HL7 history: from v2 (1987) to FHIR (2014–today)

Quick summary

• 1987–1996: HL7 v2 emerges to solve the HIS–LIS–RIS integration problem, using pipe-delimited messages over MLLP.
• 1996–2005: v2.x spreads worldwide; HL7 launches the RIM in 1997 to unify clinical semantics.
• 2005–2014: HL7 v3 and CDA Release 2 ship; v3 messaging struggles with adoption while CDA succeeds via IHE XDS.
• 2014–today: FHIR rewrites the rules with REST + JSON + the 80% rule; the 2020 ONC Cures Act makes FHIR R4 a mandatory standard for certified EHRs in the US.
• Vietnam: from scattered HL7 v2 in private hospitals (~2010), to the hl7vn/vn-core-ig repo in 2024, and the 2026 hl7.org.vn initiative led by Omi HealthTech.

1. Context — what did US healthcare need in the 1980s?

In 1987, a group of about 30 people gathered at the University of Pennsylvania to talk through a very specific problem: how to get the computers inside a hospital to talk to one another. This was not an academic conference. They were engineers from major hospitals, HIS vendors, and laboratories — people who, every morning, had to wrestle with a dozen different proprietary protocols to integrate their systems.

The 1980s context in the US was distinctive. Hospital Information Systems (HIS) were starting to spread after Meditech, Cerner, and Epic appeared. Each vendor had its own messaging format. A mid-sized hospital that wanted to connect its HIS to a laboratory information system (LIS) and a radiology information system (RIS) had to hand-build every integration bridge. The cost of integration could easily exceed the cost of the software itself.

The market signal at the time was unmistakable: the industry needed a shared messaging standard. The name "Health Level Seven" was taken from layer 7 of the OSI model — the application layer — because that is where clinical semantics actually live. The founders chose a pragmatic approach: not idealism, just something good enough that every vendor could implement it.

2. 1987–1996: HL7 v2 is born and explodes in adoption

The March 1987 workshop at UPenn is regarded as the official starting point of HL7. The first version — HL7 v1 — was an internal, conceptual document. The version that actually went into production was HL7 v2.0, released in September 1988. The HL7 organization itself became an ANSI-accredited Standards Development Organization in 1994.

The early v2 releases came out in quick succession: v2.1 in March 1990, v2.2 in December 1994, and v2.3 in March 1997. Of these, v2.3 became the most widely deployed; even today, many hospitals still run that exact version for ADT (Admission–Discharge–Transfer) flows, lab results, and prescriptions. The technical character of HL7 v2 is almost crude in its simplicity: pipes and carets as delimiters, MLLP framing over TCP, and an ABNF-style grammar that feels like programming.

MSH|^~\&|HIS|HOSP|LIS|LAB|202604301430||ORM^O01|MSG001|P|2.3
PID|1||MRN0001||NGUYEN^THI LAN||19850315|F

That very crudeness was the key to its success. An average engineer can read a v2 message with the naked eye, debug it with grep, and log it with tee. A parser library fits in a few hundred lines of code. By the mid-1990s, HL7 v2 had become the de facto standard for in-hospital integration across North America. As the internet took off at the end of the decade, new problems began to surface — but that is the next chapter's story.

3. 1996–2005: v2.x goes global, the RIM is launched

Through the 2000s, HL7 v2 entered a phase of "spreading and taking root." Versions v2.4 (2000), v2.5 (2003), v2.6 (2007), v2.7 (2011), v2.8 (2014), and v2.9 (2019) kept adding new domains — prescription management, vaccines, public health — while preserving the original syntactic philosophy. Countries outside the US started adopting it seriously: Australia, the Netherlands, Germany, then Japan and Korea.

But the community also began to notice v2's limits. Its pipe-delimited nature meant messages had no strict schema; every vendor extended things in its own way, producing incompatible "dialects" of v2. Free-form fields (Z-segments) were widely abused. When richer semantics were needed — for example, complex clinical data queries — v2 quickly showed its constraints.

In response, HL7 launched the Reference Information Model (RIM) project in 1997: a unified information model meant to cover all of clinical semantics. The RIM defined root classes such as Act, Entity, Role, Participation, and ActRelationship, from which every concrete structure would be derived. With this ambition, HL7 laid the groundwork for the next generation of standards: v3.

4. 2005–2014: the v3 + CDA ambition and what it taught us

2005 was a major milestone: HL7 published the Normative Edition of HL7 v3 together with CDA Release 2 (Clinical Document Architecture). The RIM was standardized as ISO/HL7 21731. On paper, this was a triumph. In practice, the story was much more complicated.

