
c_adiag	character	code	Action diagnosis	1977 to 2019
c_indm	character	code	Admission mode	1977 to 2019
year	integer	date	Register year

All other columns (43)

Column	Type	Role	Label	Years
c_sgh	character	code	Hospital	1977 to 2019
c_afd	character	code	Department	1977 to 2019
v_alder	numeric	value	Age at the start of the contact	1977 to 2019
c_udm	character	code	Discharge mode	1987 to 2019
c_henm	character	code	Referral mode	1987 to 2019
c_kontaars	character	code	Reason for the contact	1987 to 2019
c_bopamt	character	code	County of residence	1977 to 2004
c_amt	character	code	County	2005 to 2019
v_sengdage	numeric	value	Bed days	1994 to 2019
v_behdage	numeric	value	Treatment days	1977 to 2019
c_sex	character	code	Sex	1977 to 2019
cprtjek	character	code	CPR-tjek	1977 to 2019
cprtype	character	code	CPR-type	1977 to 2019
c_andenbeh	character	code		1977 to 1986
c_blok	character	code	Inddeling af speciale i blokke	1994 to 2019
c_eakt	character	code	Ulykkeskode, aktivitet	1987 to 2003
c_emek	character	code		1987 to 2003
c_emodpart	character	code	Ulykkeskode, modpart	1994 to 2003
c_epart	character	code	Ulykkeskode, egenpart	1994 to 2003
c_ested	character	code	Ulykkeskode, sted	1987 to 2003
c_etraf	character	code	Ulykkeskode, trafikal	1987 to 2003
c_hafd	character	code	Henvisende afdeling	2004 to 2019
c_hsgh	character	code	Henvisende sygehus	2004 to 2019
c_indform	character	code		1977 to 1986
c_indfra	character	code		1977 to 1986
c_kom	character	code	Kommune	1977 to 2019
c_nyafd	character	code		2005 to 2019
c_senstat	character	code		1977 to 1986
c_sghamt	character	code	Sygehusamt	1977 to 2019
c_udtil	character	code		1977 to 1986
c_ulykke	character	code		1977 to 1986
d_ebhdto	date	date	Data for endelig behandling (Variabel udgået efter 31.12.2003)	1996 to 2019
d_fusdto	date	date	Dato for forundersøgelse (Variabel udgået efter 31.12.2003)	1996 to 2019
d_hendto	date	date	Henvisningsdato	1977 to 2019
d_opdatdto	date	date	Intern dato for opdatering af kontakten	2005 to 2019
k_afd	character	code	Afdelingskode	2005 to 2019
leverancedato	date	date		1977 to 2019
version	character	code	Version	1977 to 2019
v_alddg	numeric	value	Alder i dage ved kontaktens start	2001 to 2019
v_aldmdr	numeric	value	Alder i måneder ved kontaktens start	2001 to 2019
v_indminut	numeric	value	Indlæggelsminut	1994 to 2019
v_indtime	numeric	date	Indlæggelsestidspunkt	1977 to 2019
v_udtime	numeric	value	Udskrivningstime	1994 to 2019

• c_udm: Starts in 1987, ten years after the register itself.
• c_bopamt: Ends with the counties themselves: the 2007 local government reform is already visible here in 2004/2005, where c_bopamt stops and c_amt starts. Neither covers the whole register, so a geographic analysis spanning that point needs both.
• v_sengdage: Starts in 1994. Before that, compute the stay from d_inddto and d_uddto instead.
• v_behdage: Ends in 2001.
• c_sex: The coding changes in 2005, from 1/2 to M/K. Prefer koen from BEF for a study variable.

No published source gives a data type for 20 of these 52 columns, so the Type column is our own assumption. Check with sapply(class) on a row of your own data before relying on it, especially for code columns, which lose their leading zeros if they arrive as numbers.

Join key: recnum.

Joins to other registers:

• recnum joins to LPR_DIAG (one-to-many).

Value sets for the coded columns (21)

Code system	Values
pattype	0 Heldoegnspatient (to 2001), Indlagt patient (2002-), 1 Dagpatient (to 1986), Deldoegnspatient (1987-2001), 2 Natpatient (to 1986), Ambulant patient (1987-), 3 Skadestuepatient
icd10_sks	Not listed here - see DST’s classification
indm	1 Akut, 2 Ikke akut, 9 Uoplyst
c_udm	1 Udskrevet/afsluttet til alment praktiserende læge, 2 Udskrevet/afsluttet til praktiserende speciallæge, 3 Udskrevet/afsluttet til eget heldøgnsafsnit eller eget deldøgnsafsnit, 4 Ingen lægelig opfølgning (må kun anvendes for psykiatriske afdelinger), 5 Udskrevet/afsluttet til andet heldøgnsafsnit eller andet deldøgnsafsnit, 6 Udskrevet/afsluttet til ambulatorium, 7 Udeblevet (kun ambulante patienter), 8 Død, 9 Uoplyst, A Andet, B Udskrevet/afsluttet til eget ambulatorium, C Udskrevet/afsluttet til andet ambulatorium, E Behandling i udlandet (hvor sygehus beslutter behandling i udlandet), F Afsluttet til sygehusafsnit, G Afsluttet til sygehusafsnit, venteforløb, K Afsluttet til sygehusafsnit (hjemmet), L Afsluttet til sygehusafsnit, venteforløb (hjemmet)
c_henm	0 Ingen henvisning, 1 Henvist fra alment praktiserende læge, 2 Henvist fra praktiserende speciallæge, 3 Henvist fra eget heldøgnsafsnit eller eget deldøgnsafsnit, 5 Henvist fra andet heldøgnsafsnit eller andet deldøgnsafsnit, 6 Henvist fra skadestue eller ambulatorium, 8 Herfødt, 9 Uoplyst, A Andet, B Eget ambulatorium, C Andet ambulatorium, D Skadestue, E Udlandet (kun direkte henvisninger), F Henvist fra sygehusafsnit, G Henvist fra sygehusafsnit, venteforløb
c_kontaars	1 Sygdom og tilstand uden direkte sammenhæng med udefra påført læsion, 2 Ulykke, 3 Voldshandling, 4 Selvmord/selvmordsforsøg, 5 Senfølge, 6 Komplet skaderegistrering foretages på efterfølgende kontakt, 7 Komplet skaderegistrering foretaget på tidligere kontakt, 8 Andet, 9 Uoplyst
sex_lpr	1 Mand (to 2004), 2 Kvinde (to 2004), M Mand (2005-), K Kvinde (2005-)
c_andenbeh	0 Ja, eget ambulatorium, 1 Ja, andet ambulatorium, eget sygehus, 2 Ja, andet ambulatorium, andet sygehus, 3 Ja, egen læge, 6 Ja, anden, 7 Ingen, 8 Død, 9 Uoplyst
c_blok	1 Medicinsk blok, 2 Kirurgisk blok, 5 Psykiatrisk blok, 6 Laboratorie blok, 8 Øvrige specialer, 9 Andre specialer, 99 Uden for specialer
c_eakt	1 Idræt, sport og motion, 2 Leg, hobby og and fritidsvirksomhed, 3 Erhvervsarbejde, 4 Vitalaktivitet, 5 Ulønnet arbejde, 8 Anden aktivitet, 9 Ikke specificeret aktivitet
c_emek	0 Slag, stød grundet fald på samme niveau, 1 Slag, stød grundet fald på trappe eller til lavere niveau, 2 Slag, stød grundet kontakt med anden genstand, person eller dyr, 3 Klemning, snit eller stik, 4 Fremmedlegeme, 5 Kvælning, 6 Kemisk påvirkning, 7 Termisk, elektrisk eller strålingspåvirkning, 8 Akut overbelastning af legeme eler legemesdel, 9 Anden og ukendt skadesmekanisme
c_emodpart	0 Ingen modpart, 1 Til fods, 2 Cykel, 3 Knallert, 4 Motorcykel eller scooter, 5 Personbil, 6 Varevogn, 7 Lastbil, bus, m.m., 8 Anden transportform, modpart, 9 Ikke specificeret transportform, modpart
c_epart	1 Til fods, 2 Cykel, 3 Knallert, 4 Motorcykel eller scooter, 5 Personbil, 6 Varevogn, 7 Lastbil, bus, m.m., 8 Anden transportform, tilskadekomne, 9 Ikke specificeret transportform, tilskadekomne
c_ested	0 Trafikområde, 1 Boligområde, 2 Produktions- og værkstedsområde, 3 Butiks- handels- og serviceområde, 4 Skole, offentlig administrations- og institutionsområde, 5 Idræts- og sportsområdet, 6 Forlystelses- og parkområdet, 7 Fri natur, 8 Hav, sø og vådområder, 9 Uoplyst
c_etraf	1 Færdselsuheld, 2 Ikke færdselsuheld
c_indform	0 Akut indlæggelse, 1 Indkaldt via forambulatorium, 2 Indkaldt via andet ambulatorium, 3 Anden indkaldelse, 4 Genindkaldt, 5 Indkaldt via koordineret forundersøgelse, 8 Herfødt, 9 Uoplyst
c_indfra	0 Hjemmet, 1 Hjemmet ekskl. døgnistitution, 2 Psykiatrisk hospital/sygehusafdeling, 3 Psykiatrisk dag/nathospital, 4 Somatisk afdeling, 5 Plejehjem/institution, 6 Andet, 8 Født her, 9 Uoplyst, A Andet
c_senstat	0 Ja, samme afdeling, 1 Ja, anden afdeling, 2 Ja, andet sygehus, 3 Ja, alderdoms- eller plejehjem, 5 Ja, rekreationshjem, 6 Ja, andre institutioner, 7 Ingen, 8 Død, 9 Uoplyst
c_udtil	0 Hjemmet, 1 Hjemmet ekskl. døgnistitution, 2 Psykiatrisk hospital/sygehusafdeling, 3 Psykiatrisk dag/nathospital, 4 Somatisk afdeling, 5 Plejehjem/institution, 6 Andet, 8 Død, 9 Uoplyst, A Andet
c_ulykke	0 Nej, 1 Ja, trafikulykke, 2 Ja, arbejdsulykke, 3 Ja, idrætsulykke, 4 Ja, hjemmeulykke, 5 Ja, anden ulykke, 6 Uoplyst, 7 Uoplyst
icd8	Not listed here - see DST’s classification

• pattype: There are four codes, not six, and three of them changed meaning. Code 1 was Dagpatient until 1986 and Deldoegnspatient from 1987, and then stops entirely at the end of 2001. Code 2 was Natpatient until 1986 and Ambulant from 1987. Code 0 was Heldoegnspatient until 2001 and Indlagt patient from 2002. The register only started using 1, 2 and 3 in 1994, so before that essentially every contact is 0. Code 3 was discontinued at the end of 2013, and from 2014 an emergency-room visit arrives as 2 with an acute admission mode in c_indm. Reading 2 as outpatient across the whole register therefore mislabels night patients before 1987 and emergency visits after 2013.
• icd10_sks: The D prefix is a Danish addition, not part of the WHO code. Matching WHO codes directly against LPR without allowing for it returns nothing. Do not carry the habit across to the cause-of-death registers: they hold the plain code, so stripping a D there removes the first real character instead.
• indm: From 2014 this is what separates an emergency-room visit from an ordinary outpatient one, because c_pattype code 3 was discontinued and both arrive as 2. Code 9 (Uoplyst) stops at the end of 2003, so a missing value after that is genuinely missing rather than coded as unknown.
• c_udm: Only codes 1, 8 and 9 reach back to 1987, and 9 stops in 2003. Discharge to another hospital unit is the trap: it was 3, 5, B or C until 2003 or 2004 and becomes F, G, K or L from 2004 onwards. Counting “discharged onwards to hospital” across the whole register therefore needs both sets, and either set alone gives a series with a hole in it. Codes 2, 4, 7 and A only begin in 1995, E in 2002, and K and L in 2006.
• c_henm: Only four codes cover the whole register: 0, 1, 2 and 8. Everything else has a window. Codes 3, 5, 9, B, C and D stop at the end of 2003 and 6 stops at the end of 1998, while F and G only begin in 2004, A in 1995 and E in 2002. The end of 2003 is a break: referrals from a hospital unit were coded 3, 5, B, C or D before it and F or G after it. Counting any of those across the whole period gives a number that changes for administrative reasons alone.
• c_kontaars: Code 4 was Selvtilføjet skade until the end of 1993 and Selvmord/selvmordsforsøg from 1994. Those are not the same population: self-inflicted injury is wider than an attempt at suicide, so a series that spans 1994 changes definition rather than changing level. Codes 5 and 9 stop at the end of 2013. Code 7 only starts in 2011 and code 6 only in 2014, so neither says anything about earlier contacts.
• sex_lpr: The coding changed at the start of 2005: 1/2 until the end of 2004, M/K from 2005. A study spanning that year that filters on c_sex == "2" keeps only the women seen before 2005 and silently drops the rest, with no error and no empty result to warn you. Take sex from BEF instead, where it is koen coded 1/2 throughout, unless you specifically need what the hospital recorded.
• c_andenbeh: Not every code covers the whole register. The periods block gives the window for each one; codes stop being used at 31.12.1986.
• c_eakt: 2 of the 7 codes were REUSED with a different meaning, so the same number does not mean the same thing across the register. The lookup above gives the current meaning; the periods block gives both. Code 1: arbejdsulykke, then idræt, sport og motion from 01.01.1995. Code 2: ikke arbejdsulykke, then leg, hobby og and fritidsvirksomhed from 01.01.1995. Every code in this variable stops by 31.12.2003, so it says nothing about later contacts.
• c_emek: Not every code covers the whole register. The periods block gives the window for each one; codes stop being used at 31.12.2003.
• c_emodpart: Not every code covers the whole register. The periods block gives the window for each one; codes stop being used at 31.12.2003.
• c_epart: Not every code covers the whole register. The periods block gives the window for each one; codes stop being used at 31.12.2003.
• c_ested: Not every code covers the whole register. The periods block gives the window for each one; codes stop being used at 31.12.2003.
• c_etraf: Not every code covers the whole register. The periods block gives the window for each one; codes stop being used at 31.12.2003.
• c_indform: Not every code covers the whole register. The periods block gives the window for each one; codes stop being used at 31.12.1986.
• c_indfra: 4 of the 10 codes were REUSED with a different meaning, so the same number does not mean the same thing across the register. The lookup above gives the current meaning; the periods block gives both. Code 1: anden afdeling, then hjemmet ekskl. døgnistitution from 01.01.1995. Code 2: andet sygehus, then psykiatrisk hospital/sygehusafdeling from 01.01.1995. Code 3: alderdoms- eller plejehjem, then psykiatrisk dag/nathospital from 01.01.1995. Code 5: skadestue, then plejehjem/institution from 01.01.1995. Every code in this variable stops by 31.12.2000, so it says nothing about later contacts.
• c_senstat: Not every code covers the whole register. The periods block gives the window for each one; codes stop being used at 31.12.1986.
• c_udtil: 4 of the 10 codes were REUSED with a different meaning, so the same number does not mean the same thing across the register. The lookup above gives the current meaning; the periods block gives both. Code 1: anden afdeling, then hjemmet ekskl. døgnistitution from 01.01.1995. Code 2: andet sygehus, then psykiatrisk hospital/sygehusafdeling from 01.01.1995. Code 3: alderdoms- eller plejehjem, then psykiatrisk dag/nathospital from 01.01.1995. Code 5: rekreationshjem, then plejehjem/institution from 01.01.1995. Every code in this variable stops by 31.12.2000, so it says nothing about later contacts.
• c_ulykke: Not every code covers the whole register. The periods block gives the window for each one; codes stop being used at 31.12.1986.
• icd8: A study whose period starts before 1994 is reading two classifications out of one column. ICD-10 codes match nothing in the early years, and the usual substr(c_diag, 2, 4) returns a meaningless fragment of an ICD-8 code rather than failing, so nothing tells you it went wrong.

