1 of 43

Terminology Mapping�in Brazil: From Local Codes to Global Meaning

Terminology Services, Governance, OCL Mapper, and the path to semantic interoperability in the Brazilian health system

Jussara Rotzsch – Chair HL7 Brazil

Jonathan Payne – Director, Open Concept Lab; Co-chair, WHO DPI-H Metadata Registry Workgroup

2 of 43

Agenda

1

Brazil Case Introduction

Interoperability paradox, barriers, RNDS, and FHIR adoption reality

2

Terminology Services in the Brazilian Context

Conformance illusion, the semantic gap, and OCL infrastructure

3

Governance Framework for National Terminology Assets

Institutional roles, one vocabulary, DPI, and the paradigm shift

4

The SGI Layer of DPI for Health (DPI-H)

5

OCL Overview and Brazil SGI Use Cases

Native authoring, automated FHIR publication, IPS Brasil

6

Live Demo: Mapping Brazilian Terminologies

Walkthrough of OCL and OCL Mapper in practice

7

Lessons Learned and Next Steps

The interoperability equation and the coalition for DPI

HL7

3 of 43

01

SECTION

Section 01

Brazil Case Introduction

Interoperability paradox, barriers, RNDS, and FHIR adoption reality

HL7

4 of 43

The Brazilian Interoperability Paradox

83% of leaders see the topic as a strategic priority. The operational reality tells another story.

Where data flows a bit better (Back-office)

HIS/EHR

14.3%

Data Management

13.4%

Where data stops (Front-office)

Patient Engagement

7.1%

Telemedicine

7.1%

Education and Training

6.2%

Brazilian interoperability is still driven by regulatory obligations and billing. The patient continues to be treated as a data object, not as the center of the strategy.

5 of 43

More integrations don't mean more efficiency.

Without data governance, ad-hoc integrations create exponential complexity and technological lock-in (vendor lock-in).

\

The Illusion of Integration

Real Interoperability

HIS

LIS

EHR

Billing

Airports without international routes. Everything works locally, but longitudinality is broken.

Patient

Internal�Systems

RNDS

Wearables

External�Labs

Patient Longitudinality. Data follows the individual, not the software.

6 of 43

The Real Barriers (It's Not Technology).

Systemic failure occurs due to absence of strategic priority from top management and ineffective governance, not for lack of standards.

Economic Block

Implementation cost / Lack of budget

58.9%

Low vendor maturity

40.2%

Executive Block

Lack of institutional priority at Board level

55.4%

Cultural resistance from professionals

31.2%

Technical Block

Lack of applied standards

51.8%

Shortage of specialized talent

50.0%

7 of 43

The Reality of Standards in Brazil

The market doesn't suffer from lack of technology, but from lack of organizational discipline to apply it.

13.1%: HL7 FHIR adoption.�(The emerging gold standard).

~40%: The fragmented legacy. Systems that don't connect, or use legacy/closed standards (HL7 v2.x, CDA, Others).

~46%: The knowledge vacuum. Can't answer or the system doesn't apply standards.

71% of countries globally already use FHIR actively. Brazil has 13.1% local adoption. But does adopting the structure solve the problem?

8 of 43

9 of 43

02

SECTION

Section 02

Terminology Services in the Brazilian Context

Conformance illusion, the semantic gap, and OCL infrastructure

HL7

10 of 43

The "Webservice" That Isn't.

RTS is a batch CSV distribution, not a live API — and why that's the gap FHIR closes.

NOT a live API

Batch distribution mechanism.

You download CSV files organized by competência (reference period).

Static exports, not runtime queries.

The name "Webservice" is misleading by modern standards.

Pull-by-period

Pull-by-period, not query-by-code.

Download the CSV snapshot for a given competência, process it locally, work from that file.

No $expand, no $lookup, no real-time validation — just periodic file retrieval.

The replacement model

Bringing real semantics into FHIR.

The replacement model — terminologia.saude.gov.br/fhir via OCL — was adopted to integrate terminology semantics directly into FHIR.

CodeSystems, ValueSets, ConceptMaps served as proper FHIR resources — meaning, not just structure.

Queryable at runtime via $expand, $lookup, $validate-code.

11 of 43

12 of 43

The Terminology Lifecycle.

