Ecosystem evaluation — external reference-data sourcing (medicines, and disease/injury terminologies)¶

Date: 2026-06-19 Status: Evaluation. Spec unchanged; no ADR minted. Captures sourcing research for the reference-data service tier — a separable service consumed by a Cairn node, not part of the wire core. Medicines sourcing (§1–§7) and disease/injury concept identifiers (§8) are both written up; a short list of human-verify items (licence clauses behind 403 walls) remains in §7 and §8.4. To be revisited before any of this is committed to. Subjects: open/government drug-reference feeds (WHO INN, RxNorm, DailyMed/openFDA, TGA ARTG, PBS) and disease/injury classifications (ICD-10/11, ICPC, alternatives) — evaluated for license compatibility with a freely-redistributable AGPL-3.0 node.

Note

This is an ecosystem evaluation — not architecture and not a decision. It records what reference-data sources exist, on what licence, and what we concluded about fit. The reference-data service is an optional external tier: a bare Cairn node records medication as INN-anchored substance UUID + dose + amount + units + formulation enum and disease as a stable concept code, and is a complete EHR without any bundled commercial drug or terminology database. Nothing here sits on the inter-node path, so a licence-encumbered source can never contaminate interoperability or the safety floor — it simply doesn't attach. That posture is exactly what founding principle 12 (uniform core, plural edges) and the language-substrate rule (§9) require of an advisory, fit-for-purpose tier.

1. Framing — why the architecture makes this tractable¶

The decisive design choice (the user's, ratified in conversation 2026-06-19) is that the EHR does not foreign-key into an integrated proprietary product database. A medication is recorded as:

a substance identity anchored on the WHO INN (International Nonproprietary Name) as the stable concept UUID — not a free-text brand name that is spelled differently across sources and drifts over time;
plus dose, amount, units, and a formulation enum (tablet / ointment / solution / …) stored explicitly.

This matters because the proprietary parts of the drug-data world are precisely the integrated, value-added product databases (First Databank, Micromedex, Multum, AMH, DrugBank-commercial). By never depending on one for core clinical recording, Cairn needs only reference and decision-support layers, and those can be assembled from public-domain government feeds. The reference-data service is therefore a two-tier problem:

Tier 1 — generic / substance-level, jurisdiction-independent (the INN-anchored substance dictionary, plus classification and decision-support reference).
Tier 2 — packaging / product / country-specific, starting with Australia.

The single hardest gap — unchanged in 20 years since the user's drugref.org work — is drug–drug interactions (§4). The mission tie-breaker throughout: license-and-redistribution terms win over convenience. A source that is free-to-use but NonCommercial or NoDerivatives is not freely redistributable and cannot be bundled into the AGPL artifact; at best it is an operator-supplied, node-local, separately-licensed plug-in.

Important

The recurring trap is the licence, not the price. Four flavours of "free" are not AGPL-bundleable: NonCommercial (CC BY-NC — DrugBank, WHO EML), NoDerivatives, member/affiliate-gated (SNOMED CT and all its national extensions incl. AMT), and registration-walled-but-otherwise-open (UK dm+d via TRUD). Each is flagged 🚩 below.