Where these values come from:

• pattype: Kodeark for Landspatientregisteret, published on www.esundhed.dk.
• icd10_sks: SKS browser (medinfo.dk).
• indm: Kodeark for Landspatientregisteret, published on www.esundhed.dk.
• c_udm: Kodeark for Landspatientregisteret, published on www.esundhed.dk.
• c_henm: Kodeark for Landspatientregisteret, published on www.esundhed.dk.
• c_kontaars: Kodeark for Landspatientregisteret, published on www.esundhed.dk.
• sex_lpr: Kodeark for Landspatientregisteret, published on www.esundhed.dk.
• c_andenbeh: Kodeark for Landspatientregisteret, published on www.esundhed.dk.
• c_blok: Kodeark for Landspatientregisteret, published on www.esundhed.dk.
• c_eakt: Kodeark for Landspatientregisteret, published on www.esundhed.dk.
• c_emek: Kodeark for Landspatientregisteret, published on www.esundhed.dk.
• c_emodpart: Kodeark for Landspatientregisteret, published on www.esundhed.dk.
• c_epart: Kodeark for Landspatientregisteret, published on www.esundhed.dk.
• c_ested: Kodeark for Landspatientregisteret, published on www.esundhed.dk.
• c_etraf: Kodeark for Landspatientregisteret, published on www.esundhed.dk.
• c_indform: Kodeark for Landspatientregisteret, published on www.esundhed.dk.
• c_indfra: Kodeark for Landspatientregisteret, published on www.esundhed.dk.
• c_senstat: Kodeark for Landspatientregisteret, published on www.esundhed.dk.
• c_udtil: Kodeark for Landspatientregisteret, published on www.esundhed.dk.
• c_ulykke: Kodeark for Landspatientregisteret, published on www.esundhed.dk.
• icd8: Retired classification, no current DST page.

Worth knowing:

• recnum: The key every other LPR2 dataset joins on. It identifies a contact, not a person.
• d_inddto: Use this as the contact date. It is the admission date, so for an outpatient course it is the date the course started, not the date of a particular visit.
• c_spec: DST publishes what each specialty code means in its department and specialty overview: https://www.dst.dk/da/Statistik/dokumentation/Times/moduldata-for-sociale-forhold–sundhedsvaesen–retsvaesen/spec (in Danish). The codes are not self-explanatory, so look them up rather than grouping on the digits.
• c_adiag: A copy of the contact’s action diagnosis. Use lpr_diag instead: it holds every diagnosis on the contact, not only the action one.
• c_indm: Used together with c_pattype to separate emergency-room contacts from ordinary outpatient ones after about 2014, see the LPR extraction chapter. Available for the register’s whole span, so a missing c_indm is an extract boundary rather than a coverage gap.
• year: Not a DST variable. It is the partition the yearly deliveries were written into, so filtering on it stops the other years being read at all. Use it to limit how much is read, not to decide when something happened: for that, use the register’s own date column.

lpr_diag - Diagnoses

Column	Type	Role	Label	Years
recnum	character	join key	Contact identifier	1977 to 2019
c_diag	character	code	Diagnosis code	1977 to 2019
c_diagtype	character	code	Diagnosis type	1977 to 2019
c_tildiag	character	code	Supplementary diagnosis	1995 to 2019
year	integer	date	Register year

All other columns (3)

Column	Type	Role	Label	Years
c_diagmod	character	code	Diagnosis modification	1977 to 1994
leverancedato	date	date		1977 to 2019
version	character	code	Version	1977 to 2019

No published source gives a data type for 5 of these 8 columns, so the Type column is our own assumption. Check with sapply(class) on a row of your own data before relying on it, especially for code columns, which lose their leading zeros if they arrive as numbers.

Join key: recnum.

Joins to other registers:

• recnum joins to LPR_ADM (many-to-one).

Value sets for the coded columns (4)

Code system	Values
icd10_sks	Not listed here - see DST’s classification
diagtype	A Aktionsdiagnose, B Bidiagnose, G Grundmorbus, naar forskellig fra aktionsdiagnose, H Henvisningsdiagnose, M Midlertidig diagnose, kun for aabne somatisk ambulante besoeg, C Komplikation
c_diagmod	0 Ingen modifikation, 1 Obs. pro., 2 Ej befundet, 3 Sequelae, 4 Antea, 5 Recidivans, 6 Traktatus, 7 Operatus
icd8	Not listed here - see DST’s classification

• icd10_sks: The D prefix is a Danish addition, not part of the WHO code. Matching WHO codes directly against LPR without allowing for it returns nothing. Do not carry the habit across to the cause-of-death registers: they hold the plain code, so stripping a D there removes the first real character instead.
• diagtype: The guide long described this as an A/B/G column. There are six codes, and three of them stop: G runs 1995-2003 only, M 1998-2013 and C 2002-2013. A and B run the whole period, H from 1995. So a comorbidity definition built on G silently covers nine years and nothing else, and filtering to A/B/G drops referral diagnoses entirely. Which types to keep is a case definition, not a technicality: outcomes usually use A and B. Carry the type column into the extract so the definition can be varied later.
• c_diagmod: Not every code covers the whole register. The periods block gives the window for each one; codes stop being used at 31.12.1986, 31.12.1994.
• icd8: A study whose period starts before 1994 is reading two classifications out of one column. ICD-10 codes match nothing in the early years, and the usual substr(c_diag, 2, 4) returns a meaningless fragment of an ICD-8 code rather than failing, so nothing tells you it went wrong.

Where these values come from:

• icd10_sks: SKS browser (medinfo.dk).
• diagtype: Kodeark for Landspatientregisteret, published on www.esundhed.dk.
• c_diagmod: Kodeark for Landspatientregisteret, published on www.esundhed.dk.
• icd8: Retired classification, no current DST page.

Worth knowing:

• year: Not a DST variable. It is the partition the yearly deliveries were written into, so filtering on it stops the other years being read at all. Use it to limit how much is read, not to decide when something happened: for that, use the register’s own date column.

3. LPR2 - Psychiatric (1995 – March 2019)

Psychiatric contacts before March 2019 are in separate registers from somatic LPR2. From March 2019, LPR3 covers both in one table.

Before 1995: inpatients only, and ICD-8

The Danish Psychiatric Central Register is electronic from 1969, but covers only inpatients until 1995 (outpatient visits were added from 1995), and diagnoses before 1994 are coded in ICD-8 (numeric codes, e.g. 290-315, where 290 covers dementia) - not ICD-10 F-codes. These older data are normally not part of the standard extract and are requested separately, via Rigsarkivet or NCRR, Aarhus University. If your study covers that period, you have to map ICD-8 to your F-code groups yourself.

If you forget to query the psychiatric registers for the period 1995–2019, you miss all dementia diagnoses (F00–F03) recorded at geriatric psychiatry outpatient clinics and memory clinics. Those patients will appear dementia-free and remain in the cohort as false negatives.

t_psyk_adm - Psychiatric contacts

The key and person columns may be renamed in your delivery. DST’s variable list gives both psychiatric tables RECNUM and PNR, the same names the somatic registers use. Several deliveries hand them over renamed: DARTER has k_recnum and v_cpr here, and v_recnum in the diagnosis table, so one key ends up with three names. That is data processing, not the register. Run colnames() first, and rename to recnum and pnr if yours differ:

library(fastreg) # read_register()
library(dplyr) # %>%, rename_with(), rename()

psyk_adm <- read_register("t_psyk_adm") %>%
  rename_with(tolower) %>%
  rename(pnr = v_cpr, recnum = k_recnum)

Column	Type	Role	Label
pnr	character	join key	Personal identifier
recnum	character	join key	Contact identifier
c_pattype	character	code	Contact type
c_adiag	character	code	Primary diagnosis
d_inddto	date	date	Admission date
d_uddto	date	date	Discharge date

All other columns (32)

Column	Type	Role	Label
c_indm	character	code	Admission mode
c_udm	character	code	Discharge mode
c_sgh	character	code	Hospital
c_afd	character	code	Department
c_spec	character	code	Specialty
v_indtime	integer	value	Admission hour
v_indminut	integer	value	Admission minute
v_udtime	integer	value	Discharge hour
cprtjek	character	code	CPR-tjek
cprtype	character	code	CPR-type
c_amt	character	code	AMT
c_blok	character	code	Inddeling af speciale i blokke
c_hafd	character	code	Henvisende afdeling
c_henm	character	code	Henvisningsmåde
c_hsgh	character	code	Henvisende sygehus
c_kom	character	code	Kommune
c_kontaars	character	code	Kontaktårsag
c_nyafd	character	code
c_sex	character	code	Køn
c_sghamt	character	code	Sygehusamt
d_ebhdto	date	date	Data for endelig behandling (Variabel udgået efter 31.12.2003)
d_fusdto	date	date	Dato for forundersøgelse (Variabel udgået efter 31.12.2003)
d_hendto	date	date	Henvisningsdato
d_opdatdto	date	date	Intern dato for opdatering af kontakten
k_afd	character	code	Afdelingskode
leverancedato	date	date
version	character	code	Version
v_alddg	numeric	value	Alder i dage ved kontaktens start
v_alder	numeric	value	Alder i år ved kontaktens start
v_aldmdr	numeric	value	Alder i måneder ved kontaktens start
v_behdage	numeric	value	Behandlingsdage (Variabel udgået efter 31.12.2001)
v_sengdage	numeric	value	Sengedage

No published source gives a data type for 6 of these 38 columns, so the Type column is our own assumption. Check with sapply(class) on a row of your own data before relying on it, especially for code columns, which lose their leading zeros if they arrive as numbers.

Join key: recnum.

Joins to other registers:

• recnum joins to T_PSYK_DIAG (one-to-many).

Value sets for the coded columns (8)

Code system	Values
pattype	0 Heldoegnspatient (to 2001), Indlagt patient (2002-), 1 Dagpatient (to 1986), Deldoegnspatient (1987-2001), 2 Natpatient (to 1986), Ambulant patient (1987-), 3 Skadestuepatient
icd10_sks	Not listed here - see DST’s classification
indm	1 Akut, 2 Ikke akut, 9 Uoplyst
c_udm	1 Udskrevet/afsluttet til alment praktiserende læge, 2 Udskrevet/afsluttet til praktiserende speciallæge, 3 Udskrevet/afsluttet til eget heldøgnsafsnit eller eget deldøgnsafsnit, 4 Ingen lægelig opfølgning (må kun anvendes for psykiatriske afdelinger), 5 Udskrevet/afsluttet til andet heldøgnsafsnit eller andet deldøgnsafsnit, 6 Udskrevet/afsluttet til ambulatorium, 7 Udeblevet (kun ambulante patienter), 8 Død, 9 Uoplyst, A Andet, B Udskrevet/afsluttet til eget ambulatorium, C Udskrevet/afsluttet til andet ambulatorium, E Behandling i udlandet (hvor sygehus beslutter behandling i udlandet), F Afsluttet til sygehusafsnit, G Afsluttet til sygehusafsnit, venteforløb, K Afsluttet til sygehusafsnit (hjemmet), L Afsluttet til sygehusafsnit, venteforløb (hjemmet)
c_blok	1 Medicinsk blok, 2 Kirurgisk blok, 5 Psykiatrisk blok, 6 Laboratorie blok, 8 Øvrige specialer, 9 Andre specialer, 99 Uden for specialer
c_henm	0 Ingen henvisning, 1 Henvist fra alment praktiserende læge, 2 Henvist fra praktiserende speciallæge, 3 Henvist fra eget heldøgnsafsnit eller eget deldøgnsafsnit, 5 Henvist fra andet heldøgnsafsnit eller andet deldøgnsafsnit, 6 Henvist fra skadestue eller ambulatorium, 8 Herfødt, 9 Uoplyst, A Andet, B Eget ambulatorium, C Andet ambulatorium, D Skadestue, E Udlandet (kun direkte henvisninger), F Henvist fra sygehusafsnit, G Henvist fra sygehusafsnit, venteforløb
c_kontaars	1 Sygdom og tilstand uden direkte sammenhæng med udefra påført læsion, 2 Ulykke, 3 Voldshandling, 4 Selvmord/selvmordsforsøg, 5 Senfølge, 6 Komplet skaderegistrering foretages på efterfølgende kontakt, 7 Komplet skaderegistrering foretaget på tidligere kontakt, 8 Andet, 9 Uoplyst
sex_lpr	1 Mand (to 2004), 2 Kvinde (to 2004), M Mand (2005-), K Kvinde (2005-)

• pattype: There are four codes, not six, and three of them changed meaning. Code 1 was Dagpatient until 1986 and Deldoegnspatient from 1987, and then stops entirely at the end of 2001. Code 2 was Natpatient until 1986 and Ambulant from 1987. Code 0 was Heldoegnspatient until 2001 and Indlagt patient from 2002. The register only started using 1, 2 and 3 in 1994, so before that essentially every contact is 0. Code 3 was discontinued at the end of 2013, and from 2014 an emergency-room visit arrives as 2 with an acute admission mode in c_indm. Reading 2 as outpatient across the whole register therefore mislabels night patients before 1987 and emergency visits after 2013.
• icd10_sks: The D prefix is a Danish addition, not part of the WHO code. Matching WHO codes directly against LPR without allowing for it returns nothing. Do not carry the habit across to the cause-of-death registers: they hold the plain code, so stripping a D there removes the first real character instead.
• indm: From 2014 this is what separates an emergency-room visit from an ordinary outpatient one, because c_pattype code 3 was discontinued and both arrive as 2. Code 9 (Uoplyst) stops at the end of 2003, so a missing value after that is genuinely missing rather than coded as unknown.
• c_udm: Only codes 1, 8 and 9 reach back to 1987, and 9 stops in 2003. Discharge to another hospital unit is the trap: it was 3, 5, B or C until 2003 or 2004 and becomes F, G, K or L from 2004 onwards. Counting “discharged onwards to hospital” across the whole register therefore needs both sets, and either set alone gives a series with a hole in it. Codes 2, 4, 7 and A only begin in 1995, E in 2002, and K and L in 2006.
• c_henm: Only four codes cover the whole register: 0, 1, 2 and 8. Everything else has a window. Codes 3, 5, 9, B, C and D stop at the end of 2003 and 6 stops at the end of 1998, while F and G only begin in 2004, A in 1995 and E in 2002. The end of 2003 is a break: referrals from a hospital unit were coded 3, 5, B, C or D before it and F or G after it. Counting any of those across the whole period gives a number that changes for administrative reasons alone.
• c_kontaars: Code 4 was Selvtilføjet skade until the end of 1993 and Selvmord/selvmordsforsøg from 1994. Those are not the same population: self-inflicted injury is wider than an attempt at suicide, so a series that spans 1994 changes definition rather than changing level. Codes 5 and 9 stop at the end of 2013. Code 7 only starts in 2011 and code 6 only in 2014, so neither says anything about earlier contacts.
• sex_lpr: The coding changed at the start of 2005: 1/2 until the end of 2004, M/K from 2005. A study spanning that year that filters on c_sex == "2" keeps only the women seen before 2005 and silently drops the rest, with no error and no empty result to warn you. Take sex from BEF instead, where it is koen coded 1/2 throughout, unless you specifically need what the hospital recorded.

Where these values come from:

• pattype: Kodeark for Landspatientregisteret, published on www.esundhed.dk.
• icd10_sks: SKS browser (medinfo.dk).
• indm: Kodeark for Landspatientregisteret, published on www.esundhed.dk.
• c_udm: Kodeark for Landspatientregisteret, published on www.esundhed.dk.
• c_blok: Kodeark for Landspatientregisteret, published on www.esundhed.dk.
• c_henm: Kodeark for Landspatientregisteret, published on www.esundhed.dk.
• c_kontaars: Kodeark for Landspatientregisteret, published on www.esundhed.dk.
• sex_lpr: Kodeark for Landspatientregisteret, published on www.esundhed.dk.