How terminologia.saude.gov.br authors, maps, manages, and distributes Brazil's health terminology.

OCL — Open Concept Lab (terminology management platform)

1

AUTHOR

Domain experts define CodeSystems for clinical concepts — BRParentesco, BRHPVInterpretacao, BRCondutaColposcopia, BRTabelaSUS

2

MAP

ConceptMaps link national codes to LOINC, SNOMED CT, ICD-10, and reconcile legacy TUSS / SUS terminologies.

3

MANAGE

Versioning, status, and governance handled in OCL — every change is reviewed, dated, and traceable.

4

DISTRIBUTE

National CodeSystems, ValueSets, ConceptMaps shipped via a FHIR Implementation Guide as installable packages — drop-in for implementers, no CSV ingest required.

FHIR Terminology Server terminologia.saude.gov.br

Implementation Guide terminologia.saude.gov.br/fhir

13 of 43

OCL: Brazil's Semantic Infrastructure.

The portal built and freely available today: terminologia.saude.gov.br

69

National CodeSystems�(Native in OCL: ICD-10, SIGTAP, LOINC)

1

FHIR Terminology�Implementation Guide (IG)

Scalable Reuse�across the entire Ecosystem

Terms governed by CGSD and the Terminology Subcommittee (INSPIRA). Mappings and "Bindings" officially aligned with RNDS, BR-Core and IPS Brasil.

14 of 43

15 of 43

03

SECTION

Section 03

Governance Framework for National Terminology Assets

Institutional roles, one vocabulary, DPI, and the paradigm shift

HL7

16 of 43

Terminology Governance Framework

1

Policy (CGSD / SEIDIGI)

Institutional authority and strategic decision making for SUS Digital.

2

Clinical Rules (Terminology Subcommittee)

Technical analysis, specialty knowledge, and expert clinical review of mappings.

3

Implementation (HL7 Brasil)

Translation of rules into standards: BR-Core, IPS Brazil profiles.

4

Execution (OCL & OCL Mapper)

Curation, versioning, and national FHIR terminology services.

Feedback Loop

OCL executes decisions. It doesn’t make them.

17 of 43

Semantic Governance: One country, one vocabulary.

Who writes

Specialists decide the words.

A clinical panel agrees on the meaning of each medical term, freezes its spelling, and revises it as medicine evolves.

Slow, careful, official.

Where it lives

One official copy.

The approved dictionary resides in a single trusted digital location that anyone can query at any time.

No outdated printed copies in hospital drawers.

Who queries it

Hospitals, Apps, Reports.

Every system queries the same dictionary instead of maintaining its own mapping spreadsheets.

18 of 43

Terminologies are Digital Public Infrastructure (DPI).

Shared Foundations

Reusable across the entire national health ecosystem. Not built for a single isolated project or lab.

National Services

Open APIs. Accessible as basic utility services that any system (public or private) can and should invoke.

Built to Scale

Architecturally designed to handle the national load (Big Data, PGHD, +2.8 billion records in RNDS). Not designed for "pilots".

Just as the CPF is the citizen's identification infrastructure, OCL is the infrastructure of clinical meaning.

19 of 43

Architectural Pattern for Terminology Governance and Runtime Services in Brazil

OCL �Mapper

OCL Terminology Server

OCL �TermBrowser

OCL �Mapper

FHIRSmith

Authoring & Governance

Publishing & Runtime

Authoring: Single source of truth for management of CodeSystems, ValueSets and ConceptMaps.

Governance: Proposals, reviews and approvals documented with full provenance traceability.

Publication-time Sync → �1 Source of Truth.

Runtime APIs: FHIR operations $lookup, $expand, $translate, $validate-code available to systems nationwide�

Terminology Publishing: IG builds, �profile validation

Brazil operationalizes the SGI pattern: OCL governs the content, FHIRSmith serves it — �one source of truth, two complementary responsibilities.

