Balsam Semantica Computable clinical meaning

Exercise · about twenty minutes

What a Coder Can Establish

A twenty-minute exercise for clinicians

The third in the series, and the only one that contains no AI at all.

The first two put a system-generated note in front of you and asked what was wrong with it. This one puts two human-written notes in front of you, for the same patient, with the same treatment and the same outcome - and asks a different question.

Not is this note good? Both are good. The question is: what can somebody else establish from it?

That question is usually treated as an administrative one. It is not. Everything that reads the record afterwards - a coder, a registry, a quality programme, a payer, a model - inherits the answer.

Part 1

The admission

A man of seventy is admitted from the emergency department with fever, productive cough and tachypnoea. Imaging shows a right lower lobe infiltrate.

Over the first hours: hypotension responding to fluids. Creatinine rising above his baseline. Confusion noticed by the nursing staff before the medical team. Raised lactate.

Antibiotics are started. He improves. He goes home after six days.

Part 2

Two notes

Both were written by competent physicians. Both describe what happened. Neither contains anything untrue.

Note A

Community-acquired pneumonia. Improved on treatment. Discharged day six.

Signed: ____________________

Note B

Sepsis secondary to community-acquired pneumonia, present on admission, meeting criteria on arrival (documented hypotension responsive to fluid resuscitation, lactate 3.1, altered mental state).

Acute kidney injury, stage 1, present on admission - creatinine 148 against a baseline of 92 from February. Resolved by day four.

Acute encephalopathy secondary to sepsis, present on admission, resolved by day three.

Right lower lobe pneumonia confirmed on chest radiography. Improved on treatment. Discharged day six.

Signed: ____________________

Part 3

The questions

Do not ask which note is better written. Ask:

  1. What can a coder establish from Note A?

    Remember that a coder may not infer. They may only code what is documented, even where the laboratory values sit in front of them.

  2. What can a coder establish from Note B that they cannot from Note A?

  3. What do the two resulting records say about this patient

    To someone reading them in three years?

  4. Which of the two patients looks sicker

    And which of them actually was?

  5. Where was the difference decided?

    Not by the coder, and not by the clinician who wrote Note A carelessly. Neither of them was careless.

Write your answers before continuing.

Part 4

The key

Do not read this until you have written your answers.

Part 5

Why this is not an administrative matter

Everything in this exercise is usually filed under coding, or billing, or documentation improvement - that is, under the parts of a hospital that clinicians are encouraged not to think about.

Follow what happens next.

These records are stored. They are aggregated. They become the denominators of quality indicators, the population in a registry, the case mix of a service, the evidence in a safety review, and eventually the training and evaluation data for models built on the institution's own history.

A model trained on records assembled this way learns what was establishable, not what was clinically true. It is then measured against a reference standard built by the same process, so it scores well while reproducing a loss that happened years earlier, in a note nobody has read since.

The revenue cycle is the one place in a hospital where the loss of clinical meaning is measured every single day - because a third party disputes it, in cash. Most institutions read only the cash value.

One question for your institution

Does your organisation measure the difference between what clinicians documented and what coders were able to establish - as a data point, aggregated and reported, rather than resolved case by case and discarded?

If it does not, it is discarding the most detailed audit of its own clinical documentation that it will ever have, and it is discarding it daily.

This exercise draws on Chapter 5 of Clinical Meaning in the Age of Health Data & AI. The clinical case is composite and illustrative, and is not drawn from any real patient record.

Clinical meaning is not generated by AI. It is captured, governed, and made computable.