Worth knowing:

• pnr: DST’s variable list calls this column PNR. Some deliveries rename it: some deliveries hand it over as v_cpr. The name is a product of the data processing, not of the register, so check your own columns and rename.
• recnum: DST’s variable list calls this column RECNUM, the same name the somatic lpr_adm uses. Some deliveries rename it, and not consistently: the contact table can arrive as k_recnum and the diagnosis table as v_recnum. Rename both to recnum before joining.
• c_pattype: Present here even where the somatic lpr_adm extract leaves it out, so the psychiatric side of a study can be classified when the somatic side cannot. Do not assume symmetry between the two extracts.

t_psyk_diag - Psychiatric diagnoses

psyk_diag <- read_register("t_psyk_diag") %>%
  rename_with(tolower) %>%
  rename(recnum = v_recnum)

Column	Type	Role	Label
recnum	character	join key	Contact identifier
c_diag	character	code	Diagnosis code
c_diagtype	character	code	Diagnosis type
c_tildiag	character	code	Supplementary diagnosis

All other columns (2)

Column	Type	Role	Label
leverancedato	date	date
version	character	code	Version

No published source gives a data type for 2 of these 6 columns, so the Type column is our own assumption. Check with sapply(class) on a row of your own data before relying on it, especially for code columns, which lose their leading zeros if they arrive as numbers.

Join key: recnum.

Joins to other registers:

• recnum joins to T_PSYK_ADM (many-to-one).

Value sets for the coded columns (2)

Code system	Values
icd10_sks	Not listed here - see DST’s classification
diagtype	A Aktionsdiagnose, B Bidiagnose, G Grundmorbus, naar forskellig fra aktionsdiagnose, H Henvisningsdiagnose, M Midlertidig diagnose, kun for aabne somatisk ambulante besoeg, C Komplikation

• icd10_sks: The D prefix is a Danish addition, not part of the WHO code. Matching WHO codes directly against LPR without allowing for it returns nothing. Do not carry the habit across to the cause-of-death registers: they hold the plain code, so stripping a D there removes the first real character instead.
• diagtype: The guide long described this as an A/B/G column. There are six codes, and three of them stop: G runs 1995-2003 only, M 1998-2013 and C 2002-2013. A and B run the whole period, H from 1995. So a comorbidity definition built on G silently covers nine years and nothing else, and filtering to A/B/G drops referral diagnoses entirely. Which types to keep is a case definition, not a technicality: outcomes usually use A and B. Carry the type column into the extract so the definition can be varied later.

Where these values come from:

• icd10_sks: SKS browser (medinfo.dk).
• diagtype: Kodeark for Landspatientregisteret, published on www.esundhed.dk.

Worth knowing:

• recnum: DST’s variable list calls this column RECNUM. Some deliveries rename it, and not consistently: this table can arrive as v_recnum while the contact table gets k_recnum. One key, three names. Check and rename to recnum.

4. LPR3 (March 2019 and onwards)

LPR3 covers both somatic and psychiatric contacts in one table. Join: lpr_a_kontakt LEFT JOIN lpr_a_diagnose ON dw_ek_kontakt.

The “a” in lpr_a_diagnose does not mean A-type diagnoses. It refers to the analysis model designation for the LPR3 series (LPR_A, introduced 2025). The table contains all types, A, B and G, so you still need to filter on diag_kode_type.

lpr_a_kontakt - Contacts

Column	Type	Role	Label
dw_ek_kontakt	character	join key	Contact identifier
dw_ek_forloeb	character	code	Course identifier
pnr	character	identifier	Personal identifier
kont_starttidspunkt	datetime	date	Contact start
kont_sluttidspunkt	datetime	date	Contact end
kont_type	character	code	Contact type
lprindberetningssystem	character	code	Reporting system
adiag	character	code	Primary diagnosis
prioritet	character	code	Priority
kont_ans_hovedspec	character	code	Responsible main specialty
year	integer	date	Register year

All other columns (43)

Column	Type	Role	Label
dw_sk_sygehusophold	character	code	Hospital stay identifier
dw_ek_helbredsforloeb	character	code	Health course identifier
dw_ek_borger	character	code	Citizen identifier
adiag_tekst	character	value	Primary diagnosis, text
kont_type_tekst	character	value	Contact type, text
kont_patient_type	character	code	Patient type
kont_patient_type_tekst	character	value	Patient type, text
prioritet_tekst	character	value	Priority, text
kont_aarsag	character	code	Reason for the contact
kont_aarsag_tekst	character	value	Reason for the contact, text
kont_henv_aarsag	character	code	Referral reason
kont_henv_aarsag_tekst	character	value	Referral reason, text
kont_henv_maade	character	code	Referral mode
kont_henv_maade_tekst	character	value	Referral mode, text
kont_henv_instans	character	code	Referring body
kont_henv_tidspunkt	datetime	date	Referral time
kont_indb_tidspunkt	datetime	date	Reporting time
beh_starttidspunkt	datetime	date	Treatment start
flag_kont_afsluttet	numeric	code	Contact closed flag
kont_ans	character	code	Responsible unit
kont_ans_inst	character	code	Responsible institution
kont_ans_hovedspec_shak	character	code	Responsible main specialty, SHAK
kont_ans_geo_reg	character	code	Region of the treating unit, geographic
kont_ans_geo_reg_tekst	character	value	Region of the treating unit, text
kont_ans_org_reg	character	code	Region of the treating unit, organisational
kont_ans_org_reg_tekst	character	value	Organisational region, text
kont_inst_ejertype	character	code	Institution ownership type
kont_fir_kode	character	code	Company code
kont_fir_tekst	character	value	Company, text
kont_fritvalg	character	code	Free choice of hospital
kont_fritvalg_tekst	character	value	Free choice, text
kont_lpr_entity_id	character	code	LPR entity identifier
borger_koen	character	code	Sex
borger_foedselsdato	date	date	Date of birth
borger_doedsdato	date	date	Date of death
borger_alder_aar_ind	numeric	value	Age in years at contact start
borger_alder_aar_ud	numeric	value	Age in years at contact end
borger_bo_kom	character	code	Municipality of residence
borger_bo_kom_tekst	character	value	Municipality of residence, text
borger_bo_reg	character	code	Region of residence
borger_bo_reg_tekst	character	value	Region of residence, text
cprtjek	character	code	CPR check
cprtype	character	code	CPR type

• dw_sk_sygehusophold: A stay can gather several contacts. Counting rows here is not the same as counting admissions.
• dw_ek_borger: An internal person key. Use pnr for joins to other registers; this one does not travel outside LPR3.
• kont_indb_tidspunkt: When the contact was reported, not when it happened. Recent months look incomplete because reporting lags.
• flag_kont_afsluttet: An open contact has no end time yet, so durations computed near the end of the data are wrong rather than missing.
• borger_koen: Sex as recorded on the contact, as text rather than a number. No published source gives its value set, so this schema records none. Checked without success: Sundhedsdatastyrelsen’s Vejledning til LPR3_F, which documents these very research tables; the LPR3 reporting guidance; esundhed’s LPR documentation, which covers LPR2 only; and DST’s variable list, which names the column but gives neither label nor values. Treat this as settled rather than as something still to look up. The neighbouring register is no guide either: LPR2’s c_sex switched from 1/2 to M/K in 2005, so both codings exist in the family and neither can be assumed here. If you need sex as a study variable, take koen from BEF, which is documented and stable; if you need what the hospital recorded, check what your own column contains before filtering on it.
• borger_foedselsdato: A copy from CPR carried on the contact, so it only exists for people who had a hospital contact. BEF is the source to use for a study variable.
• borger_doedsdato: A death date carried on the contact. It is not a death register: use DOD for mortality, or you only see people who had a hospital contact.

DST publishes no labels for 53 of these columns. Where the Label column is filled in anyway, it is this guide’s reading of the column name, not an official description.

Join key: dw_ek_kontakt.

Joins to other registers:

• dw_ek_kontakt joins to LPR_A_DIAGNOSE (one-to-many).

Value sets for the coded columns (4)

Code system	Values
kont_type	Not listed here - see DST’s classification
icd10_sks	Not listed here - see DST’s classification
reg	0 Uoplyst, 81 Nordjylland, 82 Midtjylland, 83 Syddanmark, 84 Hovedstaden, 85 Sjælland
kom	Not listed here - see DST’s classification

• kont_type: ALCA00 means physical attendance, which is the closest LPR3 gets to LPR2’s inpatient flag. It marks attendance, not admission, so a study that treats it as “was admitted” will include outpatient visits. Check what your own delivery holds before filtering: single digits and SKS codes have been seen side by side in the same year, so kont_type == "ALCA00" can silently drop rows that are the same kind of contact recorded in the other form. Cross-tabulate it against lprindberetningssystem first. MiniPAS was the route private providers reported through, so the two forms are not only two notations, they are also two different parts of the health service.
• icd10_sks: The D prefix is a Danish addition, not part of the WHO code. Matching WHO codes directly against LPR without allowing for it returns nothing. Do not carry the habit across to the cause-of-death registers: they hold the plain code, so stripping a D there removes the first real character instead.
• reg: Do not confuse these with AMT, the pre-2007 counties, which has 16 codes in the ranges 11-14, 21-24, 31-37 and 88. Different geography, different era.
• kom: These codes are valid from 1 January 2007. A study reaching further back needs the pre-reform classification, where the same number can mean a different municipality - confirmed for two reused codes against a current-only DST source: 707 is Norddjurs today, not its pre-2007 meaning, and likewise 849 is Jammerbugt. lookup: below covers only this post-2007 set (99 entries), not the full 278-code values_from file. Christiansø (411) is included in lookup: even though it is not a municipality (see description above): it is a real value a kom column can hold, and DST’s own current-only classification lists it as its own area code alongside the 98 municipalities. Excluding it would just move the “unhandled code” problem this fix is meant to solve onto that one value.

Where these values come from:

• kont_type: Vejledning til indberetning til LPR3, published on medinfo.dk.
• icd10_sks: SKS browser (medinfo.dk).
• reg: DST’s regional classification.
• kom: DST’s municipality classification (the code list as CSV).

Worth knowing:

• dw_ek_kontakt: The key the diagnosis and procedure tables join on.
• dw_ek_forloeb: One level above the contact: a course of treatment can span several contacts, so joining on this is not the same as joining on the contact.
• kont_starttidspunkt: A datetime, not a date. as.Date() it before comparing with an index date.
• lprindberetningssystem: Filter to “LPR3”. The table reaches back to 2017, and the outpatient contacts from before March 2019 are also in LPR2, so combining the two without this filter counts the same contact twice. The column also separates the two delivery formats, LPR_F and LPR_A.
• adiag: The contact’s action diagnosis, repeated here so simple analyses need not join lpr_a_diagnose. Secondary diagnoses are only in the diagnosis table, so filtering on this column alone misses them.
• prioritet: The code ATA1 marks an acute contact. Together with the contact’s duration this is how LPR3 substitutes for LPR2’s c_pattype.
• year: Not a DST variable. It is the partition the yearly deliveries were written into, so filtering on it stops the other years being read at all. Use it to limit how much is read, not to decide when something happened: for that, use the register’s own date column.

All confirmed columns in lpr_a_kontakt

pnr, dw_ek_kontakt, kont_starttidspunkt, kont_sluttidspunkt, kont_type, kont_type_tekst, kont_patient_type, kont_patient_type_tekst, kont_ans_hovedspec, kont_ans_hovedspec_shak, kont_ans_inst, kont_ans, kont_ans_geo_reg, kont_ans_geo_reg_tekst, kont_ans_org_reg, kont_ans_org_reg_tekst, borger_doedsdato, borger_foedselsdato, borger_koen, borger_alder_aar_ind, borger_alder_aar_ud, borger_bo_kom, borger_bo_kom_tekst, borger_bo_reg, borger_bo_reg_tekst, dw_sk_sygehusophold, dw_ek_helbredsforloeb, dw_ek_forloeb, dw_ek_borger, adiag, adiag_tekst, beh_starttidspunkt, flag_kont_afsluttet, kont_aarsag, kont_aarsag_tekst, kont_indb_tidspunkt, kont_fir_kode, kont_fir_tekst, kont_fritvalg, kont_fritvalg_tekst, kont_henv_aarsag, kont_henv_aarsag_tekst, kont_henv_instans, kont_henv_maade, kont_henv_maade_tekst, kont_henv_tidspunkt, kont_inst_ejertype, lprindberetningssystem, prioritet, prioritet_tekst, kont_lpr_entity_id, cprtjek, cprtype, year

DST’s variable list: LPR_A_KONTAKT → (in Danish). Look up the specialty code kont_ans_hovedspec in DST’s specialty/department overview → (in Danish).

lpr_a_diagnose - Diagnoses

Column	Type	Role	Label
dw_ek_kontakt	character	join key	Contact identifier
diag_kode	character	code	Diagnosis code
diag_kode_type	character	code	Diagnosis type
senere_afkraeftet	character	code	Later retracted
diag_kode_tekst	character	value	Diagnosis code, text
year	integer	date	Register year

All other columns (6)

Column	Type	Role	Label
diag_parent_kode	character	code	Parent diagnosis code
lprindberetningssystem	character	code	Reporting system
diag_kode_type_tekst	character	value	Diagnosis type, text
diag_parent_kode_tekst	character	value	Parent diagnosis code, text
diag_parent_kode_type	character	code	Parent diagnosis type
diag_parent_kode_type_tekst	character	value	Parent diagnosis type, text

• lprindberetningssystem: Filter to “LPR3”: the table holds rows from two reporting formats, and not doing so duplicates rows.

DST publishes no labels for 11 of these columns. Where the Label column is filled in anyway, it is this guide’s reading of the column name, not an official description.

Join key: dw_ek_kontakt.

Joins to other registers:

• dw_ek_kontakt joins to LPR_A_KONTAKT (many-to-one).

Value sets for the coded columns (2)

Code system	Values
icd10_sks	Not listed here - see DST’s classification
diagtype	A Aktionsdiagnose, B Bidiagnose, G Grundmorbus, naar forskellig fra aktionsdiagnose, H Henvisningsdiagnose, M Midlertidig diagnose, kun for aabne somatisk ambulante besoeg, C Komplikation

• icd10_sks: The D prefix is a Danish addition, not part of the WHO code. Matching WHO codes directly against LPR without allowing for it returns nothing. Do not carry the habit across to the cause-of-death registers: they hold the plain code, so stripping a D there removes the first real character instead.
• diagtype: The guide long described this as an A/B/G column. There are six codes, and three of them stop: G runs 1995-2003 only, M 1998-2013 and C 2002-2013. A and B run the whole period, H from 1995. So a comorbidity definition built on G silently covers nine years and nothing else, and filtering to A/B/G drops referral diagnoses entirely. Which types to keep is a case definition, not a technicality: outcomes usually use A and B. Carry the type column into the extract so the definition can be varied later.

Where these values come from:

• icd10_sks: SKS browser (medinfo.dk).
• diagtype: Kodeark for Landspatientregisteret, published on www.esundhed.dk.

Worth knowing:

• senere_afkraeftet: A diagnosis that was subsequently withdrawn. Keeping these counts conditions the patient turned out not to have.
• diag_kode_tekst: The code spelled out. Convenient for reading, but do not group on it: the text can change between years while the code stays the same.
• year: Not a DST variable. It is the partition the yearly deliveries were written into, so filtering on it stops the other years being read at all. Use it to limit how much is read, not to decide when something happened: for that, use the register’s own date column.

Standard filter for senere_afkraeftet:

filter(is.na(senere_afkraeftet) | senere_afkraeftet != "Ja")

5. LPR - SKS procedure codes

SKS (Sundhedsvæsenets Klassifikations System - the Danish Health Classification System) is the Danish classification system for operations and procedures - equivalent to the NOMESCO codes used in the other Nordic countries. Bariatric surgery has e.g. codes KJDF10 (RYGB) and KJDF40 (sleeve gastrectomy). Look up codes in the SKS browser → (in Danish).

SKS codes are split across two registers depending on period. For full coverage both must be queried and the results bound together.