20 of 43

04

SECTION

Section 04

The Semantic Governance Infrastructure Layer of Digital Public Infrastructure for Health (DPI-H)

HL7

21 of 43

22 of 43

23 of 43

24 of 43

DPI-H Architecture Principles

25 of 43

Layers of the WHO/ITU DPI-H Reference Architecture (DRAFT)

Health System Goals

Build Trusted Data Foundations, Digitally-enabled Health Workforce, Ensure Quality and Continuity of Care, �Establish Trusted Personal Health Records, Optimize Supply Chain Management, �Digitally strengthened Health Financing, Strengthen Climate and Epidemic Resilience

DPI-H Components

Business services: Lifelong Health Record, Computable Decision Support Engine, HMIS, Public Health Surveillance Platform, etc.�Metadata registries: Client, Facility, Health Workforce, Health Product, Terminology Service, Logical Information Model Registry, etc.

Functional Requirements and Specifications

Detailed 'shall' statements, conformance profiles, and validation rules that make the architecture testable.

Capabilities

What the health system must do to achieve its goals.

e.g. master data & registry management, decision support, care coordination, claims management, �disease surveillance, analytics

realized by

delivered by

specified by

26 of 43

What makes Terminology Services DPI-H: Semantic Governance (DRAFT)

Runtime �Layer

Publication Layer

Packaging, validation, build, distribution

FHIR operations lookup, validation, expansion, translation

“Runtime services without governance produce content that cannot be trusted, while

governance without runtime services produces decisions that cannot be operationalised.”

Application Layer

FHIR Terminology Ecosystem, CRMI, FHIR Packages

Applications that consume content and services exposed by the runtime layer

FHIR TS, CTS2

SVCM IHE Profile, Implementation Guides

Governance Layer

Stewardship, approval, authoring, lifecycle, publication authority, collaboration, provenance

No unifying governance interop framework, �but there are important fragments: �ISO 17117, SNOMED governance, HL7 UTG, etc.

Functions

Existing Standards

27 of 43

DPI-H Semantic Governance Infrastructure Capabilities (DRAFT)

SGI Capability Category

Definition

Value Proposition to Government

Intake

Managing the intake of change requests, new upstream releases, and deprecation notifications from stakeholders and standards organizations

Ensures national standards stay current with global updates and local needs without ad-hoc, error-prone processes

Authoring and Publication Processes

Coordinating the review, approval, versioning, and release of new content and changes to existing content according to defined business rules and audit requirements

Guarantees that only vetted, traceable changes reach the health system — reducing costly downstream errors

Monitoring

Tracking compliance with governance rules, content quality/staleness, and field-level uptake of published standards

Reveals whether published standards are actually being used and where gaps or drift are emerging before they cause harm

Distribution

Providing reliable, access-controlled infrastructure for delivering content across multiple consumption modalities in a defined architecture

Makes national standards practically accessible to all implementers — a standard no one can retrieve is a standard no one follows

Consumption

Enabling downstream stakeholders and upstream maintainers to discover, retrieve, and integrate published content into their systems and workflows

Lowers the barrier for systems to align with national standards, accelerating interoperability across the health ecosystem

28 of 43

05

SECTION

Section 05

Open Concept Lab Overview and Brazil SGI Use Cases

Native authoring, automated FHIR publication, IPS Brasil

HL7

29 of 43

What is the Open Concept Lab?

OCL is an open-source terminology management toolkit to collaboratively map, manage, and use health data standards alongside the global community.

Open-source

Recognized as a Digital Public Good1

Cloud-based

Use OCL Online for community collaboration

Community-driven

OpenHIE, HL7

Standards-based

FHIR, IHE SVCM

Globally adopted

Implemented globally

30 of 43

Snapshot of Global OCL Implementations

Brazil International Patient Summary (IPS) Exchange

Kenya National Health Terminology Service

OpenMRS Concept Dictionary Management

Learn how 1,000s of health facilities use metadata managed in OCL

31 of 43

Architectural Pattern for Terminology Governance and Runtime Services in Brazil

OCL �Mapper

OCL Terminology Server

OCL �TermBrowser

OCL �Mapper

FHIRSmith

Authoring & Governance

Publishing & Runtime

Authoring: Single source of truth for management of CodeSystems, ValueSets and ConceptMaps.

Governance: Proposals, reviews and approvals documented with full provenance traceability.

Publication-time Sync → �1 Source of Truth.

Runtime APIs: FHIR operations $lookup, $expand, $translate, $validate-code available to systems nationwide�

Terminology Publishing: IG builds, �profile validation

Brazil operationalizes the SGI pattern: OCL governs the content, FHIRSmith serves it — �one source of truth, two complementary responsibilities.

