This page is part of the MII Implementation Guide Seltene Erkrankungen (v2027.0.0-ballot: 2027 Ballot 1) based on FHIR (HL7® FHIR® Standard) R4. This is the current published version. For a full list of available versions, see the Directory of published versions

Newborn Screening

Newborn screening is where a rare disease is most often found first. In Germany it is the one population-wide programme that systematically detects rare conditions before symptoms appear — which makes its results a natural starting point for the patient journeys this module describes.

What this page is, and what it is not. This module does not specify newborn screening. The module description names a dedicated KDS module for newborn screening as a sensible future addition, and this page is groundwork for it: a terminology handout that pins down which codes exist, which do not, and how a screening result would be represented if a project needs it today. Nothing here is normative for this module, and no profile in this module requires it.

The target conditions

The G-BA Kinder-Richtlinie (version of 15 May 2025, in force since 1 January 2026) lists 20 target conditions in § 17 (1). Screening for cystic fibrosis is a legally separate programme under 23 ff. and is listed separately below. Since May 2026 the list includes vitamin B12 deficiency, homocystinuria, propionic acidemia and methylmalonic aciduria — four additions that older overviews of "the German newborn screening" do not have.

Each condition is detected through a lead analyte. The assignment is not one-to-one in either direction: propionylcarnitine (C3) is the lead analyte for three conditions at once, while the single condition "carnitine cycle defects" needs three analytes. That is why the corresponding ValueSet holds 21 codes rather than 21 conditions.

ICD-10-GM is coarser than the programme: E71.3 covers MCAD, LCHAD, VLCAD and all three carnitine cycle defects, and E71.1 covers isovaleric acidemia, propionic acidemia and methylmalonic aciduria together. Only ORPHA separates them — which is precisely the argument this module makes for ORPHAcodes throughout.

The ValueSets, and where they come from

Six ValueSets ship with this module. Five of them are generated from the LOINC-SNOMED ontology, one is curated from the Richtlinie. The difference matters for anyone who wants to trust or extend them.

The generated five start from one expression describing "anything measured in a dried blood spot":

< 363787002 |Observable entity| :
    704327008 |Direct site| = 440500007 |Dried blood spot specimen|

which the LOINC-SNOMED edition answers with 335 concepts. Each ValueSet then narrows that set along one axis of the ontology — by component (<< 102651000 |Acylcarnitine|, << 52518006 |Amino acid|, << 38082009 |Haemoglobin|) or by property (118524006 |Catalytic concentration|, 118563004 |Substance ratio|). Membership is therefore a fact about the ontology, not an editorial choice, and scripts/generate-nbs-valuesets.py --check re-derives the files and fails if they have drifted.

Why the ValueSets contain LOINC codes even though the query runs against SNOMED. Every LOINC-SNOMED concept carries its LOINC code as an alternateIdentifier, so the ontology can select the concepts while LOINC supplies the codes. That indirection is deliberate: tx.fhir.org — the terminology server this IG builds and validates against — does not know the LOINC-SNOMED edition, so an intensional ECL ValueSet would not expand at all. All 264 codes were confirmed individually with CodeSystem/$lookup before being written.

The generated ValueSets describe what can be measured in a dried blood spot, not what the German programme measures. They therefore also contain analytes that are not part of it — the lysosomal enzymes of conditions screened elsewhere, and a few analytes used for entirely different purposes. Use the curated lead-analyte ValueSet when you mean the German programme.

And they are not complete against LOINC — grouping costs coverage. The LOINC-SNOMED edition only covers the part of LOINC that has been mapped to SNOMED: 335 concepts, against 608 LOINC codes whose SYSTEM is a dried blood spot. Roughly 45 % of the LOINC codes are therefore absent from the ontology, and the missing ones are not exotic — 50086-8 (C5-OH, the lead metabolite for 3-MCC deficiency) and 53166-5 (C4) are among them, as are most of the diagnostic ratios built from them. Measured per group: acylcarnitines cover about half of LOINC's dried-blood-spot acylcarnitine codes, haemoglobin fractions about a third.

That is why a sixth ValueSet exists. All DBS analytes binds intensionally to LOINC's own SYSTEM property (LP21304-8), so it is complete by construction and does not age — a new dried-blood-spot code from LOINC is in it without anyone touching a file. What it cannot offer is the clinical grouping: analytes, ratios, panels, interpretation codes and unrelated assays (drug levels, serology, PSA) all sit side by side. Use the grouped sets when you need a clinically bounded list, the complete one when you need coverage.