Neither procedure table holds pnr. It is fetched by joining to lpr_adm (LPR2) or lpr_a_kontakt (LPR3).

lpr_sksopr - LPR2 SKS procedures (up to 2018)

Location (DARTER): parquet-registers/lpr_sksopr

lpr_sksopr <- read_register("lpr_sksopr") %>%
  rename_with(tolower)

Column	Type	Role	Label
recnum	character	join key	Contact identifier
c_opr	character	code	Procedure code
c_oprart	character	code	Procedure type
c_osgh	character	code	Hospital performing the procedure
c_tilopr	character	code	Supplementary code
d_odto	date	date	Procedure date
year	integer	date	Register year

All other columns (5)

Column	Type	Role	Label
c_oafd	character	code	Procedurende afdeling (operation)
leverancedato	date	date
version	character	code	Version
v_ominut	numeric	value	Procedureminut (operation)
v_otime	numeric	value	Proceduretime (operation)

No published source gives a data type for 8 of these 12 columns, so the Type column is our own assumption. Check with sapply(class) on a row of your own data before relying on it, especially for code columns, which lose their leading zeros if they arrive as numbers.

Join key: recnum.

Joins to other registers:

• recnum joins to LPR_ADM (many-to-one).

Value sets for the coded columns (2)

Code system	Values
sks	Not listed here - see DST’s classification
oprart	V Vigtigste operation i afsluttet kontakt, P Vigtigste operation i operativt indgreb, D Deloperation, + Tillaegskode

• sks: The codes are hierarchical, so a prefix match selects a whole branch. That also makes it easy to select more than you meant: check how many characters your intended group actually needs before filtering with starts_with().
• oprart: Counting rows in the procedure table counts add-on codes and sub-procedures as procedures. If you want one row per operation, filter to V or P first. All four codes run from 1996 with no breaks.

Where these values come from:

• sks: SKS browser (medinfo.dk).
• oprart: Kodeark for Landspatientregisteret, published on www.esundhed.dk.

Worth knowing:

• c_opr: The SKS procedure code. Surgical codes start with K.
• year: Not a DST variable. It is the partition the yearly deliveries were written into, so filtering on it stops the other years being read at all. Use it to limit how much is read, not to decide when something happened: for that, use the register’s own date column.

lpr_a_procregistrering - LPR3 SKS procedures (2019 and onwards)

The LPR3 procedure table. It sits alongside lpr_a_kontakt and lpr_a_diagnose, and joins to contacts on dw_ek_kontakt exactly like the diagnosis table does. Column names confirmed against DST’s variable list →.

Column	Type	Role	Label
dw_ek_kontakt	character	join key	Contact identifier
proc_kode	character	code	Procedure code
proc_starttidspunkt	datetime	date	Procedure start
proc_kode_type	character	code	Procedure code type
proc_sluttidspunkt	datetime	date	Procedure end
proc_parent_kode	character	code	Parent procedure code
dw_ek_forloeb	character	code	Course identifier
flag_proc_uden_kont	character	code	Procedure without a contact
lprindberetningssystem	character	code	Reporting system

All other columns (11)

Column	Type	Role	Label
dw_ek_procedureregistrering	character	identifier	Procedure registration identifier
dw_sk_sygehusophold	character	code	Hospital stay identifier
proc_indb_tidspunkt	datetime	date	Reporting time
proc_kode_tekst	character	value	Procedure code, text
proc_kode_type_tekst	character	value	Procedure code type, text
proc_parent_kode_tekst	character	value	Parent procedure code, text
proc_parent_kode_type	character	code	Parent procedure code type
proc_parent_kode_type_tekst	character	value	Parent procedure code type, text
proc_lpr_entity_id	character	code	LPR entity identifier
prod_enh	character	code	Performing unit
prod_inst	character	code	Performing institution

• proc_indb_tidspunkt: When the procedure was reported, not when it happened. Recent months look incomplete because reporting lags.

DST publishes no labels for 20 of these columns. Where the Label column is filled in anyway, it is this guide’s reading of the column name, not an official description.

No published source gives a data type for 20 of these 20 columns, so the Type column is our own assumption. Check with sapply(class) on a row of your own data before relying on it, especially for code columns, which lose their leading zeros if they arrive as numbers.

Join key: dw_ek_kontakt.

Joins to other registers:

• dw_ek_kontakt joins to LPR_A_KONTAKT (many-to-one).

Value sets for the coded columns (1)

Code system	Values
sks	Not listed here - see DST’s classification

• sks: The codes are hierarchical, so a prefix match selects a whole branch. That also makes it easy to select more than you meant: check how many characters your intended group actually needs before filtering with starts_with().

Where these values come from:

• sks: SKS browser (medinfo.dk).

Worth knowing:

• dw_ek_kontakt: The key to lpr_a_kontakt, which is where pnr lives. This table carries no person identifier of its own.
• proc_starttidspunkt: A datetime, not a date. as.Date() it before comparing with an index date.
• proc_kode_type: “P” marks a procedure, “+” an add-on code. An add-on code modifies the procedure above it and is not a procedure in its own right, so counting all rows overcounts.
• proc_parent_kode: LPR3 nests procedures the same way it nests diagnoses: an add-on code points at the procedure it belongs to.
• dw_ek_forloeb: One level above the contact. Where dw_ek_kontakt is empty, this is the only route back to a person, and it reaches a whole course of treatment rather than a single contact.
• flag_proc_uden_kont: Flags a procedure with no contact attached. Those rows cannot be joined to lpr_a_kontakt at all, so a plain inner join drops them silently.
• lprindberetningssystem: Filter to “LPR3” when combining with LPR2, or contacts reported under both systems are counted twice.

There is no pnr here, as in every LPR procedure table: it identifies procedures, not people. Join to lpr_a_kontakt to get one.

A few rows have no contact to join to, and DST flags them for you in flag_proc_uden_kont. Count them before you drop them.

library(fastreg) # read_register()
library(dplyr) # filter, select, inner_join, mutate, collect

proc <- read_register("lpr_a_procregistrering") %>%
  rename_with(tolower) %>%
  filter(lprindberetningssystem == "LPR3") %>% # same duplicate filter as the contacts
  filter(proc_kode %in% !!MY_CODES) %>% # your SKS codes
  select(dw_ek_kontakt, proc_starttidspunkt)

surgery <- read_register("lpr_a_kontakt") %>%
  rename_with(tolower) %>%
  filter(lprindberetningssystem == "LPR3") %>%
  select(dw_ek_kontakt, pnr) %>% # pnr lives here
  inner_join(proc, by = "dw_ek_kontakt") %>%
  mutate(date_procedure = as.Date(proc_starttidspunkt)) %>%
  select(pnr, date_procedure) %>%
  collect()

lpr_sksube - SKS examinations and treatments (ZZ codes)

Examination and treatment codes (ZZ codes) live in lpr_sksube (LPR2), separate from the operation codes in lpr_sksopr above. Join like the other SKS tables, via recnum. The register is shaped exactly like lpr_sksopr.

Column	Type	Role	Label
recnum	character	join key	Contact identifier
c_opr	character	code	Procedure code
d_odto	date	date	Procedure date
year	integer	date	Register year

All other columns (8)

Column	Type	Role	Label
c_oprart	character	code	Procedure type
c_osgh	character	code	Hospital performing the procedure
c_tilopr	character	code	Supplementary code
c_oafd	character	code	Procedurende afdeling (undersøgelse & behandling)
leverancedato	date	date
version	character	code	Version
v_ominut	numeric	value	Procedureminut (undersøgelse & behandling)
v_otime	numeric	value	Proceduretime (undersøgelse & behandling)

No published source gives a data type for 8 of these 12 columns, so the Type column is our own assumption. Check with sapply(class) on a row of your own data before relying on it, especially for code columns, which lose their leading zeros if they arrive as numbers.

Join key: recnum.

Joins to other registers:

• recnum joins to LPR_ADM (many-to-one).

Value sets for the coded columns (2)

Code system	Values
sks	Not listed here - see DST’s classification
oprart	V Vigtigste operation i afsluttet kontakt, P Vigtigste operation i operativt indgreb, D Deloperation, + Tillaegskode

• sks: The codes are hierarchical, so a prefix match selects a whole branch. That also makes it easy to select more than you meant: check how many characters your intended group actually needs before filtering with starts_with().
• oprart: Counting rows in the procedure table counts add-on codes and sub-procedures as procedures. If you want one row per operation, filter to V or P first. All four codes run from 1996 with no breaks.

Where these values come from:

• sks: SKS browser (medinfo.dk).
• oprart: Kodeark for Landspatientregisteret, published on www.esundhed.dk.

Worth knowing:

• c_opr: Check that you actually have this column. DST documents it for every year 1999-2019, but a delivery can arrive with only recnum, d_odto and year, which leaves no way to tell one procedure from another. Without it the table is unusable, and no filtering recovers it.
• year: Not a DST variable. It is the partition the yearly deliveries were written into, so filtering on it stops the other years being read at all. Use it to limit how much is read, not to decide when something happened: for that, use the register’s own date column.

Check that you actually have c_opr before planning around this register. DST documents it for every year 1999-2019, but at least one delivery contains only recnum, d_odto and year, which cannot tell one procedure from another. Run colnames() first.

Combination across the full period

See the code example: combination across the full period

# Replace [projectnumber] with your own project number
# With fastreg: read_register("registername") instead of open_dataset("path/to/registername/")

# SKS from LPR2 (up to 2018)
surg_lpr2 <- read_register("lpr_sksopr") %>%
  rename_with(tolower) %>%
  filter(toupper(c_opr) %in% !!SKS_CODES) %>%   # !! sends the local R vector to DuckDB
  left_join(
    read_register("lpr_adm") %>%
      rename_with(tolower) %>%
      select(recnum, pnr, d_inddto),
    by = "recnum"
  ) %>%
  select(pnr, surgery_date = d_odto, surgery_code = c_opr) %>%
  collect()

# SKS from LPR3 (2019 and onwards) - join via dw_ek_kontakt
surg_lpr3 <- read_register("lpr_a_procregistrering") %>%
  rename_with(tolower) %>%
  filter(toupper(proc_kode) %in% !!SKS_CODES) %>%   # !! sends the local R vector to DuckDB
  left_join(
    read_register("lpr_a_kontakt") %>%
      rename_with(tolower) %>%
      select(dw_ek_kontakt, pnr),
    by = "dw_ek_kontakt"
  ) %>%
  mutate(surgery_date = as.Date(proc_starttidspunkt)) %>%
  select(pnr, surgery_date, surgery_code = proc_kode) %>%
  collect()

# Combined
surg_all <- bind_rows(surg_lpr2, surg_lpr3)

6. Other clinical registers

These registers are used less often than those above and are not column-verified here - the descriptions give an overview, but look up the exact variable names in DST’s overview of registers and variable lists and confirm against your own files with colnames().

MFR - Medical Birth Register

One row per birth, carrying both mother and child. The mother appears once per birth, so a mother of three has three rows; the register is keyed on the child through cpr_barn.

Two versions of the same MFR register exist

DST delivers MFR as one wide table, 1997-2018. Sundhedsdatastyrelsen publishes it split across MFR_MOR_* and MFR_BARN_* tables. Check which one your project has before writing code against column names.

gestationsalder_dage is in days, not weeks. paritet counts previous births, which is not the same as the number of children currently alive.

Column	Type	Role	Label	Years
alder_moder	numeric	value	ALDER_MODER
bmi_moder	numeric	value	BMI_MODER	2003 to 2018
cpr_barn	character	join key	CPR_BARN
cpr_moder	character	join key	CPR_MODER
flerfoldsfoedsel_beregnet	character	code	FLERFOLDSFOEDSEL_BEREGNET
gestationsalder_dage	numeric	value	GESTATIONSALDER_DAGE
hoejde_moder	numeric	value	HOEJDE_MODER	2003 to 2018
laengde_barn	character	code	LAENGDE_BARN
paritet	numeric	value	PARITET
rygerstatus_moder	character	code	RYGERSTATUS_MODER
vaegt_barn	numeric	value	VAEGT_BARN
vaegt_moder	numeric	value	VAEGT_MODER	2003 to 2018

All other columns (79)

Column	Type	Role	Label	Years
abdominalomfang	numeric	value	ABDOMINALOMFANG
abruptio	character	code	ABRUPTIO
afdeling	character	code	AFDELING
alderveddoed_dage_barn	numeric	value	ALDERVEDDOED_DAGE_BARN
alder_fader	numeric	value	ALDER_FADER
amnioinfusion	character	code	AMNIOINFUSION	1998 to 2018
amnitomi_under_foedsel_hsp	character	code	AMNITOMI_UNDER_FOEDSEL_HSP
andensutur	character	code	ANDENSUTUR
apgarscore_efter5minutter	numeric	value	APGARSCORE_EFTER5MINUTTER
barnslevendenr_flerfoldfoedsel	numeric	value	BARNSLEVENDENR_FLERFOLDFOEDSEL
barnsnummer_flerfoldsfoedsel	numeric	value	BARNSNUMMER_FLERFOLDSFOEDSEL
besoeghosjordemoder	character	code	BESOEGHOSJORDEMODER
besoeghoslaege	character	code	BESOEGHOSLAEGE
besoeghosspeciallaege	character	code	BESOEGHOSSPECIALLAEGE
bopaelskommune_moder	character	code	BOPAELSKOMMUNE_MODER
cpapbeh_neonatalafdeling	character	code	CPAPBEH_NEONATALAFDELING	2000 to 2018
cpr_fader	character	join key	CPR_FADER
disproportio	character	code	DISPROPORTIO
doedsdato_barn	date	date	DOEDSDATO_BARN
doedsdato_moder	date	date	DOEDSDATO_MODER
epiduralblokade	character	code	EPIDURALBLOKADE	2000 to 2018
episiotomi	character	code	EPISIOTOMI
fastsiddendemoderkage	character	code	FASTSIDDENDEMODERKAGE
flerfoldsgraviditet	character	code	FLERFOLDSGRAVIDITET
foedested	character	code	FOEDESTED
foedselsaar	character	code	FOEDSELSAAR
foedselsdato	date	date	FOEDSELSDATO
foedselsdiagnose_moder	character	code	FOEDSELSDIAGNOSE_MODER
foedselsloebenummer	numeric	value
foedselstime	character	code	FOEDSELSTIME
fosterpraesentation	character	code	FOSTERPRAESENTATION
hjemmebesoeg	character	code	HJEMMEBESOEG	2003 to 2018
hovedomfang	numeric	value	HOVEDOMFANG
intrauterin_asfyxi	character	code	INTRAUTERIN_ASFYXI
intrauterin_palpation	character	code	INTRAUTERIN_PALPATION
kejsersnit_modersoenske	character	code	KEJSERSNIT_MODERSOENSKE	2002 to 2018
koen_barn	character	code	KOEN_BARN
levende_eller_doedfoedt	character	code	LEVENDE_ELLER_DOEDFOEDT
markoer_accreta	character	code	MARKOER_ACCRETA
markoer_anaestesi_til_operation	character	code	MARKOER_ANAESTESI_TIL_OPERATION	2000 to 2018
markoer_andre_foedselskomplikati	character	code
markoer_b_misdannelse	character	code	MARKOER_B_MISDANNELSE
markoer_cardiomyopati	character	code	MARKOER_CARDIOMYOPATI
markoer_graviditetskomplikatio	character	code	MARKOER_GRAVIDITETSKOMPLIKATIO
markoer_haemoperitoneum	character	code	MARKOER_HAEMOPERITONEUM
markoer_hjemmefoedsel_beregnet	character	code	MARKOER_HJEMMEFOEDSEL_BEREGNET
markoer_igangsaettelse	character	code	MARKOER_IGANGSAETTELSE
markoer_infektioner	character	code	MARKOER_INFEKTIONER
markoer_kejsersnit	character	code	MARKOER_KEJSERSNIT
markoer_medicinske_sygdomme	character	code	MARKOER_MEDICINSKE_SYGDOMME
markoer_navlesnorsblod_analyse	character	code	MARKOER_NAVLESNORSBLOD_ANALYSE	2003 to 2018
markoer_perineal_bristning	character	code	MARKOER_PERINEAL_BRISTNING
markoer_post_partum_bloedning	character	code	MARKOER_POST_PARTUM_BLOEDNING
markoer_ruptur	character	code	MARKOER_RUPTUR
markoer_smertelindring	character	code	MARKOER_SMERTELINDRING	1999 to 2018
markoer_ultralyd	character	code	MARKOER_ULTRALYD	1999 to 2018
markoer_vestimulation	character	code	MARKOER_VESTIMULATION	1999 to 2018
markoer_ydre_vending	character	code	MARKOER_YDRE_VENDING
navlesnorsfremfald	character	code	NAVLESNORSFREMFALD
pk_mfr	character	code	PK_MFR
placentavaegt	numeric	value	PLACENTAVAEGT
polyhydramnios	character	code	POLYHYDRAMNIOS
pprom	character	code	PPROM
praevia	character	code	PRAEVIA
prom	character	code	PROM
respiratorbeh_neonatalafdeling	character	code	RESPIRATORBEH_NEONATALAFDELING	2000 to 2018
sengedage_beregnet_barn	numeric	value	SENGEDAGE_BEREGNET_BARN
sengedage_beregnet_moder	numeric	value	SENGEDAGE_BEREGNET_MODER
sengedage_neonatalafdeling_barn	numeric	value	SENGEDAGE_NEONATALAFDELING_BARN
sepsis_barn	character	code	SEPSIS_BARN
skalp_blodproeve	character	code	SKALP_BLODPROEVE	2000 to 2018
suturcollum	character	code	SUTURCOLLUM
sygehus	character	code	SYGEHUS
tang_forloesning	character	code	TANG_FORLOESNING
tegn_paa_asphyxi	character	code	TEGN_PAA_ASPHYXI
tidligerefoedsler_i_danmark	character	code	TIDLIGEREFOEDSLER_I_DANMARK
tidligerekejsersnit_i_danmark	character	code	TIDLIGEREKEJSERSNIT_I_DANMARK
tidligerespontaneaborter	character	code	TIDLIGERESPONTANEABORTER
vakuumekstraktion	character	code	VAKUUMEKSTRAKTION

No published source gives a data type for 91 of these 91 columns, so the Type column is our own assumption. Check with sapply(class) on a row of your own data before relying on it, especially for code columns, which lose their leading zeros if they arrive as numbers.