2. Tier 1 — generic / substance-level (jurisdiction-independent)¶

Source	Body	Licence	AGPL-bundleable?	Verdict
WHO INN	WHO INN Programme	Public domain ("placed in the public domain … used without any restriction whatsoever")	✅ Yes	✅ Anchor. Names are free; only friction is access — no clean bulk file (registration-gated INN Hub API, or parse the biannual public-domain PDF lists). An engineering problem, not a legal one. Biannual updates.
RxNorm — Current Prescribable Content	US NLM	Public domain (only `SAB=RXNORM` + `SAB=MTHSPL`)	✅ Yes	✅ Primary normalization layer. Downloadable without a UMLS login. Ingredient / clinical-drug / dose-form / strength.
RxNorm — full release / RxNav-in-a-Box	US NLM	Public-domain core + UMLS-restricted SABs (First DataBank, Micromedex, Multum, VA)	🚩 No	⚠️ Avoid for redistribution. The offline-friendly RxNav-in-a-Box is UMLS-gated. Build your own offline store from the prescribable subset instead.
DailyMed / SPL	US NLM / FDA	Public domain	✅ Yes	✅ Active ingredient, strength, form, route, NDC; LOINC-coded label sections. Also the DDI-mining substrate (§4).
openFDA	US FDA	CC0 1.0	✅ Yes	✅ NDC directory, drug labels, FAERS. Narrow caveat: a few privately-submitted copyrightable items are marked.
Drugs@FDA	US FDA	Public domain	✅ Yes	✅ Approved products, ingredients, strengths.
PubChem	NCBI	Public domain	✅ Yes	✅ Chemistry layer (structures, InChI, identifiers). Per-depositor caveat for sub-collections.
ChEMBL	EMBL-EBI	CC BY-SA 3.0	⚠️ Caveat	Usable, but ShareAlike is copyleft — keep it an isolated, attributed data layer; don't mix into a combined work. Bioactivity-oriented; low priority for prescribing.
WHO ATC/DDD	WHO CC, Oslo (atcddd.fhi.no)	NC + no-derivatives + attribution	🚩 No	🚩 €200 fee + forbids commercial redistribution + forbids modification. Operator-supplied / separately-licensed only. Community scraper `fabkury/atcd` outputs CC BY-NC-SA, confirming it is not free.
WHO Essential Medicines List	WHO	CC BY-NC-SA 3.0 IGO	🚩 No (NC)	Free to access (eEML export at list.essentialmeds.org), but NC blocks bundling. Factual "on-list y/n" flag may be reconstructable from non-copyrightable facts — needs legal review.
DrugBank (full)	OMx / U. Alberta	CC BY-NC 4.0 / paid commercial	🚩 No	🚩 NonCommercial — excluded. Only the tiny DrugBank Open Data ID-mapping subset (CC0) is usable.
KEGG DRUG	Kanehisa Labs	Academic-only / paid commercial	🚩 No	🚩 Excluded.

3. Tier 2 — packaging / product (Australia first)¶

Source	Body	Licence	AGPL-bundleable?	Role
PBS API / CSV	Dept. of Health	CC BY 3.0 AU	✅ Yes	✅ Primary packaging backbone — item code, form & strength, pack size/quantity, manner of administration, restrictions, ATC linkage, pricing. See §3.1 for the 2026 format cutover.
TGA ARTG	TGA	CC BY (3.0 AU → 4.0 on newer releases)	✅ Yes	✅ Primary product registry — sponsor, trade name, ARTG ID, active ingredients, dosage form, status.
AMT (in SNOMED CT-AU)	ADHA / NCTS	SNOMED National/Affiliate licence	🚩 No	⚠️ *Free to use inside* Australia (free registration), NOT internationally redistributable. SNOMED affiliate licensing charges fees in non-member territories and forbids sub-licensee redistribution. Treat as a per-node, per-jurisdiction-licensed plug-in** fetched over the distribution plane — never bundled. Maps cleanly onto the ADR-0014 content-addressed-component posture.
NPS MedicineWise	—	—	—	☠️ Defunct (ceased 31 Dec 2022). Ignore.
AMH; Therapeutic Guidelines (eTG)	proprietary	All rights reserved	🚩 No	🚩 Paid/proprietary subscription products. Exclude.

3.1 PBS feed — viable, but the format just shifted (time-sensitive)¶

The user fed drugref from PBS data in EasyGP; the old pain was a PBS schedule format that "kept shifting like quicksand," eventually settling on a mostly-standards-compliant XML that nearly always had easy-to-fix parsing errors. As of 2026 that avenue is still viable but the XML is now retired — the format shifted one final time:

PBS XML and PBS Text files: discontinued 1 May 2026. Gone. Any ingest pointed at the XML broke this May.
PBS Offline: discontinued 1 March 2026. Vendor Schedule distribution ceased October 2024.
Replacement: PBS API v3 + monthly "PBS API CSV files" (every API endpoint/table exported as CSV, published monthly). The CSV bundle is the direct modern equivalent of the old bulk XML dump, and the structured CSV/JSON should largely end the parsing-error pain.