HL7 v3 messaging used XML, was built directly on the RIM, came with a strict schema and rich semantics. The price tag, though, was enormous complexity. A simple v3 message such as a patient registration could run hundreds of lines of nested XML, with every attribute bound to a controlled vocabulary. Writing a correct message required deep RIM-savvy business analysts — a profile of staff that was extremely scarce.

The consequence: large investment programs in v3 messaging hit hard walls. The UK NHS National Programme for IT (NPfIT) — one of the most ambitious health IT programs in history — eventually scrapped most of its v3 content because of the cost and complexity of implementation. Germany and Korea adopted v3 only at limited scale. Only a few specific areas, such as public health reporting, retained v3 messaging.

CDA, on the other hand, took a different path. CDA used the RIM but packaged it as clinical documents — each document is an XML file with a header and a body, can be digitally signed, and can be stored independently. The IHE XDS (Cross-Enterprise Document Sharing) integration profile chose CDA as the standard container for documents shared across organizations. As a result, CDA has endured in Europe, the US, and Japan — to this day, it remains the dominant standard for discharge summaries and administrative healthcare forms in many countries.

The big lesson from the v3 cycle: an idealized model on paper does not automatically translate into adoption. If the cost of implementation outweighs the short-term value, organizations will pick something simpler — or wait for the next generation. That lesson was the seed of FHIR.

5. 2014–today: FHIR and the healthcare API revolution

In 2011, Grahame Grieve — an Australian engineer with deep experience in the HL7 community — published a blog post titled "Resources for Health." The idea was simple: instead of trying to cram every clinical concept into the RIM, break information into independent Resources (Patient, Observation, MedicationRequest, and so on), each with its own schema, exchanged over a REST API and serialized as JSON or XML. The post spread quickly through the community. In 2012, HL7 officially adopted the project and gave it a name: FHIR — Fast Healthcare Interoperability Resources.

The first milestone was FHIR DSTU1, released on 30/09/2014. The "80% rule" philosophy was stated explicitly from the beginning: the spec would standardize 80% of common use cases, leaving the remaining 20% for the community to handle through extensions and profiles. This philosophy was the polar opposite of v3's "cover everything" ambition, and it landed exactly where modern developers were — people already comfortable with REST + JSON from the wider web world.

The release cadence that followed was steady. DSTU2 shipped in 2015. STU3 shipped in 2017. R4 reached its Normative Edition on 27/12/2018, with the technical 4.0.1 release — still called R4 — landing in 2019; this was the first time FHIR carried Normative content (specifically Patient and Observation), while most other Resources — including AllergyIntolerance, Condition, and Encounter — remained in Trial Use status. Because of this hybrid structure, R4 is often described as "Mixed Normative + STU."

In 2020, the ONC Cures Act Final Rule in the US issued a critical requirement: certified health IT modules under §170.315(g)(10) had to provide standardized FHIR API technology based on FHIR R4.0.1, with a compliance deadline of 31/12/2022. The rule did not blanket "every EHR" — it targeted certified modules — but because virtually every major US EHR needs certification to participate in federal insurance programs, the chain reaction was massive. Apple Health, Epic, Cerner, and Allscripts all adopted FHIR R4 within the next few years.

In 2022, HL7 released FHIR R4B (4.3.0) — a Trial Use release with limited changes from R4, not a new Normative edition. In 2023, FHIR R5 (5.0.0) launched officially, with significant improvements in pharmaceuticals, genomics, and medical devices. Even so, R4 remains the default choice for most countries building national IGs: US Core, JP Core, KR Core, CH Core, and AU Core all sit on R4.

By the mid-2020s, FHIR has become the de facto standard for healthcare interoperability worldwide. Apple, Google, and Microsoft all offer FHIR-native cloud services. WHO, Gavi, and the World Bank use FHIR for global health initiatives. When the wave of healthcare AI agents arrived in 2024–2025, FHIR naturally became the data layer underneath — because it was already present in every system that mattered.

6. Vietnam in the HL7 timeline

Vietnam came to HL7 about two decades later than the developed world, but its trajectory has its own character. Around 2010, a few private hospitals investing in new systems — Vinmec being the canonical example — began requiring HIS, LIS, and RIS to communicate over HL7 v2 instead of homegrown protocols. That was the first time HL7 v2 appeared in production in Vietnam, although the scope was still narrow: private hospitals and international facilities.

During the same period, the BHYT (social health insurance, BHYT) data track went its own way. Decision 4210/QĐ-BYT standardized the output data format for BHYT claim adjudication — but used plain XML, not HL7 v2 or CDA. It was later replaced by Decision 130/2023 and Decision 4750/2023, then by Decision 3176/QĐ-BYT (29/10/2024). The BHYT ecosystem therefore grew on a parallel track rather than merging into HL7.