32 of 43

Use Case #1: IPS Brasil – runtime translation of local codes to SNOMED CT via governed ConceptMaps

OCL governs the SIGTAP/ICD-10 → SNOMED CT mappings; FHIR service executes $translate; output conforms to IPS with meaning preserved.

Step 1: Data Capture

A SUS patient is coded locally using SIGTAP or ICD-10.

Step 2: The Translation Engine (ConceptMap)

The system queries the public infrastructure (OCL + FHIRSmith) using the FHIR $translate operation. The official mapping converts the local code into SNOMED CT.

Step 3: Global Destination (IPS)

The data is packaged into the International Patient Summary using SNOMED CT, enabling global continuity of care and portable patient control via the "Meu SUS Digital" app.

The Challenge

Brazil relies heavily on localized terminologies (SIGTAP, CBHPM) while transitioning to international clinical standards (SNOMED CT, ICD-11). Manually mapping thousands of concepts across spreadsheets is slow, error-prone, impossible to version control, and hard to govern.

33 of 43

Local Medication Table

Meddra

OCL Mapper

  • Resolving complex synonymy
  • Handling pharmaceutical presentations
  • Aligning generic vs. brand names

Ontologia Brasileira de Medicamentos (OBM)

Governance Loop

Concepts without an OBM equivalent are tagged as “Unmatched (0)” and exported as governance reports to expand the national ontology.

Use Case #2: National Medication Governance (OBM)

34 of 43

06

SECTION

Section 06

Live Demo: Mapping Brazilian Terminologies

Walkthrough of OCL and OCL Mapper in practice

HL7

35 of 43

36 of 43

What does an AI-enabled Terminology Mapping Workflow look like?

Recommend

e.g. LLM-as-judge

Preprocess

e.g. keyword expansion

Multi-algorithm Retrieval

e.g. semantic search, scispacy, bridge search, etc.

User initiates matching process

Accompanied review process

Save/export final mappings in desired format

Auto-match Process

User Workflow

Validate & �Learn

Auto-match

Ranked Candidates

Candidate Pool

37 of 43

Input Data

LOINC Mapping Example 1

Order Mnemonic

Order Description

Tube Type

Specimen

Test Mnemonic

Test Description

Result Type

Units of Measure

BNP

B type Natriuretic Peptide, Plasma

Pink

Plasma

BNP

B type Natriuretic Peptide, Plasma

N

pg/mL

AI Assistant – RECOMMEND

30934-4 shows excellent alignment: exact component match (Natriuretic peptide.B), system compatibility (Ser/Plas covers Plasma), matching units (pg/mL), and strong algorithm consensus (100% normalized score). The LOINC system 'Ser/Plas' explicitly includes plasma as indicated in the long common name 'in Serum or Plasma'. High COMMON_TEST_RANK (180) and COMMON_ORDER_RANK (57) indicate this is a frequently used, well-established code. Verify that your local term measures mass concentration rather than molar concentration - if molar units are used, consider 47092-2 instead.

AI Assistant Analysis

Correct candidate RECOMMENDED with explanation of each matching axis.

Grade

Candidate Pool: A

AI Assistant: A

PRIMARY – 30934-4 Natriuretic peptide B [Mass/volume] in Serum or Plasma

ALTERNATES

  • 47092-2 Natriuretic peptide B [Moles/volume] in Serum or Plasma

PRIMARY – 30934-4 Natriuretic peptide B [Mass/volume] in Serum or Plasma

38 of 43

Input Data

LOINC Mapping Example 2

CONDITIONAL assessment identified correct alternate and guided mapper with verification steps.