Practical specifics for rebuilding the ingest:

Licence: CC BY 3.0 Australia (confirmed on the data.gov.au PBS datasets — Item Report, ATC Report, Patient Category Report). Redistributable, commercial OK, attribution only — AGPL-compatible. ✅
Public API is free, no login (default subscription key) but brutally rate-limited: 1 request / 20 s, shared across all users globally. Per-item API enumeration of the full schedule is therefore a non-starter — use the monthly CSV bundle for bulk load; the dept's own developer docs recommend weekly snapshots.
Free registration via the PBS Data API Portal (data-api-portal.health.gov.au) yields a subscription key, higher limits, and — important for a dispensing system — embargo access to future schedules, so next month's data can be staged before it goes live. The public API exposes only current + trailing 12 months.
~14 tables, base URL https://data-api.health.gov.au/pbs/api/v3: items, item-overview, prescribers, schedules, atc-codes, organisations, restrictions, criteria, copayments, fees, programs, summary-of-changes, … The summary-of-changes endpoint is new vs. the EasyGP era — it gives the monthly delta directly, so full-dump diffing is no longer required.

Conclusion: PBS + TGA ARTG (both CC BY) give the entire Australian product/packaging layer with no licence fees and full redistribution rights — the EasyGP→drugref pattern, minus the brittle XML parser. Because the format just shifted again, the ingest should treat the PBS schema as a versioned external dependency (founding principle 11, legibility across time, applied to an upstream feed), not as a fixed column set.

4. The drug–drug interaction gap (still the hard part)¶

No comprehensive, clinically-validated, openly-redistributable, AGPL-compatible DDI database exists. Every clinical-grade source (Stockley's, Lexicomp, Micromedex, First Databank, DrugBank) is paywalled. The defensible open composite, in descending licence-safety:

ONC high-priority DDI list — expert-consensus set (Phansalkar et al., JAMIA 2012; + CredibleMeds QT drugs) in open-access literature. Clean and authoritative but minimal — a high-severity safety floor, not a full checker. Note: NLM's RxNav Interaction API that served it was permanently retired 2 Jan 2024; re-encode from the papers.
Mine DailyMed SPL interaction sections (public domain) with NLP. ONSIDES (Tatonetti lab) is MIT-licensed and proves the pattern for adverse events — reuse as precedent. Engineering- and clinical-validation-heavy, but the strongest from-scratch open foundation.
DDInter 2.0 (Central South University) — ~302k DDI records with severity + mechanism + management. The single best candidate for structured open DDI — but its exact download-page licence needs a human (browser) confirm (aggregators tag it CC0; that may describe metadata only; the site 403s automated fetch).
TWOSIDES / OFFSIDES, Hetionet — research-only FAERS signals; ambiguous/likely-NC; never primary alerting.

The honest conclusion: the curation is the moat, and it is the part no one gives away. This is exactly where Cairn could revive the spirit of the hand-curated drugref DDI set — but as an institutionally-owned, append-only overlay (principle 1), not a volunteer wiki (see §5).

5. Sustainability — why drugref.org soft-died, and the lesson¶

drugref.org (the user's ~20-year-old wiki, seeded by the Mercy Ships foundation, with a hand-curated DDI database) soft-died from structural volunteer attrition + grant-cycle mismatch (funders pay for novelty, never maintenance). But it survived in Canada because OSCAR's drugref2 replaced volunteer data-authoring with an automated feed off Health Canada's open Drug Product Database (DPD). That is the template:

Don't crowd-source the clinical facts. Aggregate them from institutionally-funded government feeds, and reserve scarce human curation for the thin, high-value layer machines can't supply (DDI severity, clinical judgment).

Sustainable funding models, most-durable first: (1) government public-good feeds (RxNorm, openFDA, DPD, dm+d, PBS, TGA — standing budgets, not grants); (2) consortium/member dues (SNOMED — durable funding, conditional openness); (3) foundation/nonprofit subscription (TAIR → Phoenix Bioinformatics became self-sustaining in a year); (4) commercial-services-around-open-core. Avoid CIEL-style single-maintainer key-person dependency.

