Business and Clinical
Intelligence on FHIR

Nikolai Ryzhikov, CTO, Health Samurai

Vitalis 2026 · May 7 · Hall F3

1 / 28

What is intelligence?

2 / 28

What is intelligence?

inter + legere = to choose between

The ability to make decisions based on incomplete information about the world

3 / 28

What type of intelligence?

  • Business Intelligence — administrative decisions
  • Clinical Intelligence — clinical decisions
4 / 28
  • +Artificial Intelligence — routine inelligence is getting crazy cheap!
5 / 28

Data → Model → Decision

Better data

↓

Better decisions

6 / 28

What is good data?

Accessible

Data you can't reach is bad data.

7 / 28

Standard format

No standard → mapping & transformation → money, time, errors

Or: you just don't use that data

8 / 28

FHIR gives us all of this

  • Standard model - Resources
  • Terminologies
    LOINC, ICD, SNOMED CT, ATC — shared meaning
  • API
    SMART on FHIR, Bulk Export, Subscriptions...
  • Framework for agreement
    Implementation Guides = profiles on top of resources
9 / 28

Why FHIR?

  • OpenEHR — subset - clinical model, no API, no standard infrastructure
  • OMOP — research clinical subset - could be derivative from FHIR

There are progress with integration of OpenEHR and FHIR Community projects for FHIR to OMOP

10 / 28
FHIR Zulip community — implementers feed
11 / 28

FHIR tools to reflect reality

  • International (IPS)
  • Europe (EHDS)
  • National Core
  • Regional IGs
12 / 28

Future on FHIR

All data accessible through open APIs
in the same semantic format

Plug and Play components, subsystems

13 / 28

FHIR-native systems

Interesting trend - just make a system on FHIR

14 / 28

FHIR is good storage format!

  • generic
  • vendor neutral
15 / 28

FHIR is good for analytics!

Load FHIR into Postgres, ClickHouse, BigQuery…
start writing queries. Works.

But: every DB has its own SQL dialect

Analytics is possible but not interoperable

16 / 28

SQL on FHIR

New standard. Part of FHIR R6 core.

  • ViewDefinition — DSL for flattening FHIR resources
  • Query — standard analytical queries (Library)
  • Runner API — apply ViewDefinitions to data

Written once — runs on any engine:
Postgres, ClickHouse, BigQuery, Spark, DuckDB

17 / 28

ViewDefinition example

Patient → flat table

{
  "resource": "Patient",
  "select": [{
    "column": [
      {"path": "getResourceKey()", "name": "id"},
      {"path": "gender",            "name": "gender"}
    ]
  }, {
    "forEach": "name.where(use='official').first()",
    "column": [
      {"path": "given.join(' ')", "name": "given_name"},
      {"path": "family",          "name": "family_name"}
    ]
  }]
}

Result:

id gender given_name family_name
pt-1maleJohnSmith
pt-2femaleAnnaBerg
18 / 28

SQL query on views

Blood Pressure Trend Report

SELECT
  p.id        AS patient_id,
  p.family_name,
  bp.systolic,
  bp.diastolic,
  bp.effective_date
FROM patient_demographics p
JOIN us_core_blood_pressures bp
  ON p.id = bp.patient_id
WHERE bp.effective_date >= :from_date
  AND bp.effective_date <= :to_date
ORDER BY bp.effective_date

ViewDefinitions create flat tables → standard SQL joins them

Same query works on Postgres, BigQuery, DuckDB, Spark

19 / 28

SoF Implementations

SQL on FHIR conformance test matrix across implementations
20 / 28

Enrich IG

  • Profiles — what data to collect
  • ViewDefinitions — how to flatten for analytics
  • Queries — ready dashboards & reports

App store for analytics:
load IG → get working dashboards

21 / 28

SQL on FHIR API

  • ViewDefinition/$run - realtime flatening
  • ViewDefinition/$export -flatten and export data
  • SQLQuery/$run - execute query
22 / 28
Join us — Analytics on FHIR and SQL on FHIR community sessions
23 / 28

Closing the loop

Deploy intelligence back into care

  • SMART on FHIR — apps inside EHR
  • CDS Hooks — real-time decision support
24 / 28

AI + FHIR + SQL on FHIR

  • LLMs understand FHIR, SQL
  • FHIR and specs is guardrails for AI
25 / 28

The future is near

For routine intelligence

programmers won't be needed

"Show readmission rate by DRG
for last quarter"

Agent writes SQL → builds visualization
on standard data, validatable result

26 / 28

Agents on FHIR

Weekly community meetup — every Thursday

Coding agents + FHIR knowledge bases + skills

  • FHIR skills for agents — IGs, profiles, terminologies
  • Vibe-coding on FHIR — apps generated from text
  • Data pipelines — legacy → FHIR with AI assist
  • CDS & quality measures — agent-written, human-validated

Participants: Google, Clalit, Intermountain,
Health Samurai, Nurset, Microsoft …

chat.fhir.org → #Agents on FHIR · YouTube · Open demos welcome

27 / 28

Thank you!

Nikolai Ryzhikov

niquola@health-samurai.io

health-samurai.io · fhiranalytics.org

LinkedIn QR code — Nikolai Ryzhikov

LinkedIn

28 / 28
1 / 28