Join key: cpr_barn.

Joins to other registers:

• pnr joins to BEF (many-to-one).

Worth knowing:

• bmi_moder: Only from 2003, unlike most of the register, which starts in 1997.
• cpr_barn: The child’s CPR number, which is how the register joins to every other register about the child.
• cpr_moder: The mother’s CPR number. There is a row per child, so a mother of three appears three times.
• gestationsalder_dage: Gestational age in days, not weeks. Divide by 7 for the usual clinical scale.
• paritet: Parity. Counts previous births, so it is not the same as the number of children currently alive.

The Cancer Register

The gold standard for incident cancer diagnoses - more complete and precise for cancer than LPR diagnoses. Running since 1943, which makes it one of the longest continuous cancer registries anywhere. Use it for cancer as an outcome or an exclusion.

It is not a DST register. The Cancer Register is held by Sundhedsdatastyrelsen and does not appear in DST’s register list at all: a project reaches it through Forskerservice, on a separate application. The person key is k_cprnr, not pnr, so rename before joining.

One row per tumour, not per person. Somebody with two primary cancers has two rows, distinguished by k_tumornr. Deduplicating on the person alone collapses second primaries and undercounts incidence.

Column	Type	Role	Label	Years
c_icd10	character	code	ICD10 diagnose	1978 to 2024
c_morfo03	character	code	ICD3 - Histologi og morfologi	1978 to 2024
c_topo3	character	code	Tumorens lokalisation	1978 to 2024
d_diagnosedato	date	date	Diagnosedato
k_cprnr	character	join key	CPR-nummer
k_tumornr	character	code	Tumor løbenr.
v_diagnosealder	numeric	value	Patientens alder

All other columns (28)

Column	Type	Role	Label	Years
c_amtf07	character	code	Amtskode	1968 to 2006
c_behandling	character	code	Behandlingskode	1943 to 2003
c_diaggr	character	code	Diagnosegruppering ICD7
c_diaggr_kp	character	code	Diagnosegruppering KP	1978 to 2024
c_diaggr_nordcan	character	code	Diagnosegruppering NordCan
c_grad	character	code	Gradering	1943 to 2008
c_idc7	character	code	ICD7 diagnose	1943 to 1977
c_komf07	character	code	Patientens bopælskommune	1968 to 2006
c_kommune	character	code	Bopælskommune	1978 to 2024
c_lateral	character	code	Lateralitet
c_makrogrundlag	character	code	Makroskopisk grundlag
c_mikrogrundlag	character	code	Mikroskopisk grundlag
c_orggr	character	code	ICD7 diagnosegruppering
c_orggr_idc10	character	code	ICD10 diagnosegruppering	1978 to 2024
c_region	character	code	Patientens bopælsregion	2006 to 2024
c_sarc	character	code	Sarcom. eller ej	1943 to 2006
c_sex	character	code	Patientens køn
c_status	character	code	Patientens status
c_tnm_m	character	code	Angiver fjernmetastaser	2004 to 2024
c_tnm_n	character	code	Angiver lymfeknudemetastaser	2004 to 2024
c_tmn_t	character	code	Størrelse af tumor	2004 to 2024
c_udbred	character	code	Tumorens udbredelse	1943 to 2003
c_udbred_klassifikation	character	code	Anvendt udbredelsesklassifikation
c_aa	character	code	Ann Arbour klassifikation	2004 to 2024
d_fdsdato	date	date	Fødselsdato
d_statdato	date	date	Status dato
v_diagmd	numeric	value	Diagnosemåned
v_diagaar	numeric	value	Diagnoseår

• c_grad: Only filled in for urinary tract tumours.
• c_idc7: ICD-7, the coding used before ICD-10. A series running back before 1978 needs both this and c_icd10.
• c_udbred: Replaced by the TNM columns and Ann Arbor from 2004, when the register moved from paper coding to electronic reporting through LPR. Nothing maps one to the other, so a stage variable across 2004 is two different things.

Join key: k_cprnr.

Joins to other registers:

• k_cprnr joins to BEF (many-to-one).

Value sets for the coded columns (3)

Code system	Values
icd10	Not listed here - see DST’s classification
kom	Not listed here - see DST’s classification
reg	0 Uoplyst, 81 Nordjylland, 82 Midtjylland, 83 Syddanmark, 84 Hovedstaden, 85 Sjælland

• icd10: Do not strip a leading D from these codes. The habit comes from LPR, where the D is really there, and applying it here removes the first character of a real code: E119 becomes 119, which matches nothing and raises no error. The danger is worst where a code genuinely begins with D. ICD-10 chapter D covers in-situ and benign neoplasms, so D46 is myelodysplastic syndrome, a whole code. Strip its “prefix” and you get 46, which looks like a code and is not one.
• kom: These codes are valid from 1 January 2007. A study reaching further back needs the pre-reform classification, where the same number can mean a different municipality - confirmed for two reused codes against a current-only DST source: 707 is Norddjurs today, not its pre-2007 meaning, and likewise 849 is Jammerbugt. lookup: below covers only this post-2007 set (99 entries), not the full 278-code values_from file. Christiansø (411) is included in lookup: even though it is not a municipality (see description above): it is a real value a kom column can hold, and DST’s own current-only classification lists it as its own area code alongside the 98 municipalities. Excluding it would just move the “unhandled code” problem this fix is meant to solve onto that one value.
• reg: Do not confuse these with AMT, the pre-2007 counties, which has 16 codes in the ranges 11-14, 21-24, 31-37 and 88. Different geography, different era.

Where these values come from:

• icd10: WHO ICD-10 browser.
• kom: DST’s municipality classification (the code list as CSV).
• reg: DST’s regional classification.

Worth knowing:

• c_icd10: The modern diagnosis code. Older years also carry ICD-7 in c_idc7, so a series running back before ICD-10 has to use both.
• c_morfo03: Histology and morphology, ICD-O-3. Together with c_topo3 (site) this is what distinguishes tumour types; the ICD-10 code alone does not.
• d_diagnosedato: The diagnosis date, which is the incidence date for this register.
• k_cprnr: The person key. Named k_cprnr rather than pnr, so rename before joining to a DST register.
• k_tumornr: One person can appear several times: the register counts incident tumours, not people. Deduplicating on the person alone collapses second primaries.
• v_diagnosealder: Age at diagnosis, precomputed. Recompute from d_fdsdato if your index date differs.

Primary sector - sysi and sssy (Health Insurance Register)

Contacts and services in the primary sector (general practice, practising specialists, physiotherapy etc.). As with LPR, the register is split over time: sysi covers the older years, sssy the newer. Join via pnr. Use it for e.g. GP/specialist contacts, screening or vaccinations billed in primary care. Key fields are named the same in both, but the two registers do not carry an identical column set, so verify with colnames(). They overlap in 2005: both report that year, so reading both without picking a source double-counts it.

sysi (1990-2005):

Column	Type	Role	Label
pnr	character	join key	Personal identifier
ydernr	character	identifier	Provider number
speciale	character	code	Specialty, 6-digit
ydlant	integer	value	Number of services under the specialty
afrper	character	date	Settlement period
sikgrup	character	code	Insurance group
year	integer	date	Register year

All other columns (15)

Column	Type	Role	Label	Years
ydtyp	character	code	Provider type
ydltid	character	code	Service timing code
ydersamt	character	code	Provider’s county
bruhon	numeric	value	Gross fee to the provider
honuge	character	date	Fee week
barnmak	character	code	Child marker
pattyp	character	code	Patient type
praktyp	character	code	Practice type
sikrekom	character	code	Municipality of the insured
grdhon	character	code	GRUNDHONORAR	1997 to 2005
henvisni	character	code	Henvisningsydernummer
paragraf	character	code	PARAGRAFRELATION	1997 to 2005
praksiso	character	code	PRAKSISOMRÅDE	1997 to 2005
sikreamt	character	code	Sikredes amt
vagtomr	character	code	VAGTOMRÅDE	1997 to 2005

No published source gives a data type for 21 of these 22 columns, so the Type column is our own assumption. Check with sapply(class) on a row of your own data before relying on it, especially for code columns, which lose their leading zeros if they arrive as numbers.

Join key: pnr.

Joins to other registers:

• pnr joins to BEF (many-to-one).

Value sets for the coded columns (1)

Code system	Values
kom	Not listed here - see DST’s classification

• kom: These codes are valid from 1 January 2007. A study reaching further back needs the pre-reform classification, where the same number can mean a different municipality - confirmed for two reused codes against a current-only DST source: 707 is Norddjurs today, not its pre-2007 meaning, and likewise 849 is Jammerbugt. lookup: below covers only this post-2007 set (99 entries), not the full 278-code values_from file. Christiansø (411) is included in lookup: even though it is not a municipality (see description above): it is a real value a kom column can hold, and DST’s own current-only classification lists it as its own area code alongside the 98 municipalities. Excluding it would just move the “unhandled code” problem this fix is meant to solve onto that one value.

Where these values come from:

• kom: DST’s municipality classification (the code list as CSV).

Worth knowing:

• ydernr: A provider number, not a person. DST’s variable list does not say what unit it identifies or whether it is stable when a practice changes hands, so do not use it to follow an individual clinician over time without checking that first.
• speciale: The 6-digit specialty code is what distinguishes a GP contact from a specialist one. There is no separate contact-type column: the specialty is the classification.
• ydlant: One row can cover several services, so counting rows undercounts activity. Sum this column instead.
• afrper: DST labels this “Afregningsperiode”, a settlement period rather than a treatment date. Nothing in the variable list says how far settlement can lag the contact, so check the distribution against honuge before using it as a date.
• sikgrup: Group 1 patients need a referral from their GP to see a specialist, physiotherapist, chiropodist or psychologist, and pay nothing. Group 2 patients may go directly to any GP or specialist, but pay the difference between the fee and the regional subsidy themselves. The two groups therefore leave different traces for the same clinical need, so group membership is a confounder in any analysis of specialist use. Source: borger.dk, “Sygesikring og sikringsgrupper”.
• year: Not a DST variable. It is the partition the yearly deliveries were written into, so filtering on it stops the other years being read at all. Use it to limit how much is read, not to decide when something happened: for that, use the register’s own date column.

sssy (2005-2025):

Column	Type	Role	Label
pnr	character	join key	Personal identifier
ydernr	character	identifier	Provider number
speciale	character	code	Specialty, 6-digit
ydlant	numeric	value	Number of services under the specialty
afrper	character	date	Settlement period
sikgrup	character	code	Insurance group
year	integer	date	Register year

All other columns (18)

Column	Type	Role	Label	Years
ydtyp	character	code	Provider type
ydltid	character	code	Service timing code
ydersamt	character	code	Provider’s county
bruhon	numeric	value	Gross fee to the provider
honuge	character	date	Fee week
barnmak	character	code	Child marker
kontakt	numeric	value	Contact
patgrp	character	code	Patient group
koenimp	character	code	Sex, imputed values included
alderimp	numeric	value	Alder ultimo inkl. imputerede
behandlingsdato	date	date		2021 to 2025
cprtjek	character	code	CPR-tjek
cprtype	character	code	CPR-type
registreringstid	character	code		2021 to 2025
spec2	character	code	2-cifret speciale
spec80	character	code		2021 to 2025
statpop	character	code		2021 to 2025
version	numeric	date	Version pr. referencetidspunkt for Moduldata

• kontakt: Not in SYSI. Do not assume a comparable count of contacts before 2005.
• koenimp: Imputed where the source was missing, so it is not identical to koen in BEF. Prefer BEF when you need sex as a study variable.

No published source gives a data type for 6 of these 25 columns, so the Type column is our own assumption. Check with sapply(class) on a row of your own data before relying on it, especially for code columns, which lose their leading zeros if they arrive as numbers.

Join key: pnr.

Joins to other registers:

• pnr joins to BEF (many-to-one).

Value sets for the coded columns (1)

Code system	Values
koen	1 Mand, 2 Kvinde, 9 Uoplyst

• koen: DST’s classification KOEN_V1_1980 also defines 9 for not stated, which a delivery may not contain but a value set should. Sex is taken from the tenth digit of the CPR number: even is female, odd is male.

Where these values come from:

• koen: DST’s classification KOEN_V1_1980 (the code list as CSV).

Worth knowing:

• ydernr: A provider number, not a person. DST’s variable list does not say what unit it identifies or whether it is stable when a practice changes hands, so do not use it to follow an individual clinician over time without checking that first.
• speciale: The 6-digit specialty code is what distinguishes a GP contact from a specialist one. There is no separate contact-type column: the specialty is the classification.
• ydlant: One row can cover several services, so counting rows undercounts activity. Sum this column instead.
• afrper: DST labels this “Afregningsperiode”, a settlement period rather than a treatment date. Nothing in the variable list says how far settlement can lag the contact, so check the distribution against honuge before using it as a date.
• sikgrup: Group 1 patients need a referral from their GP to see a specialist, physiotherapist, chiropodist or psychologist, and pay nothing. Group 2 patients may go directly to any GP or specialist, but pay the difference between the fee and the regional subsidy themselves. The two groups therefore leave different traces for the same clinical need, so group membership is a confounder in any analysis of specialist use. Source: borger.dk, “Sygesikring og sikringsgrupper”.
• year: Not a DST variable. It is the partition the yearly deliveries were written into, so filtering on it stops the other years being read at all. Use it to limit how much is read, not to decide when something happened: for that, use the register’s own date column.

Laboratory results - Laboratoriedatabasens Forskertabel

Laboratory and blood test results: HbA1c, lipids, creatinine. A very large register, well over a billion rows, so filter with arrow or duckplyr before you collect() into R.

It is not a DST register. This is Laboratoriedatabasens Forskertabel (LAB_F), held by Sundhedsdatastyrelsen and reached through Forskerservice. It is not in DST’s register list, the table names are project-specific, and the columns are named in English: patient_cpr, analysiscode, samplingdate, value, unit. NPU is the coding system used in analysiscode, not a column name.

value is text, not a number. Alongside numeric results it holds a fixed set of words - POSITIV, NEGATIV, IKKE PÅVIST, INGEN VÆKST and the blood type codes. as.numeric() turns every one of them into NA without warning, which drops exactly the samples where something was found.