6. Recommended sourcing stack¶

Substance anchor (Tier 1, global): WHO INN (public domain) as the UUID, + RxNorm Current Prescribable Content (public domain) for normalization/clinical-drug/dose-form, + DailyMed/SPL + openFDA (public domain) for ingredients/labels. All four bundle freely under AGPL.
Australian product/packaging (Tier 2): PBS API/CSV + TGA ARTG (both CC BY) as the freely-shippable layer. AMT/SNOMED CT-AU as a per-node NCTS-licensed plug-in, never bundled.
Other jurisdictions (same shape): open regulatory registry bundled (US FDA/NDC, Canada DPD, UK dm+d via free TRUD account); national SNOMED extension as a licensed plug-in.
DDI: ONC high-priority floor (re-encoded) → SPL-mined layer (ONSIDES-style, MIT) → DDInter if licence confirms → Cairn's own curated append-only overlay as the durable value-add.
Hard excludes: WHO ATC/DDD, WHO EML (NC), DrugBank-full, KEGG, AMH, eTG.

7. Verify-before-relying (could not auto-fetch; all 403'd)¶

DDInter exact download-page licence (the linchpin for structured open DDI).
Canada DPD and EU Article 57 per-dataset licence tags (expected OGL-Canada / CC-BY; confirm).
Literal CC BY version on current TGA/PBS downloads and the SNOMED CT-AU National Licence redistribution clause text from NCTS.

8. Disease & injury concept identifiers¶

Medication is as central as disease, and the requirement is identical: the decision-making pathway must key on solid, stable concept identifiers — never free-text names that are spelled differently across sources and drift over time. This is the same discipline as the INN anchor in §1, applied to the morbidity/injury axis. SNOMED CT is clinically the richest option but is excluded on the mission: it is member/affiliate-gated, charges fees in non-member territories, forbids sub-licensee redistribution — a money-spinner behind a paywall, the same defect that put AMT (§3) out. The realistic field is the WHO ICD family, the WONCA ICPC family, and a handful of genuinely-open biomedical ontologies.

Important

The key nuance: NoDerivatives is not fatal for the identifier use-case. Cairn needs to use classification codes verbatim as stable concept anchors — it does not need to modify the classification. WHO's CC BY-ND licence permits exactly that: copy, redistribute, and commercial use of the codes with attribution. The ND clause only bites if you ship a modified codelist, a translation, or your own crosswalk (ICD↔SNOMED, ICD-11↔ICD-10) — each of which needs a separate WHO agreement. So ICD is usable as the identifier substrate; the boundary to document is "verbatim codes yes, derived maps/translations no."

8.1 The candidates¶

Classification	Body	Licence	Bundle verbatim codes?	Modify / own crosswalks?	Stable IDs	Fit
ICD-11	WHO	CC BY-ND 3.0 IGO; API/software royalty-free (no standalone resale)	✅ Yes (attribution, commercial OK)	🚩 No (ND → separate WHO agreement)	Persistent URIs `id.who.int/icd/entity/{id}` + stem codes	★ Best technical fit — see §8.2
ICD-10	WHO	CC BY-ND 3.0 IGO (historically licence-application-gated)	✅ Yes (verbatim)	🚩 No (ND)	Alphanumeric codes (e.g. `J18.9`)	Legacy/bridging where ICD-11 not yet adopted; effectively frozen (~2019)
ICD-10-CM	US NCHS/CDC	Public domain (US gov)	✅ Yes	✅ Yes	Annual codes + addenda	🚩 US-specific, code-incompatible with WHO ICD-10/AM — licence-cleanest but least portable
ICD-10-AM	IHACPA (Sydney/NCCH origin)	Paid licence; free tier = NonCommercial, AU-internal, no-redistribution	🚩 No	🚩 No	Per-edition codes	🚩 Exclude / site-provided plug-in only — categorically AGPL-incompatible
ICPC-3	WONCA / WICC	"Openly available under a Creative Commons licence" — exact variant UNCONFIRMED	⚠️ Depends on variant	⚠️ Only if CC BY/CC0/CC BY-SA	Concept codes	GP-aligned, the natural primary-care coder — gated on §8.4 verify #1
ICPC-2 / 2e	WONCA / WICC	WONCA copyright, licence-gated (all rights reserved)	🚩 No	🚩 No	Rubric codes, mapped to ICD-10	🚩 Encumbered — exclude
ICPC-2 PLUS	Univ. Sydney FMRC → NCCH	Paid annual licence (~AUD 120–420/site, renewing as of Feb 2023)	🚩 No	🚩 No	Interface terms → ICPC-2	🚩 Exclude — the copyright-encumbered Australian derivative; Cairn ships only the capability to load it
SNOMED CT (full) / national extensions (incl. AMT)	SNOMED International / NRCs	Member/affiliate-gated; fee in non-member territories	🚩 No	🚩 No	SCTIDs	🚩 Excluded by mission — node-local licensed plug-in only