The turning point came in 2024. The Department of Health Information Technology under the Ministry of Health (the unit's name as of publication) released the hl7vn/vn-core-ig repo — the first draft of a Vietnam Core IG built on FHIR R4, with canonical http://fhir.ehealth.gov.vn/core/. Around the same time, the fhir.chiaseyhoc.vn community mirrored the content and began conversations around what profile structures should look like for Vietnam.

2025 brought the legal jolt. Circular 13/2025/TT-BYT (issued 06/06/2025, effective 21/07/2025) requires hospitals to complete electronic medical records (EMR) by 30/09/2025 at the latest, and all healthcare facilities by 31/12/2026. The same year, Law 91/2025/QH15 on Personal Data Protection (effective 01/01/2026) and Decree 356/2025/NĐ-CP with implementing guidance — classifying health data as sensitive personal data — tightened security requirements across the board. When mandatory EMR meets sensitive data protection, FHIR becomes the natural choice for being interoperable while keeping consent under control.

In 2026, the hl7.org.vn initiative led by Omi HealthTech (OmiGroup) picks up the work — relaunching the Vietnam Core IG at higher quality, openly under a CC-BY-4.0 license, with the ambition of becoming the official HL7 Affiliate Vietnam. The new canonical is http://fhir.hl7.org.vn/core/; the project is community-driven, open to contributions from any vendor, hospital, or researcher. It is the natural reconnection between Vietnam and the global HL7 timeline that it has been twenty years late to join.

7. Lessons for Vietnam

Forty years of HL7 history give Vietnam several very concrete lessons as it builds the VN Core IG.

One: do not repeat the v3 mistake. Do not wait for a perfect model before launching. Embrace FHIR's 80% rule: standardize what is common, let extensions handle the long tail. VN Core should ship as a useful draft and iterate based on community feedback — not sit in a drawer waiting to be "fully ripe."

Two: community matters more than administrative orders. International HL7 succeeded by remaining vendor- and organization-neutral. The hl7vn/vn-core-ig repo was bootstrapped through initial efforts by the regulator — the next step is to move toward a community-driven model, with an HL7 Affiliate as the official representative and working groups open to every interested organization.

Three: R4 is the sweet spot for Vietnam right now. R5 is newer, but the tooling, tutorials, and other national IGs all cluster around R4. Choosing R4 means standing on the shoulders of giants — JP Core, US Core, AU Core are all on R4; SUSHI, the IG Publisher, and validators are all stable on R4.

Four: invest in terminology in parallel with profiles. FHIR is hollow without ICD-10 VN, SNOMED CT VN, LOINC VN, the medical service catalog, and the post-Resolution-202 administrative division catalog of 34 provinces. Profiles and terminology must move together starting from version 0.1.0.

Five: anchor FHIR to the legal framework. Circular 13/2025/TT-BYT (electronic medical records), Law 91/2025/QH15 (personal data protection), Decree 356/2025/NĐ-CP (DPIA and DPO guidance), and Decision 1332/QĐ-BYT (the personal health record on VNeID, Vietnam's national digital identification app) — these are the demand drivers; FHIR is just the tool. Any IG that does not cite its legal basis clearly will struggle to be adopted by the public sector.

8. References and further reading

Primary references

• HL7 International — v2 history page: hl7.eu/HL7v2x/v291/hl7v291tab0104.htm
• FHIR R4B — version history page: hl7.org/fhir/R4B/history.html
• FHIR R4 — AllergyIntolerance (status Trial Use): hl7.org/fhir/R4/AllergyIntolerance.html
• ONC — Key Dates Cures Act: healthit.gov/sites/.../Key_Dates_2020.pdf (FHIR R4.0.1, §170.315(g)(10), deadline 31/12/2022)
• Original Vietnam Core IG repo: github.com/hl7vn/vn-core-ig
• Community mirror: fhir.chiaseyhoc.vn

Relevant Vietnamese legal documents

• Circular 13/2025/TT-BYT — Electronic medical records (issued 06/06/2025, effective 21/07/2025).
• Law 91/2025/QH15 — Personal Data Protection Law (effective 01/01/2026).
• Decree 356/2025/NĐ-CP — Implementing guidance for the Personal Data Protection Law (effective 01/01/2026).
• Decision 3176/QĐ-BYT — Output data standard for healthcare services (29/10/2024).
• Resolution 202/2025/QH15 — Reorganization of provincial-level administrative units, from 63 down to 34 (operational from 01/07/2025).

More from the knowledge hub