Grade

Candidate Pool: A

AI Assistant: A

Order Mnemonic

Order Description

Tube Type

Specimen

Test Mnemonic

Test Description

Result Type

Units of Measure

SPEP

Protein Electrophoresis, Serum

SST

Serum

GLOBS

Globulin

N

g/dL

AI Assistant – CONDITIONAL

The input term 'Globulin' with system 'Serum' creates ambiguity between gamma globulin (2874-6) and total globulin (10834-0, 2336-6). The order mnemonic 'SPEP' (Serum Protein Electrophoresis) and units 'g/dL' suggest this is likely gamma globulin from electrophoresis, supporting 2874-6. However, the generic term 'Globulin' could refer to total globulin. Key decision point: Does your local 'Globulin' term specifically mean gamma globulin from SPEP, or does it represent total globulin? Verify: (1) Whether SPEP order produces gamma globulin results or total globulin, (2) If your lab reports gamma globulin separately from other globulin fractions, (3) Whether the result represents a calculated total or electrophoresis-derived gamma fraction. If SPEP context confirms gamma globulin measurement, proceed with 2874-6.

AI Assistant Analysis

PRIMARY – 2874-6 Gamma globulin [Mass/volume] in Serum or Plasma by Electrophoresis

ALTERNATES

  • 10834-0 Globulin [Mass/volume] in Serum by calculation
  • 2336-6 Globulin [Mass/volume] in Serum
  • 2336-6 Globulin [Mass/volume] in Serum

39 of 43

Input Data

LOINC Mapping Example 3

Order Mnemonic

Order Description

Tube Type

Specimen

Test Mnemonic

Test Description

Result Type

Units of Measure

UAMICRO

Microscopic Urinalysis

URN

Urine

UAWBC

UA WBC

N

per HPF

AI Assistant – REJECT

All current candidates have System='WBC' (white blood cells as specimen) but the input term 'UA WBC' refers to white blood cells IN urine (System should be 'Urine'). The candidates measure various enzymes/chemicals within isolated white blood cells, not white blood cell counts in urine sediment. The metadata confirms this is for 'Microscopic Urinalysis' with units 'per HPF' (high power field), indicating microscopic counting of leukocytes in urine. ��Key decision point: Search for LOINC codes where System='Urine' and Component='Leukocytes' with counting properties (NCnc/Naric) rather than the current WBC-system enzyme measurements.

AI Assistant Analysis

NO CANDIDATES IN POOL RECOMMENDED

Correctly REJECTED candidates in pool and provided steps to find the right match

Grade

Candidate Pool: F

AI Assistant: A

40 of 43

LOINC Mapping Evaluation: Results

Candidate Pool Generation

  • Correct match in the candidate pool: 86% (172 / 200)
  • Correct match missing from candidate pool: 14% (28 / 200)

OCL AI Assistant for automated recommendation

  • Correct match identified: 88% (151 / 172)
  • Incorrect recommendation despite correct candidate being available: 12% (21 / 172)
  • Correct rejection / abstention: 11% (3 / 28)

Top-n Recall (Aggregate)

top-1

top-5

top-10

53%

80%

86%

For “human in the loop” mapping workflow, the goal is increased accuracy/ efficiency and reduced cognitive load for the human mapper

Source: P Naliyatthaliyazchayl et al. “Automated Logical Observation Identifiers Names and Codes mapping with biomedical natural language processing models: enabling scalable health information exchange via OCL.” JAMIA. May 2026. https://doi.org/10.1093/jamia/ocag010

41 of 43

07

SECTION

Section 07

Lessons Learned and Next Steps

HL7

42 of 43

The Interoperability Equation.

FHIR

FHIR

Structure

+

IGs

IGs

Usage Profiles

+

Terminology

Terminology

Meaning�(OCL)

+

Governance

Governance

Trust�(HL7 + MoH)

=

Digital Public Infrastructure for Health (DPI)

Together, we can build a connected ecosystem.

www.hl7.org.br | terminologia.saude.gov.br | openconceptlab.org

43 of 43

  1. Do you agree with the assertion that semantic governance is what differentiates a national, DPI-H terminology service from a local one? If not (or if insufficient), then what is missing?
  2. Is the 4-layer split of the SGI conceptual framework the right shape? (e.g. governance, publication, runtime, applications) – or should authoring be its own layer? Should publication fold back into governance?
  3. What is the architecture for the semantic governance infrastructure in your country and how does it compare with Brazil and the work-in-progress WHO-ITU DPI-H architecture?
  4. Do you agree with the lack of a unifying semantic governance framework? What role can HL7 as a community take to advance semantic governance?

Questions for Discussion

www.hl7.org.br | terminologia.saude.gov.br | openconceptlab.org