8.2 Why ICD-11 is the best technical fit for an offline-first, stable-ID record¶

Persistent URIs against name-drift. Every concept has a durable identifier rooted at https://id.who.int/icd/entity/{entityId}, plus codeable MMS stem codes. A URI + entity ID is exactly the "stable identifier, not a free-text name" the decision pathway needs — and it composes with principle 11 (legibility across time): the coded event carries the stable anchor, while its plaintext legibility twin records the human label as asserted at the time, so the event stays readable even as the classification moves.
Genuinely offline. WHO ships an official Docker container (whoicd/icd-api, ARM-supported) that runs the full Coding Tool + browser + API with no internet connection, mirroring the canonical URIs locally (id.who.int/icd/entity → yourserver/icd/entity). This is a clean fit for the fractal-topology node (ADR-0001) and the availability floor — no cloud dependency on the decision pathway. The API is free (cloud needs free registration; local needs acceptLicense=true).
The licence boundary to honour: ship codes/URIs verbatim with WHO attribution; do not redistribute a modified ICD, a translation, or a Cairn-built ICD↔SNOMED / cross-version map without a separate signed WHO agreement. Treat any such map as a separately-licensed artifact, never folded into the AGPL corpus.

8.3 Genuinely-open ontologies (clean AGPL-compatible enrichment substrate)¶

Not primary-care morbidity coders, but useful as a free, stable-ID semantic layer that cross-maps to the encumbered ones — and unlike ICD/ICPC they are fully modifiable, so Cairn can derive from them:

Mondo Disease Ontology — CC BY 4.0, OWL/OBO, stable MONDO: IDs, integrates/maps across ICD, SNOMED, Orphanet, OMIM. The broadest clean disease ontology. ✅
ORDO (Orphanet Rare Disease Ontology) — CC BY 4.0, stable ORPHAcodes; rare-disease-focused. ✅
HPO (Human Phenotype Ontology) — phenotype/sign-symptom layer, stable HP: IDs, but a bespoke licence (not plain CC) — usable-pending-check (§8.4 verify #3).
SNOMED CT Global Patient Set (GPS) — the only free SNOMED content for non-members, stable SCTIDs; a 2026 source suggests the licence may have shifted to CC BY-ND (was CC BY 4.0) — verify (§8.4 #2). ND would still permit verbatim-code use, same posture as ICD.
MedDRA — 🚩 subscription/paywalled, exclude.

8.4 Recommendation and open verifies¶

Note

Ratified. The ICD-11 decision below graduated from this evaluation to ADR-0025 (canonical home data-model §3.16), which fixes ICD-11 as the canonical classification interlingua with a local-terminology overlay (map-once, offered-not-forced, open mappings deferrable to a professional coder). Spec → v0.27.

Recommended disease/injury identifier stack: - Primary concept anchor: ICD-11 (CC BY-ND 3.0 IGO) — verbatim entity-URI/stem-code identifiers, free offline Docker container, commercial OK with attribution. The ND boundary documented in §8.2. (Ratified — ADR-0025.) - Bridging: ICD-10 (CC BY-ND) where ICD-11 isn't yet the local standard — same verbatim posture. - Primary-care layer: ICPC-3 — conditionally. If its open licence confirms as CC BY (not NC/ND), it becomes the natural GP-aligned coder and should be adopted for the primary-care reason-for-encounter axis. Until the variant is confirmed, treat as pending; do not assume usable. - Free semantic substrate: Mondo + ORDO (CC BY 4.0) for derivable, modifiable cross-mapping; HPO for phenotype pending its licence check. - Exclude / site-plug-in only: SNOMED CT full + AMT, ICD-10-AM, ICPC-2/2e/2-PLUS — each a node-local, separately-licensed dependency the deploying site supplies under its own licence; Cairn ships the load capability, never the data.