Coverage starts in 2008, not the 1990s, and complete coverage arrives per laboratory between 2010 and 2016. Only NPU- and DNK-coded results reach the table, about 95 per cent; results with a laboratory’s own local code are absent, as are patients who declined consent. See Laboratoriedatabasen →.

DARTER: a dedicated extraction guide for laboratorieproevesvar_ is in progress - see DARTER - Register paths and datastores.

Healthcare costs (under development)

Under development - confirm everything in your own delivery. This section describes cost sources at the register level. Exact table and column names and the available years change from year to year (the DRG rate system is updated annually) and are not verified here. Use it as a pointer, and clarify the specific files with your data manager.

To compute healthcare use/costs per person, they are typically assembled from several sources:

• Somatic hospital contacts: DRG-grouped rates from the Danish Health Data Authority (DRG = inpatient, DAG = outpatient). Rates express average operating expenses per DRG group and are computed annually - see SDS DRG rates.
• Psychiatric hospital contacts: are not billed by DRG. The main principle is a bed-day rate for inpatients and a visit rate for outpatients - keep somatic and psychiatric separate.
• Primary sector (general practice, specialists etc.): fees in sysi/sssy (see the section above).
• Medication: patient co-payment vs. reimbursement in LMDB (section 7).

Availability of cost data varies (some years/sources are missing, e.g. more recent DRG years). The pattern is inspired by the Plana-Ripoll group’s code on OSF, but the variable names there are from a 2022 delivery and should not be assumed current.

7. LMDB - Prescription Register

Owned by the Danish Health Data Authority as the Prescription Register: Lægemiddelstatistikregisteret →.

One row per dispensed prescription. Covers approximately 1994 onwards.

Column	Type	Role	Label	Years
pnr	character	join key	Personal identifier	1995-Q2 to 2025-Q2
eksd	date	date	Dispensing date	1995-Q2 to 2025-Q2
atc	character	code	ATC code, full 7 characters	1995-Q2 to 2025-Q2
atc1	character	code	ATC level 1 (1 character)	1995-Q2 to 2025-Q2
atc2	character	code	ATC level 2 (3 characters)	1995-Q2 to 2025-Q2
atc3	character	code	ATC level 3 (4 characters)	1995-Q2 to 2025-Q2
atc4	character	code	ATC level 4 (5 characters)	1995-Q2 to 2025-Q2
vnr	character	code	Item number (product key)	1995-Q2 to 2025-Q2
apk	numeric	value	Number of packages	1995-Q2 to 2025-Q2
year	integer	date	Dispensing year

All other columns (55)

Column	Type	Role	Label	Years
indo	character	code	Indication code	2004-Q2 to 2025-Q2
packsize	numeric	value	Package size	1995-Q2 to 2025-Q2
strnum	numeric	value	Strength, numeric	1995-Q2 to 2025-Q2
strunit	character	value	Unit for the numeric strength	1995-Q2 to 2025-Q2
aldr	numeric	value	Age at dispensing	1995 to 2025
abc	character	code	ABC-kode	2007 to 2025
aip	numeric	value	Registerpris (‘Apotekets indkøbspris’) (s)	1995 to 2025
aref	character	code	Andre refusionsordninger	1995 to 2025
aup	numeric	value	Registerpris (‘Apotekets udsalgspris’) (s)	1995 to 2025
bald	character	code		1995 to 2011
cprtjek	character	code	CPR-tjek	1995 to 2025
cprtype	character	code	CPR-type	1995 to 2025
cpr_kom	character	code	CPR-bopælskommune på ekspeditionsdatoen	2005 to 2025
cpr_reg	character	code	CPR-bopælsregion på ekspeditionsdatoen	2005 to 2025
dosform	character	code	Lægemiddelform (s)	1995 to 2025
doso	character	code	Doseringskode	2004 to 2025
edbl	character	code		2020 to 2025
ejs	character	code	Fravalg af substitution	1997 to 2025
eksp	numeric	value	Ekspeditionspris	1995 to 2025
ekst	character	code	Ekspeditionstype	1995 to 2025
etid	numeric	date	Ekspeditionstidspunkt	1997 to 2025
ibgp	numeric	value	Indberettet beregningsgrundlagspris	2000 to 2025
ibnr	character	code	Indberetternummer	1995 to 2025
itype	character	code	Indberettertype	1995 to 2025
kom	character	code	Kommunekode	1995 to 2025
korr	character	code	Korrektionskode	1995 to 2025
name	character	code	Lægemidlets navn (Store bogstaver) (s)	1995 to 2025
ovnr	character	code	Ordineret varenummer	1997 to 2025
packtext	character	code	PACKTEXT	1995 to 2025
patt	character	code	Patienttype	2000 to 2025
pksubgr	character	code	Pakningssubstitutionsgruppe	2007 to 2025
pnr12	character	join key	CPR-nummer	1995 to 2025
pprs	character	code	Udleveringsbestemmelse	1995 to 2025
ptp	character	code	Patientbetaling	1995 to 2025
ramt	character	code	Amtskommunekode	1995 to 2025
reca	character	code	Autorisationskode for receptudsteder	2005 to 2025
recu	character	code	Receptudsteder	1995 to 2025
rgl1	character	code	1. Kommuneregelnummer	1995 to 2025
rgl2	character	code	2. kommuneregelnummer	1995 to 2025
rgla	character	code	Amtskommunalt reglnummer	1995 to 2025
rimb	character	code	Tilskudskode	1995 to 2025
rinr	character	code	Reiterationsnummer	1995 to 2025
sektor	character	code	Sektor	1995 to 2025
streng	character	code	Styrke, klartekst (s)	1995 to 2025
takd	date	date	Takstdato	1995 to 2025
tard	date	date	Takseringsdato	2000 to 2025
tilpris	numeric	value	Tilskudspris pr. pakning	1995 to 2025
tsk1	character	code	1. Kommunale tilskud	1995 to 2025
tsk2	character	code	2. Kommunale tilskud	1995 to 2025
tsk3	character	code	Andre tilskud	1995 to 2025
tska	character	code	Amtskommunalt tilskud	1995 to 2025
udlv	character	code	Udleveringssted	1995 to 2025
voltypecode	character	code		1995 to 2025
voltypetxt	character	code		1995 to 2025
volume	character	code	Volume	1995 to 2025

• indo: Recorded only when the prescriber picks an indication from the drop-down. Typed as free text it is not carried over, so the column is often empty.

No published source gives a data type for 28 of these 65 columns, so the Type column is our own assumption. Check with sapply(class) on a row of your own data before relying on it, especially for code columns, which lose their leading zeros if they arrive as numbers.

Join key: pnr.

Joins to other registers:

• pnr joins to BEF (many-to-one).

Value sets for the coded columns (3)

Code system	Values
atc	Not listed here - see DST’s classification
kom	Not listed here - see DST’s classification
reg	0 Uoplyst, 81 Nordjylland, 82 Midtjylland, 83 Syddanmark, 84 Hovedstaden, 85 Sjælland

• atc: As a rule, filter on the full 7-character code rather than on the level columns: atc2 holds three characters, so a longer pattern matched against it can never match, and it returns nothing at all with no error. The level columns are well suited to grouping, and to filtering when every code you want is the same length as the column.
• kom: These codes are valid from 1 January 2007. A study reaching further back needs the pre-reform classification, where the same number can mean a different municipality - confirmed for two reused codes against a current-only DST source: 707 is Norddjurs today, not its pre-2007 meaning, and likewise 849 is Jammerbugt. lookup: below covers only this post-2007 set (99 entries), not the full 278-code values_from file. Christiansø (411) is included in lookup: even though it is not a municipality (see description above): it is a real value a kom column can hold, and DST’s own current-only classification lists it as its own area code alongside the 98 municipalities. Excluding it would just move the “unhandled code” problem this fix is meant to solve onto that one value.
• reg: Do not confuse these with AMT, the pre-2007 counties, which has 16 codes in the ranges 11-14, 21-24, 31-37 and 88. Different geography, different era.

Where these values come from:

• atc: WHO ATC/DDD Index.
• kom: DST’s municipality classification (the code list as CSV).
• reg: DST’s regional classification.

Worth knowing:

• pnr: DST’s variable list calls this column PNR12. Check whether your variable is named pnr or pnr12. Do not confuse this with the pnr documented on esundhed’s page for this register (DocumentationExtended?id=14): that pnr is the pharmacy or manufacturer’s production-unit number, length 10, not a person. Same name, unrelated variable, on the two pages that between them cover this register.
• eksd: The date the prescription was collected at the pharmacy. Not the date it was prescribed, and not evidence that the medicine was taken.
• vnr: The only reliable way to isolate one specific product. Two brands with the same active substance share an ATC code but have different item numbers.
• year: Not a DST variable. It comes from fastreg’s parquet conversion, which concatenates the yearly deliveries, so it exists in the data you read but not in DST’s own documentation of this register.

All confirmed columns in LMDB

pnr, eksd, ekst, atc, atc1, atc2, atc3, atc4, indo, vnr, apk, aldr, bald, eksp, korr, rinr, name, streng, packtext, volume, voltypecode, voltypetxt, dosform, strnum, strunit, packsize, cprtjek, cprtype, year, etid, ovnr, patt, doso, reca, abc

Filtering on atc2 fails silently. The level columns are shorter than the full code: atc2 holds three characters, so grepl("N02A", atc2) matches nothing and returns zero rows without an error. Filter on atc and keep the level columns for grouping. Details, and a check you can run on your own extract, are in Medication (ATC).

Data quality of indo and doso

indo (indication code) is recorded only when the doctor picks an indication from the drop-down menu in the electronic prescription; typed as free text, it is not carried over. It is therefore missing on about 12-18% of prescriptions (more before 1 October 2017) and is often nonspecific. The dosage field doso is effectively empty (recorded for ~0.06% of prescriptions). See Medication (ATC) for how to handle this.

8. Socioeconomic registers

All three registers are used for SEP extraction following SEPLINE guidelines (Hjorth et al. 2025). No single combined SEP variable is calculated - three separate dimensions.

UDDA - Education Register

One record per person per year - updated when the education level changes.

Column	Type	Role	Label	Years
pnr	character	join key	Personal identifier	1980 to 2025
hfaudd	character	code	Highest completed education	1980 to 2025
udd	character	code	Education code	1980 to 2025
hf_vfra	date	date	Date the education was completed	1980 to 2025
year	integer	date	Register year

All other columns (13)

Column	Type	Role	Label	Years
hf_kilde	character	code	Source of the education record	1980 to 2025
hfinstnr	character	code	Institution that awarded the education	1980 to 2025
almaudd	character	code	Highest completed general education	1980 to 2025
erhaudd	character	code	Highest completed vocational education	1980 to 2025
alm_vfra	date	date	Date the general education was obtained	1980 to 2025
erh_vfra	date	date	Date the vocational education was obtained	1980 to 2025
ig_vfra	date	date	Start date of the ongoing education	1980 to 2025
alminstnr	character	code	Institution, general education	1980 to 2025
erhinstnr	character	code	Institution, vocational education	1980 to 2025
iginstnr	character	code	Institution, ongoing education	1980 to 2025
cprtjek	character	code	CPR check	2005 to 2025
cprtype	character	code	CPR type	2005 to 2025
version	character	code	Module data version	2005 to 2025

• hfinstnr: The guide previously referred to this column as INSTNR. DST’s list has no INSTNR; the institution columns are HFINSTNR, ALMINSTNR, ERHINSTNR and IGINSTNR, one per kind of education.
• almaudd: The general-education track only. hfaudd is the highest completed education of any kind, so the two answer different questions and are not interchangeable.
• ig_vfra: Pairs with udd: this is when the ongoing education began. An education with a start and no completion is either still running or was interrupted, and the register does not distinguish the two.

No published source gives a data type for 17 of these 18 columns, so the Type column is our own assumption. Check with sapply(class) on a row of your own data before relying on it, especially for code columns, which lose their leading zeros if they arrive as numbers.

Join key: pnr.

Joins to other registers:

• pnr joins to BEF (many-to-one).

Value sets for the coded columns (1)

Code system	Values
hfaudd	Not listed here - see DST’s classification

• hfaudd: This is an identifier, not a scale. The level (short, medium, long) has to be looked up in a separate table, and cannot be read off the digits: 4112 is an electrician, and taking the first two digits as a level code makes it a long higher education. DST documents the ongoing-education codes separately as DISCED-15 UDD.

Where these values come from:

• hfaudd: DST’s DISCED-15 classification.

Worth knowing:

• hfaudd: A code for which education, not for its level. UDDA carries no level column at all, so the level has to come from a lookup table.
• udd: Not the same code system as hfaudd. Under DISCED-15, AUDD codes describe a completed education and UDD codes one that is ongoing or was interrupted, so a lookup table built for one will not fit the other.
• year: Not a DST variable. It is the partition the yearly deliveries were written into, so filtering on it stops the other years being read at all. Use it to limit how much is read, not to decide when something happened: for that, use the register’s own date column.

Importanthfaudd is not an education level - do not read the digits

This is the single easiest mistake to make with UDDA, and it produces wrong numbers without any error message. There are two different code systems, and they look confusingly alike:

	What it identifies	Example
Education code (hfaudd)	Which education, as a four-digit DISCED-15 code	4112 = electrician
Level code	How long/high the education is, two digits	30 = vocational, 70 = master’s

hfaudd is an identifier, not a scale. The first two digits are simply the start of a serial number, and there is no arithmetic relationship between the two systems. Taking substr(hfaudd, 1, 2) and treating it as the level is wrong, and it fails in both directions:

• hfaudd = 4112 is an electrician - vocational, level 30. Reading 41 and comparing it to 40 makes it a long higher education.
• hfaudd = 141 is genuinely level 40, short cycle higher education. Reading 14 matches no level at all, so a real education becomes missing data.

UDDA does not contain the level. DST’s education register holds only HFAUDD, HF_KILDE, HF_VFRA, HF_VTIL and INSTNR - there is no level column to read. The level has to come from a lookup table that maps each education code to its level. See Socioeconomic variables for how to do it.

How often the substr() shortcut is actually wrong

Tested against the full DISCED-15 list, the rule gets 58% of the 4,621 codes right, and 1,149 of them land in a confidently wrong category rather than in “unknown”. So it is not a rule that fails loudly on odd cases: it is right often enough to look like it works, and wrong often enough to move real numbers.

It looks plausible because the level codes (10, 15, 20, 30, 40, 70, 90) have the same shape as the first two digits of an education code. That resemblance is a coincidence of how the serial numbers were assigned, not a mapping.

AUDD or UDD - pick the right lookup. DISCED-15 has one code for an education that is ongoing or interrupted (UDD) and one for a completed education (AUDD), and DST publishes format tables for both. hfaudd is højest fuldførte AUDD, the highest completed education, so an audd lookup is the one that fits. A udd table will still join - it will just match some rows, miss others, and never tell you. Naming and paths are in Format tables.

FAIK - Family Income

Household-equivalised disposable income per year. Link: join BEF (pnr, familie_id, year) with FAIK (familie_id, year).

The structure can differ between years. On DARTER, FAIK from 2022 onward also carries pnr, and the family’s row is repeated once per family member instead of appearing once. Joining on familie_id alone then multiplies your rows silently. Check your own delivery before you rely on the join - how, and what to do about it, is in Socioeconomic variables.