Representing a result

A single measurement is an Observation. The LOINC analyte code already carries both the specimen (DBS) and the property, so the code alone answers "what was measured, in what, and as what kind of quantity":

• code — the analyte code from the table above
• value[x] — valueQuantity in µmol/L (UCUM umol/L) for the [Moles/volume] codes; a unitless valueQuantity (UCUM 1) for ratios; a valueCodeableConcept for [Presence] codes
• interpretation — for screening, the LOINC LA answer codes are more precise than the generic HL7 interpretation codes: LA18592-8 In range, LA18593-6 Out of range
• category — laboratory
• specimen — a Specimen of type SNOMED 440500007 Dried blood spot specimen

A complete report nests three levels: the report panel 57128-1, an overall interpretation 57130-7 (answered with codes such as LA12428-1 All screening is in range for the conditions tested or LA18944-1 Screen is out of range for at least one condition), and per-condition interpretations that sit between the panel and the individual analytes.

57128-1  Newborn Screening Report summary panel
 ├─ hasMember → 57130-7  overall interpretation        → LA12428-1 / LA18944-1 …
 ├─ hasMember → 46752-2  MCAD/GA-2 interpretation      → LA18592-8 / LA18593-6
 │    └─ derivedFrom → 53175-6  C8       (valueQuantity µmol/L)
 │    └─ derivedFrom → 53177-2  C8/C10   (valueQuantity 1)
 └─ hasMember → 46762-1  hypothyroidism interpretation → LA18592-8 …
      └─ derivedFrom → 29575-8  TSH      (valueQuantity)

Where a screening result leads to a diagnosis, that diagnosis belongs in this module's own profiles — a suspicion from screening is a clinical diagnosis with verificationStatus = unconfirmed, and its genetic confirmation is a genetic diagnosis. The SMA case example walks exactly that path, from a positive newborn screen to molecular confirmation.

Limits — what LOINC does not offer

The table below lists the codes this module looked for and did not find. Each entry was checked against LOINC directly, rather than inferred — so "missing" means searched-for and absent, not merely unknown to us.

Why this matters: wherever a code is missing, an implementer needs one anyway and will invent a local one. Naming the gaps here means the same gap gets the same workaround across sites, instead of a different private code in each.

LOINC has retired most of the per-condition interpretation codes. A second verification pass over the whole code set found 18 DISCOURAGED codes, and for eight of them there is no active condition-specific successor — only group codes such as 46736-5 Fatty acid oxidation defects, 46744-9 Organic acidemias and 46733-2 Amino acidemias. That affects LCHAD, VLCAD, CPT2/CACT, glutaric aciduria I, propionic and methylmalonic acidemia, homocystinuria and tyrosinemia. Anyone who wants to keep naming the individual condition has to carry the specificity elsewhere — through the group code on Observation.code plus a link to the Condition, or through the underlying analyte observations. Two codes that still look like obvious successors, 46766-2 and 46767-0, are DISCOURAGED themselves.

The same pass established that the sickle cell screen has changed shape rather than merely changed code: the three method-specific "haemoglobin pattern" codes (54104-5, 54103-7, 54105-2) are all DISCOURAGED, and LOINC now models the finding as a ranking of the haemoglobins actually detected (64117-5 Most predominant hemoglobin in DBS and its siblings) plus a suspicion code (71592-0). The table above uses the screening panel code 54081-5; a project that needs the finding itself, not the fact that screening happened, should use 64117-5.

One code was caught by the build, not by the research. 54104-5 Hemoglobin pattern in DBS by HPLC resolves perfectly well and looks like the obvious lead analyte for sickle cell disease — but LOINC gives it the status DISCOURAGED, which the IG Publisher reported once the ValueSet existed. It has been replaced with the panel code 54081-5 (ACTIVE); the specific finding for sickle cell disease is the HbS fraction 56476-5, which the generated haemoglobin ValueSet contains. The generator now reads the LOINC STATUS property alongside the display, so a discouraged or deprecated code is reported rather than silently published.

One methodological caveat: the display texts come from tx.fhir.org, which does not expose LOINC's LONG_COMMON_NAME as a separate property. They are the server's display values — the ones a FHIR Coding.display is validated against — and were not additionally reconciled against the official LOINC release file.

Sources

• G-BA Kinder-Richtlinie, version of 15 May 2025, in force 1 January 2026 — § 17 (target conditions), 23 ff. (cystic fibrosis)
• LOINC 2.82 via tx.fhir.org
• LOINC-SNOMED edition, Snowstorm branch MAIN/SNOMEDCT-LOINC, version 2026-03-21
• ORPHA 2025 and ICD-10-GM 2026 via the MII Ontoserver