Open verifies (all sites 403'd automated fetch — need a human browser read): 1. ICPC-3 exact CC variant (CC BY vs CC BY-NC vs CC BY-ND) — icpc-3.info licence page + the WONCA "ICPC-3 to Become Openly Licensed" announcement. Highest priority — it decides whether the GP coder is in. See §8.5 for a full verification-attempt log and a manual recipe. 2. SNOMED GPS current licence — CC BY 4.0 vs CC BY-ND 4.0. 3. HPO custom licence full text (hpo.jax.org/app/license). 4. WHO crosswalk/translation separate-agreement terms — needed before Cairn ships any ICD-derived map.

8.5 ICPC-3 licence variant — verification (2026-06-19): exact CC variant still UNNAMED, but inference sharpened¶

A second, focused attempt to pin the exact ICPC-3 Creative Commons variant did not succeed automatically; the official WONCA announcement was then supplied directly (HH, from the WONCA site, January 2026 Working Party News, published Feb 2026). Decisive finding: even the primary source does not name the variant — the SPDX identifier will live on the licence deed attached to the data/download, not in the announcement.

What is firmly established (now from the primary source): - WONCA (the licensor of ICPC, via the WICC) "has now decided to make ICPC-3 openly available under a Creative Commons licence," "to remove barriers to adoption, implementation, and innovation worldwide" and to "strengthen primary care documentation, research, education, and digital health development globally." - ICPC-3 "is designed to interoperate with major international classifications and terminologies such as ICD-11, ICF, and SNOMED CT, supporting semantic interoperability" — i.e. it is purpose-built to slot in as a pluggable primary-care layer that produces ICD-11, exactly the ADR-0025 shape. It also carries "extensive inclusion terms and synonyms… a practical thesaurus," which maps neatly onto the local-terminology overlay (it can bulk-populate local-term→ICD-11 bindings). - The exact CC variant is not stated in the announcement or in any publicly machine-accessible source.

What was tried, and why it failed: - WebSearch (many phrasings): consistently returns "a Creative Commons licence," never the variant. - Authoritative pages all return HTTP 403 to automated fetch: the WONCA announcement (globalfamilydoctor.com/News/ICPC-3OpenLicense.aspx), icpc-3.info and its sub-tools (book./browser./claw.icpc-3.info), wicc.one, and the Wikipedia ICPC page. - Wayback Machine (web.archive.org) is blocked from this tool. - The ICPC-3 User Manual PDF was retrieved and text-extracted (the encrypted, "not-for-extraction" Routledge file — decrypted and parsed). It carries only the book's notice — "Copyright Material – Provided by Taylor & Francis – Not for Redistribution" — i.e. the commercial book's copyright, not the classification's open-data licence. (Note the split, and the irony: the manual is a paywalled T&F book even though the classification is openly licensed — exactly the trap to avoid conflating.) - A third-party GitHub LICENSE (Karim-53/Docs-for-ICPC) is GPLv3 for that repo's own docs — not authoritative for ICPC-3.

Best current inference (MODERATE confidence — still confirm before relying): the announcement's own wording now leans clearly toward a permissive CC BY: - "WICC and WONCA will provide guidance and support for translations and implementations" — translations are derivative works; an actively-supported translation programme is hard to reconcile with a NoDerivatives (-ND) clause. - "remove barriers to… innovation… digital health development globally" — cuts against a NonCommercial (-NC) clause (NC is the classic barrier to commercial digital-health adoption). Together these point to CC BY (commercial + derivatives allowed). Countervailing: one earlier search summary inferred CC BY-NC, and WONCA has not published the SPDX identifier — so this remains an inference, not a confirmation. Do not bundle until the licence deed on the data/download is read directly.