Column	Type	Role	Label	Years
pnr	character	join key	Personal identifier
familie_id	character	join key	Household key
famaekvivadisp_13	numeric	value	Household-equivalised disposable income
year	integer	date	Register year
famdisponibel_13	numeric	value	Disposable income
famindkomstialt_13	numeric	value	Total income before taxes
famsociogrup_13	numeric	code	Socioeconomic group, 2013 definition	1993 to 2024
famtype	numeric	code	Family type

All other columns (79)

Column	Type	Role	Label	Years
famaekvivadisp	numeric	value	Equivalised disposable family income	1990 to 2012
famaekvivaindknetto	numeric	value	Equivalised total income including net interest	1990 to 2012
famaktieindk	numeric	value	Share income	1990 to 2009
famandenpension	numeric	value	Other pension payments to the family
famandoverforsel	numeric	value	Other transfers to the family	1990 to 2012
famantalfskattepligtige	numeric	value	Number of fully taxable adults in the family
famarbejdsloesp	numeric	value	Unemployment benefit and training allowance
famarbmabidrag	numeric	value	Labour market contributions
famboernetilskud	numeric	value	Child benefit and family allowances
famboligform	numeric	code	Housing tenure	2000 to 2024
famboligstoette	numeric	value	Housing benefit paid to the family
famboligtype	numeric	code	Dwelling type	2000 to 2024
fambruttoindk	numeric	value	Gross family income	1990 to 2009
famdagpenge_kontant_13	numeric	value	Total benefits and social assistance
famdisponibel	numeric	value	Disposable family income	1990 to 2012
famefterloen	numeric	value	Early retirement pay
famejdskat_ejerbolig	numeric	value	Property tax, homeowners	2010 to 2024
famejdskat_lejerbolig	numeric	value	Property tax, tenants	2010 to 2024
famejendomsvurdering	numeric	value	Cash property value
famerhvervsindk	numeric	value	Business income for the family	1990 to 2012
famerhvervsindk_13	numeric	value	Business income: wages and net profit
famfolkefortid_13	numeric	value	State and early retirement pension in the family
famformrest_ny05	numeric	value	Net residual wealth at year end
famformueaktiver	numeric	value	Total assets
famformueindk	numeric	value	Total capital income for the family	1990 to 2012
famformueindk_brutto	numeric	value	Capital income, gross
famfradragialt	numeric	value	Total calculated deductions	1990 to 2012
famfrynsegoder	numeric	value	Taxable value of fringe benefits	1993 to 2009
famgaeldialt	numeric	value	Liabilities
famgron_check	numeric	value	Green check	2010 to 2024
famhoejstudda	character	code	Highest education among the adults	2000 to 2024
famhonny	numeric	value	Fees liable to labour market contributions
famindkomstialt	numeric	value	Total family income before tax	1990 to 2012
famkontanthjaelp	numeric	value	Social assistance for the family	1990 to 2009
famkontanthjaelp_13	numeric	value	Social assistance in the family
famlejevaerdi	numeric	value	Imputed rental value of owner-occupied dwelling	1990 to 2012
famlejev_egen_bolig	numeric	value	Imputed rental value of owner-occupied dwelling
famloenmv	numeric	value	Total wage income in the family	1990 to 2012
famloenmv_13	numeric	value	Total wage income in the family
fammidlertidyd	numeric	value	Temporary transfer incomes	1990 to 2012
famoevrigformue	numeric	value	Other capital income for the family	1990 to 2009
famoevrigformue_13	numeric	value	Other capital incomes in the family
famoffpens_eftlon_13	numeric	value	Public pensions in the family
famoff_overforsel_13	numeric	value	Public transfers to the family
famorlovsydelse	numeric	value	Leave benefits paid to the family	1994 to 2009
famoverfoerindk	numeric	value	Total transfer incomes	1990 to 2012
famovergangyd	numeric	value	Total transitional allowance	1994 to 2006
famovrig_dagpenge_akas_13	numeric	value	Other benefits from unemployment funds
famovrig_kontanthjalp_13	numeric	value	Activation, unemployment and rehabilitation allowance
famovrig_overforsel_13	numeric	value	Other transfer incomes in the family
fampensionatp	numeric	value	ATP pension payments
fampensionialt	numeric	value	Total pensions for the family	1990 to 2012
fampensoffentlig	numeric	value	State and early retirement pension	1990 to 2009
fampenstjeneste	numeric	value	Civil servant pension
famprivat_pension_13	numeric	value	Private pensions in the family
famrenteindk	numeric	value	Total interest income from Denmark	1990 to 2009
famrenteindk_13	numeric	value	Total taxable interest income
famrenteudgifter	numeric	value	Total deductible interest expenses	1990 to 2012
famrenteudgifter_13	numeric	value	Interest expenses
famrestbistandsyd	numeric	value	Other benefits from municipalities and unemployment funds	1990 to 2009
famrestindk	numeric	value	Miscellaneous unclassified income	1990 to 2012
famrestindk_13	numeric	value	Other personal income in the family
famsamletindk	numeric	value	Total income for the family	1990 to 2009
famskatfriyd	numeric	value	Tax-free incomes in the family	1990 to 2012
famskatmvialt	numeric	value	Tax and labour market contributions paid	1990 to 2012
famskatmvialt_13	numeric	value	Tax, labour market contributions and special pension
famskatpligtindk	numeric	value	Taxable income for the family
famskattot	numeric	value	Total income tax paid	1990 to 2009
famskattot_13	numeric	value	Total personal final tax
famsociogrup	numeric	code	Socioeconomic group	1994 to 2009
famsu	numeric	value	State education grants (SU)
famsumindknettorent	numeric	value	Total income including net interest	1990 to 2012
famsyg_barsel_13	numeric	value	Sickness and maternity benefits
famtransportfradrag	numeric	value	Total commuting deduction	1990 to 2009
famunderhbidrag	numeric	value	Maintenance payments made by the family
famvirkordind	numeric	value	Amounts placed in the business tax schemes
famvirkoverskud	numeric	value	Net profit from self-employment	1990 to 2009
famvirkoverskud_13	numeric	value	Total profit from self-employment
version	numeric	code	Module data version

• famaekvivadisp: The same weighting scheme as FAMAEKVIVADISP_13 below, just algebraically rearranged on the source page: the first adult over 14 counts as 1.0, each further person over 14 as 0.5, and each child under 15 as 0.3. All persons belonging to the same family on 31 December of the income year (same E-familienummer), including resident children under 25, are assigned the family’s equivalised income. FAMDISPONIBEL is the same money before that division.

No published source gives a data type for 67 of these 87 columns, so the Type column is our own assumption. Check with sapply(class) on a row of your own data before relying on it, especially for code columns, which lose their leading zeros if they arrive as numbers.

Join key: familie_id.

Joins to other registers:

• familie_id joins to BEF (one-to-many).

Value sets for the coded columns (3)

Code system	Values
famboligform	1 Ejerbolig, 2 Lejebolig, 9 Uoplyst
socio13	11 Selvstændige, 110 Selvstændige, 111 Selvstændige erhvervsdrivende med 10 eller flere ansatte, 112 Selvstændige erhvervsdrivende med 5-9 ansatte, 113 Selvstændige erhvervsdrivende med 1-4 ansatte, 114 Selvstændige erhvervsdrivende uden ansatte, 12 Medarbejdende ægtefælle, 120 Medarbejdende ægtefælle, 13 Lønmodtagere, 131 Lønmodtagere med ledelsesarbejde, 132 Lønmodtager i arbejde der forudsætter færdigheder på højeste niveau, 133 Lønmodtager i arbejde der forudsætter færdigheder på mellemniveau, 134 Lønmodtager i arbejde der forudsætter færdigheder på grundniveau, 135 Andre lønmodtagere, 139 Lønmodtager uden nærmere angivelse, 21 Arbejdsløs mindst halvdelen af året, 210 Arbejdsløse mindst halvdelen af året, 22 Sygedagpenge, orlov mv., 220 Modtager af sygedagpenge, uddannelsesgodtgørelse, orlovsydelser mv., 31 Uddannelsessøgende, 310 Under uddannelse, inkl. skoleelever på min. 15 år, 32 Pensionist/efterløn, 321 Førtidspensionister, 322 Folkepensionister, 323 Efterlønsmodtagere mv., 33 Kontanthjælp, 330 Kontanthjælpsmodtagere, 41 Andre, 410 Andre, 42 Børn, 420 Børn under 15 år, ultimo året
famtype	0 Ingen oplysninger, 1 Ægtepar forskelligt køn, 2 Registreret partnerskab, 3 Samlevende par, 4 Samboende par, 6 Enlig mænd (herunder også ikke hjemmeboende børn), 7 Enlig kvinder (herunder også ikke hjemmeboende børn), 8 Ægtepar samme køn

• socio13: Two codes are easy to misread. 410 is Andre (other), not unemployed: the unemployed are 210. And 420 is children under 15, a known category rather than a missing value, so it appearing in an adult cohort means the index date is wrong rather than the data being incomplete.
• famtype: There is no code 5. The sequence runs 0 to 4 and then 6 to 8, which is easy to read as a missing value rather than as a code that never existed.

Where these values come from:

• famboligform: DST TIMES documentation: famboligform.
• socio13: DST’s SOCIO classification (the code list as CSV).
• famtype: DST high-quality variable documentation: famtype.

How it is computed:

famaekvivadisp_13

FAMAEKVIVADISP_13 = FAMDISPONIBEL_13 / (1 + (0.5 * (number of people over 14 in the family - 1)) + (0.3 * number of people under 15 in the family))

famaekvivadisp

FAMAEKVIVADISP = FAMDISPONIBEL / (0.5 + (0.5 * number of people over 14 in the family) + (0.3 * number of people under 15 in the family))

famaekvivaindknetto

FAMAEKVIVAINDKNETTO = FAMSUMINDKNETTORENT / (0.5 + (0.5 * number of people over 14 in the family) + (0.3 * number of people under 15 in the family))

• FAMANDOVERFORSEL = FAMOEVRIG_OVERFORSEL_13 + FAMGRON_CHECK (from 2010) famdagpenge_kontant_13

FAMDAGPENGE_KONTANT_13 = FAMARBLHUMV + FAMOVRIG_DAGPENGE_AKAS_13 + FAMKONTANTHJ_13 + FAMOVRIG_KONTANTHJALP_13 + FAMSYG_BARSEL_13

• FAMDISPONIBEL = FAMSUMINDKNETTORENT - FAMSKATMVIALT - FAMUNDERHBIDRAG famdisponibel_13

FAMDISPONIBEL_13 = FAMINDKOMSTIALT_13 + FAMLEJEV_EGEN_BOLIG - FAMRENTEUDGIFTER_13 - FAMSKATMVIALT_13 - FAMUNDERHBIDRAG

• FAMERHVERVSINDK = FAMLOENMV + FAMVIRKOVERSKUD
• FAMFORMUEINDK = FAMRENTEINDK + FAMOEVRIGFORMUE + FAMLEJEVAERDI famfradragialt

FAMFRADRAGIALT = FAMBRUTTOINDK - FAMSKATPLIGTIGINDK - FAMARBMABIDRAG (from 1994)

famindkomstialt_13

FAMINDKOMSTIALT_13 = FAMERHVERVSINDK_13 + FAMOFF_OVERFORSEL_13 + FAMPRIVAT_PENSION_13 + FAMRENTEINDK_13 + FAMOEVRIGFORMUE_13 + FAMRESTINDK_13

fammidlertidyd

FAMMIDLERTIDYD = FAMKONTANTHJAELP + FAMARBEJDSLOESP + FAMORLOVSYDELSE + FAMRESTBISTANDYD

fampensionialt

FAMPENSIONIALT = FAMPENSOFFENTLIG + FAMPENSIONATP + FAMEFTERLOEN + FAMOVERGANGYD (1994-2006) + FAMPENSTJENESTE (from 1991) + FAMANDENPENSION

famrestindk

FAMRESTINDK = FAMINDKOMSTIALT - FAMERHVERVSINDK - FAMOVERFOERINDK - FAMFORMUEINDK

• FAMSKATMVIALT = FAMSKATTOT + FAMARBMABIDRAG famsumindknettorent

FAMSUMINDKNETTORENT = FAMINDKOMSTIALT + FAMLEJEVAERDI - FAMRENTEUDGIFTER

Worth knowing:

• pnr: Not in DST’s variable list for FAIK, which documents the register as keyed on the household. Where it is present, the household’s row is repeated once per family member, so joining on familie_id alone multiplies rows.
• familie_id: There is no person identifier here. Fetch familie_id from BEF for the relevant year, then join on it.
• famaekvivadisp_13: This is the family’s disposable income divided by an equivalence factor, so that families of different sizes can be compared. The factor counts the first adult as 1.0, each further person over 14 as 0.5, and each child under 15 as 0.3: a couple with two young children comes to 1 + 0.5 + 0.6 = 2.1, so a family income of 420,000 is recorded as 200,000 for each of the four. FAMDISPONIBEL_13 is the same money before that division, and FAMAEKVIVAINDKNETTO is a different income concept (total income with net interest), so check which one your analysis plan means. The _13 suffix marks the definition that replaced the older variables from 2013: DST states that FAMDISPONIBEL was discontinued in 2013 and replaced by FAMDISPONIBEL_13. A study spanning 2013 has to know which side of that change it is on. DST’s year ranges settle the relationship with the sister column FAMAEKVIVADISP, which has no _13: the old one runs 1990-2012, this one runs 1987-2024. The _13 definition was applied backwards as well as forwards, so it is not a newer variable covering later years only. A study spanning the 2013 changeover should use _13 for the whole period rather than splicing the two, which would mix two definitions in one series.
• year: Not a DST variable. It comes from fastreg’s parquet conversion, which concatenates the yearly deliveries, so it exists in the data you read but not in DST’s own documentation of this register.

Income quintiles are calculated as 3-year averages compared against Q20/Q40/Q60/Q80 cut-points from the full BEF population stratified by sex × 5-year age group × reference year.

AKM - Labour Classification Module

Labour market status per person per year.

Column	Type	Role	Label	Years
pnr	character	join key	Personal identifier
socio13	integer	code	Socioeconomic classification, 2013 version	1991 to 2024
socio02	integer	code	Socioeconomic classification, 2002 version	2002 to 2013
socio	integer	code	Socioeconomic classification, 1994-2001 version	1994 to 2001
socio_gl	integer	code	Socioeconomic classification, 1976-1990 version	1976 to 1990
year	integer	date	Register year

All other columns (41)

