Balsam Semantica Computable clinical meaning

Grounding Clinical AI in Semantic Integrity

Clinical AI depends on more than the ability to process clinical language.
 It depends on whether the information it receives still means what the clinical encounter established.

Balsam Semantica
is an independent, physician-led practice that reviews how clinical information is represented in machine-readable form, and where that representation stops being faithful to its source.
Asynchronous by design. In writing, with a defined scope, and remote delivery.

[About the Practice →]

THE SEMANTIC GAP

Clinical information can remain readable while losing its meaning

Clinical information is not a collection of words. Its meaning depends on context: who reported it, what was observed, what was concluded, and what remains uncertain.
When these distinctions disappear during documentation, terminology mapping, or AI processing, the record can still be read, and still be wrong. Data that no longer represents the encounter becomes the input for every system downstream.

THE FOUNDATIONAL PRINCIPLE

What Is Computable Clinical Meaning?

Computable clinical meaning is the accurate, context-aware representation of what a clinical encounter established, in a form machines can act on without obscuring what matters clinically.
That includes what the patient reported, what the examination showed, and what the clinician concluded.
It depends on keeping these distinctions intact:
• Patient-reported information versus clinically observed findings
• Clinical findings versus diagnostic conclusions
 • Present conditions versus past medical history
• Positive findings versus negation
 • Certainty versus uncertainty
• Clinical meaning versus its coded representation
The objective is not to make clinical information machine-readable. It is to make its meaning dependable once machines use it.

The instrument

The assessment

CIRI™ - the Clinical Intelligence Readiness Instrument. 

A framework for assessing an organization's readiness to govern clinical AI. It uses 18 metrics across three dimensions:
• Governance • Clinical Practice • Technical and Semantic Readiness
Its purpose is to show whether the organizational, clinical, and technical foundations for responsible clinical AI are in place. CIRI™ is currently in development and not yet validated.

The CIRI™ assessment→  ·  See a sample metric→

18metrics across three conditions
3respondent groups, answering independently
0–5scale, evidenced above the second level
~20ppwritten report, delivered in writing

The guide

Clinical Meaning in the Age of Health Data & AI. A developing guide on the relationship between clinical meaning, health data, standardization, and artificial intelligence. It follows clinical information as it moves from human interpretation into structured data and machine-driven processes.Its central question:
How can health data become computable without losing the clinical meaning it is meant to represent?

Part I - Seeing the Data Layer→  ·  Part III - The AI Layer→ 

Documentation

What the clinician meant is selected, not recorded. Everything that does not fit the form is left out, and nothing records that it was ever considered.

Select a stage below to see what happens to clinical meaning there.

FROM PRINCIPLE TO PRACTICE 

Meaning Must Be Preserved, Not Assumed

"Clinical meaning is not generated by AI. It is captured, governed, and made computable."
The reviews below test that principle where it is most often broken: in product claims, in AI-generated documentation, and in terminology mapping.

a question of meaning

Three Sentences, Three Different Meanings

The three examples illustrated below all center on the same clinical condition, yet they are not interchangeable. Each carries a different source and a different level of clinical interpretation. When an AI system, documentation workflow, or terminology mapping treats them as equivalent, the record loses the very distinctions a clinician would rely on.
Semantic integrity begins here.

  • PolarityNegated
  • CertaintyAsserted
  • TemporalityCurrent
  • SubjectThe patient
  • LateralityNot applicable
  • EpisodicityUnstated
  • PolarityPositive
  • CertaintySuspected
  • TemporalityCurrent
  • SubjectImplied, not named
  • LateralityNot applicable
  • EpisodicityUnstated
  • PolarityPositive
  • CertaintyConfirmed
  • TemporalityTen years ago
  • SubjectThe patient
  • LateralityNot applicable
  • EpisodicityFirst or recurrence, not stated

Writing 

Selected writing on clinical meaning, health data standards, semantic interoperability, and the foundations of trustworthy clinical AI, parts of the guide Clinical Meaning in the Age of Health Data & AI, free written exercises for clinicians, sample reviews and CIRI™  method, are all published here. 

Explore Writing→

Written enquiries about related work are welcome.