Why it's decisive: CC BY / CC0 / CC BY-SA → AGPL-compatible, and ICPC-3 becomes the natural pluggable primary-care layer producing ICD-11 (ADR-0025). CC BY-NC or CC BY-ND → not freely bundleable, usable only as a node-local plug-in the deployment licenses — the same posture as SNOMED/AMT.

Manual verification recipe (for HH): 1. Open https://www.icpc-3.info/ in a browser → look for a Licence / Terms / Copyright footer; the classification download and the API doc (icpc-3.info/documents/extra/API-Calls.pdf) usually state the data licence. 2. Read the WONCA announcement directly: https://www.globalfamilydoctor.com/News/ICPC-3OpenLicense.aspx — it should name the variant. 3. Check the Classification Workbench (https://claw.icpc-3.info/) and browser (https://browser.icpc-3.info/) footers. 4. If still unclear, email the WICC / ICPC-3 consortium (contact on icpc-3.info) and ask for the SPDX identifier of the data licence. 5. The decisive question: "May we redistribute the ICPC-3 classification verbatim, commercially, inside an AGPL-3.0 product?" — CC BY / CC0 / CC BY-SA = yes; CC BY-NC / any-ND = no.

Sources¶

Tier 1: WHO INN (who.int/teams/health-product-and-policy-standards/inn) · RxNorm Current Prescribable Content (nlm.nih.gov/research/umls/rxnorm/docs/prescribe.html) + Terms of Service + UMLS License Agreement · DailyMed SPL Resources · openFDA License (open.fda.gov/license) · ChEMBL licensing (chembl.github.io) · WHO ATC/DDD copyright (atcddd.fhi.no/copyright_disclaimer/) · WHO EML (list.essentialmeds.org) · DrugBank Terms of Use (go.drugbank.com/legal/terms_of_use) · KEGG Legal (kegg.jp/kegg/legal.html). Tier 2 / PBS: PBS New API & API CSV files news (pbs.gov.au/info/news/2024/12/...) · PBS Download (pbs.gov.au/info/browse/download) · data.pbs.gov.au documents 91327/91602/90834 · Accessing PBS embargo data (hpp.health.gov.au) · data.gov.au PBS Item Report (CC BY 3.0 AU) · TGA datasets (tga.gov.au/resources/datasets) · SNOMED licensing (snomed.org/get-snomed, snomed.org/licensing). DDI: Phansalkar 2012 (PMC3422823) · NLM RxNav Interaction API retirement · ONSIDES (github.com/tatonetti-lab/onsides, MIT) · DDInter 2.0 (NAR 2025) · nsides.io. Sustainability: OSCAR drugref2 (oscaremr.atlassian.net) · Health Canada DPD (open.canada.ca) · TAIR/Phoenix (PMC4795935) · OpenMRS CIEL. Disease/injury terminology (§8): WHO FAQ Licensing ICD-10 (cdn.who.int/.../who-faq-licensing-icd-10.pdf) · WHO Copyright policy (who.int/about/policies/publishing/copyright) · ICD-11 License (icd.who.int/en/docs/icd11-license.pdf) · ICD-API License + Docker container + Local Deployment (icd.who.int/icdapi/docs2/...) · ICD-10 CDN (icdcdn.who.int/icd10) · CDC NCHS ICD-10-CM Files (cdc.gov/nchs/icd/icd-10-cm/files.html) · IHACPA products & licenses (ihacpa.gov.au/health-care/products-and-licenses) · Lane Print Electronic Code Lists (ar-drg.laneprint.com.au) · WONCA "ICPC-3 to Become Openly Licensed" (globalfamilydoctor.com/News/ICPC-3OpenLicense.aspx) · icpc-3.info · WICC (wicc.one/icpc-classification) · ICPC-2 PLUS (en.wikipedia.org/wiki/ICPC-2_PLUS, sydney.edu.au NCCH) · Mondo (mondo.monarchinitiative.org) · ORDO (sciences.orphadata.com/ordo) · HPO (hpo.jax.org/app/license) · SNOMED GPS (snomed.org/gps).