Column	Type	Role	Label	Years
beskst13	integer	code	Main source of income	1991 to 2024
disco08_alle_indk_13	character	code	Occupation code, DISCO-08	2010 to 2024
nace_db07_13	character	code	Industry, DB07	2007 to 2024
alder_ult_ink	integer	value	Age at 31 December	1991 to 2024
ant_ansat_arbsted	numeric	value	Antal ansatte for væsentligste arbejdssted ( optælling af ansatte sker for hver arbejdessted under SENR (AKM)	2006 to 2013
ant_ansat_arbsted_13	numeric	value	Antal ansatte for væsentligste arbejdssted ( optælling af ansatte sker for hver arbejdessted under SENR (AKM)	2010 to 2024
ant_ansat_senr	numeric	value	Antal ansatte for væsentligste beskæftigelse (AKM)	1976 to 2005
atpsum2	numeric	value	Beskæftigelsesmål baseret på ATP- indbetalinger (AKM)	1976 to 2010
beskst	numeric	code	Beskæftigelsesstatus 1980 til 2001 (Indkomst/AKM)	1976 to 2001
beskst02	numeric	code	Beskæftigelsesstatus fra 2002 (Indkomst/AKM)	2002 to 2013
branche_77	character	code	Dansk branchekode 1977 til 1993 (AKM)	1980 to 1999
brchi	character	code	Branchekode for indehaver	1976 to 2002
brchl	character	code	Branchekode for lønmodtager	1976 to 1999
cprtjek	character	code	CPR-tjek	1991 to 2024
cprtype	character	code	CPR-type	1991 to 2024
disco08_alle_indk	character	code	Fagklassifikation for beskæftigelsesforhold, fra 2010 (AKM)	2010 to 2013
disco08_loen_indk	character	code	Fagkode for væsentligste lønmodtagerbeskæftigelse i året. (AKM)	2010 to 2024
disco08_sel_indk	character	code	Fagkode for arbejde i selvstændig virksomhed (AKM)	2010 to 2024
discoalle_indk	character	code	Fagklassifikation for beskæftigelsesforhold, fra 1991 TIL 2009 (AKM)	1993 to 2009
discoloen_indk	character	code	Fagkode for væsentligste lønmodtagerbeskæftigelse i året. (AKM)	1991 to 2009
discosel_indk	character	code	Fagkode for arbejde i selvstændig virksomhed (AKM)	1991 to 2009
discotyp	character	code	Kilde til lønmodtager DISCO-koden (AKM)	1991 to 2024
disco_alle_indk_13	character	code	Fagklassifikation for beskæftigelsesforhold, fra 1991 TIL 2009 (AKM)	1991 to 2009
funk_timeant	numeric	value	Personens samlede antal arbejdstimer i året.	2008 to 2024
nace	character	code	Branche for væsentligste beskæftigelse, fra 1992 til 2007 (AKM)	1992 to 2007
nacea	character	code	Branchegruppering for arbejdssted (1993 til 2007) (AKM)	1992 to 2007
nacea_db07	character	code	Branchegruppering for arbejdssted ( fra 2007) (AKM)	2007 to 2024
nacei	character	code	Branchekode for selvstændige og medarbejdende ægtefællers virksomhed(1993 til 2007) (AKM)	1993 to 2007
nacei_db07	character	code	Branchegruppering for indehaver (Selvstændig eller medhjælpende ægtefælle) (fra 2007) (AKM)	2007 to 2024
nace_13	character	code	Branche for væsentligste beskæftigelse, fra 1993 til 2007 (AKM)	1993 to 2007
nace_db07	character	code	Branche for væsentligste beskæftigelse, fra 2008-2013 (AKM)	2007 to 2013
nystgr	character	code	Stillingsgruppering 1980 til 1995 (AKM)	1980 to 1999
omfang	character	code	Omfang af skattepligt	1991 to 2024
senr	character	code	SE-nummer	1985 to 2002
senri	character	code	SE-nummer for den virksomhed som personen ejer	1985 to 2002
senrl	character	code	SE-nummer for lønmodtager	1985 to 1998
typ	character	code	Stillingstype	1980 to 1999
version	character	code	Moduldata version	1991 to 2024
virkf	numeric	value	Virksomhedskode angiver hvilken form for ejerskab der er på den arbejdsplads personen får størst erhrvsindkomst fra (AKM)	2001 to 2013
virkfa	numeric	value	Virksomhedskode angiver hvilken form for ejerskab der er på den arbejdsplads personen får størstklønindkomst/arbejdstimer (AKM)	2001 to 2024
virkf_13	numeric	value	Virksomhedskode angiver hvilken form for ejerskab der er på den arbejdsplads personen får størst erhrvsindkomst fra (AKM)	2001 to 2024

• beskst13: A different question from socio13: where the money came from, rather than what the person’s labour market position was. A source does now exist (https://www.dst.dk/da/TilSalg/data-til-forskning/generelt-om-data/ dokumentation-af-data/hoejkvalitetsvariable/ Personers-tilknytning-til-arbejdsmarkedet-set-over-hele-aaret–AKM-/BESKST13, checked 2026-09-09), a large classification with its own PDF supplement for edge cases, linked rather than transcribed. beskst13 differs from beskst02 in named ways (net rather than gross income for the self-employed, price-adjusted thresholds), which DST calls a difference rather than a break: see beskst02’s reader_note for the sharper break, between beskst02 and the older beskst.
• disco08_alle_indk_13: Occupation, not socioeconomic position. DISCO-08 only starts in 2010; 1991-2009 uses the older disco_alle_indk_13 with a different code set, so an occupation series across 2010 is not continuous.
• nace_db07_13: Industry classification from 2007. The pre-2007 series uses nace_13, which is a different classification rather than a renamed one.
• alder_ult_ink: Age at the end of the year, not at your index date. Recompute from a birth date if the exact age matters.
• beskst: Not comparable with beskst02, in DST’s own words, not just a difference of degree. Source: https://www.dst.dk/da/TilSalg/data-til-forskning/ generelt-om-data/dokumentation-af-data/hoejkvalitetsvariable/ Personers-tilknytning-til-arbejdsmarkedet-set-over-hele-aaret–AKM-/BESKST02 (checked 2026-09-09). People with an unemployment share over 50 percent, and people on efterløn (early retirement pay), get their own codes under beskst02 that beskst does not have, and the rules for who counts as self-employed also changed. A study spanning 2001-2002 should not treat this as a relabelled continuation of the same variable.
• beskst02: Not comparable with beskst (see beskst’s own reader_note): DST states this directly, not just a difference of degree. Comparable with beskst13 with named exceptions (net vs gross self-employment income, price adjustment); see beskst13’s reader_note.

No published source gives a data type for 46 of these 47 columns, so the Type column is our own assumption. Check with sapply(class) on a row of your own data before relying on it, especially for code columns, which lose their leading zeros if they arrive as numbers.

Join key: pnr.

Joins to other registers:

• pnr joins to BEF (many-to-one).

Value sets for the coded columns (11)

Code system	Values
socio13	11 Selvstændige, 110 Selvstændige, 111 Selvstændige erhvervsdrivende med 10 eller flere ansatte, 112 Selvstændige erhvervsdrivende med 5-9 ansatte, 113 Selvstændige erhvervsdrivende med 1-4 ansatte, 114 Selvstændige erhvervsdrivende uden ansatte, 12 Medarbejdende ægtefælle, 120 Medarbejdende ægtefælle, 13 Lønmodtagere, 131 Lønmodtagere med ledelsesarbejde, 132 Lønmodtager i arbejde der forudsætter færdigheder på højeste niveau, 133 Lønmodtager i arbejde der forudsætter færdigheder på mellemniveau, 134 Lønmodtager i arbejde der forudsætter færdigheder på grundniveau, 135 Andre lønmodtagere, 139 Lønmodtager uden nærmere angivelse, 21 Arbejdsløs mindst halvdelen af året, 210 Arbejdsløse mindst halvdelen af året, 22 Sygedagpenge, orlov mv., 220 Modtager af sygedagpenge, uddannelsesgodtgørelse, orlovsydelser mv., 31 Uddannelsessøgende, 310 Under uddannelse, inkl. skoleelever på min. 15 år, 32 Pensionist/efterløn, 321 Førtidspensionister, 322 Folkepensionister, 323 Efterlønsmodtagere mv., 33 Kontanthjælp, 330 Kontanthjælpsmodtagere, 41 Andre, 410 Andre, 42 Børn, 420 Børn under 15 år, ultimo året
beskst02	01 Selvstændig, 02 Medarbejdende ægtefælle, 03 Lønmodtager og ejer af virksomhed, 04 Lønmodtager, 05 Lønmodtager med understøttelse, 06 Pensionist og ejer af virksomhed, 07 Pensionist, 08 Øvrige, 09 Efterlønsmodtager, 10 Arbejdsløs mindst halvdelen af året (nettoledighed), 11 Modtager af dagpenge (aktivering og lign., sygdom, barsel og orlov), 12 Kontanthjælpsmodtager, 99 Ikke i AKM
disco08	Not listed here - see DST’s classification
nace_db07	Not listed here - see DST’s classification
beskst	01 Selvstændig, 02 Medarbejd., 03 Lønmodt. m. virksom., 04 Lønmodtager, 05 Lønmodtager med understøt., 06 Pensionist m. virksom., 07 Pensionist, 08 Øvrige, 99 Ikke i AKM
branche_77	Not listed here - see DST’s classification
disco_old	Not listed here - see DST’s classification
discotyp	0 Uoplyst, 1 Stat, 2 Kommune, 3 A-kasse, 4 Privat virksomhed, mindst 10 ansatte, 5 CPR (1991-1999), 6 Foregående år, samme arbejdsgiver og arbejdssted (2004-), 7 Uddannelse igangværende, 8 A-kasse og branchekode, 9 Uddannelse afsluttet + branche, 10 Danmarks Statistiks e-indkomstregister
nace_old	Not listed here - see DST’s classification
nystgr	Not listed here - see DST’s classification
omfang	0 Børn under 15 år og uden indkomst og formue (1998-2001), 1 Fuldt skattepligtig og har bopæl både primo og ultimo året, 2 Har ikke boet i Danmark hele året, eller er død i løbet af året, 3 Findes i oplysningsseddelregisteret men ikke i slutligningsregisteret (fra 2002), 4 Personer under 15 år uden indkomst, samt personer der ikke er skattepligtige, 5 Er død (bobehandlingskode) og ingen indkomst i slutligningsregisteret, 6 Indkomst usandsynlig for mindst en person i familien, kan ikke rettes konsistent

• socio13: Two codes are easy to misread. 410 is Andre (other), not unemployed: the unemployed are 210. And 420 is children under 15, a known category rather than a missing value, so it appearing in an adult cohort means the index date is wrong rather than the data being incomplete.
• beskst02: Not comparable with beskst (see beskst’s own reader_note): DST states this directly, not just a difference of degree. Comparable with beskst13 with named exceptions: self-employment income is measured net of capital income in beskst13 rather than gross, and beskst13’s amount thresholds are price-adjusted where beskst02’s are not. A study spanning both should read DST’s own change note before treating them as one continuous series.
• disco08: The AKM columns do not draw purely from this file. Checked against disco08_alle_indk’s own value list on DST’s high-quality variable page: only 564 of its 790 distinct codes appear in this CSV’s 564 level-5 codes. disco_typ (see discotyp.yaml) explains why: the occupation code for a given person-year can come from several source registers of varying vintage and quality, including imputation, so a study will see codes this file’s current edition does not carry. Do not treat a code missing from this CSV as an error in the data before checking discotyp.
• nace_db07: Two things to check before using this as a lookup. First, format: this file writes the detailed code with dots (01.11.00); nace_db07 and nace_db07_13 hold it without them (011100). Second, coverage: checked against nace_db07’s own value list on DST’s high-quality variable page, only 490 of its 743 distinct codes match this file’s 738 level-2 codes even after stripping dots. As with DISCO-08 (see disco08.yaml), the industry code for a given person-year can be sourced from more than one register, so a code absent from this CSV is not necessarily an error.
• beskst: Not comparable with beskst02, in DST’s own words, not just a difference of degree: people with an unemployment share over 50 percent, and people on efterløn (early retirement pay), get their own codes under beskst02 that this code list does not have, and the rules for who counts as self-employed also changed. See beskst02’s reader_note.
• branche_77: Superseded by NACE (see nace), which the AKM register also carries for overlapping years. A study reaching back before 1992 needs this code list; one starting later should prefer NACE or DB07 rather than reading a code DST had already retired.
• disco_old: Do not read a code here against DISCO-08’s published list (disco08.yaml): the two classifications share the general numeric shape but DST revised the detail at the 2010 switch, and a code that means one occupation in DISCO-08 is not guaranteed to mean the same thing, or to exist at all, in the pre-2010 scheme.
• discotyp: A different question from the DISCO occupation code itself: this says where that code came from, not what the job was. DST’s own page states that DISCO code quality is best when discotyp is 1, 2, 4 or 10, and that the remaining codes mean the DISCO code was set by imputation.
• nace_old: Do not read a code here against DB07/NACE Rev.2’s published list (nace_db07.yaml): the 2007-2008 revision changed the code structure, and a code that means one industry under DB07 is not guaranteed to mean the same thing, or to exist at all, under this older scheme.
• nystgr: Retired in 1995 and replaced by DISCO. A code’s meaning can shift within the window a study covers: DST’s own page notes that code 5100 changed meaning in 1992 (the group it now names used code 5500 before that year), so a series spanning 1992 needs the year-by-year note on the source page, not just the current label.
• omfang: Code 2 changed meaning at the 2002 boundary: before 2002 (and again 2002-2009) it also absorbed children under 15 with no income or wealth, a group that codes 0 and 4 cover separately in other years. Code 5 (died, not settled) only exists from 2002; before that, a death with no settled income was coded 2 instead. DST’s own publications and statistikbanktabeller are restricted to omfang=1: an analysis that does not filter on it silently includes partial-year residents, the newly dead, and unsettled records alongside ordinary full-year taxpayers.

Where these values come from:

• socio13: DST’s SOCIO classification (the code list as CSV).
• beskst02: DST high-quality variable documentation: BESKST02.
• disco08: DST’s DISCO-08 classification (the code list as CSV).
• nace_db07: DST’s DB07 classification (the code list as CSV).
• beskst: DST high-quality variable documentation: BESKST.
• branche_77: DST high-quality variable documentation: BRANCHE-77.
• disco_old: DST high-quality variable documentation: DISCOALLE_INDK.
• discotyp: DST high-quality variable documentation: DISCOTYP.
• nace_old: DST high-quality variable documentation: NACE.
• nystgr: DST high-quality variable documentation: NYSTGR.
• omfang: DST high-quality variable documentation: OMFANG.

Worth knowing:

• socio13: This is the version to use. It runs from 1991, so it reaches back further than its name suggests and covers the periods of the three older versions as well.
• socio02: Uses socio13’s lookup, but the two do not fully agree. Checked 2026-09-09 against DST’s own value tables (https://www.dst.dk/da/TilSalg/ data-til-forskning/generelt-om-data/dokumentation-af-data/hoejkvalitetsvariable/ Personers-tilknytning-til-arbejdsmarkedet-set-over-hele-aaret–AKM-/SOCIO02): code 111 here is “10 or more employees”, the same boundary socio13 uses, so socio02 and socio13 read together safely. socio (the 1994-2001 predecessor) is the one that disagrees: see its own reader_note.
• socio: Uses socio13’s lookup, but the two do not agree on code 111. Checked 2026-09-09 against DST’s own value tables (https://www.dst.dk/da/TilSalg/ data-til-forskning/generelt-om-data/dokumentation-af-data/hoejkvalitetsvariable/ Personers-tilknytning-til-arbejdsmarkedet-set-over-hele-aaret–AKM-/SOCIO): here, 111 is “selvstaendig med 50 eller flere ansatte” (self-employed, 50+ employees). In socio02 and socio13, the same code 111 is “10 eller flere ansatte” (10+ employees). The size boundary moved and the code did not change, so a self-employed person with, say, 20 employees is 111 in one era and a different code in the other, with nothing in the data to flag the switch.
• year: Not a DST variable. It comes from fastreg’s parquet conversion, which concatenates the yearly deliveries, so it exists in the data you read but not in DST’s own documentation of this register.

SEPLINE categorisation of socio13:

• Employed: 110–114, 120, 131–135, 139
• Student: 310
• Unemployed: 210, 410
• Outside labour market: 220, 321, 330
• Retired: 322, 323
• Unknown: 0, 420 or missing

9. Project-specific registers

Many projects have access to registers beyond the standard list above - e.g. quality registers from clinical databases or pre-computed classification files.

These are project-specific and not available in all projects on DST.

Examples are private hospitals (priv_adm, priv_diag, priv_skspor - structured in parallel with LPR) and various clinical quality registers. Availability varies from project to project.

Working on DARTER / project 708421?

The project uses among others DBSO (the Danish Obesity Treatment Database) and OSDC (Open Source Diabetes Classifier).

• All applied-for variables and registers in a searchable table: steno-aarhus.github.io/darter-project →
• Confirmed paths and column names: DARTER - Register paths and datastores →

Data resource profiles and reporting

Data resource profiles are the papers you cite when you describe a register in your methods section: they document the register’s content, coverage and validity. Cite the profile for each register you use.

Register	Data resource profile
The Danish health-care system and epidemiological research (overview)	Schmidt et al. 2019, Clin Epidemiol - doi:10.2147/CLEP.S179083
CPR (Civil Registration System)	Schmidt, Pedersen & Sørensen 2014, Eur J Epidemiol - doi:10.1007/s10654-014-9930-3
LPR (National Patient Registry)	Schmidt et al. 2015, Clin Epidemiol - doi:10.2147/CLEP.S91125
LMDB (Prescription Registry)	Pottegård et al. 2017, Int J Epidemiol - doi:10.1093/ije/dyw213
Cause of Death Register	Helweg-Larsen 2011, Scand J Public Health - doi:10.1177/1403494811399958

Reporting: report observational studies following STROBE. For register-based / routinely-collected-data studies, RECORD extends STROBE, and RECORD-PE covers pharmacoepidemiology specifically - see RECORD-PE (EQUATOR Network).