Anamnesis Was Never About Documentation

The most diagnostic thing a patient does

The endurance of the interview is not sentiment, and it is not tradition for its own sake. It is one of the most reproducible findings in clinical medicine.

In 1975, Hampton and colleagues followed medical outpatients and recorded how each diagnosis was actually reached. In 66 of 80 patients, around 82%, the history alone was enough to arrive at the correct diagnosis. The physical examination and every laboratory investigation combined accounted for the remaining fraction. The study has been repeated across decades and settings, and the result barely moves. A 1992 replication put the history at 76% of final diagnoses. The proportion shifts a little by specialty, but the conclusion is stable: most of the time, the diagnosis is already present in what the patient tells you, before an instrument is picked up and before a test is sent.

Think about that for a minute. The single most powerful diagnostic tool in medicine is what the patient tells you. Not the labs you ran, or the images you took. The story told.

The quintessential diagnostic act

This is why anamnesis is not a preface to diagnosis. It is diagnosis. When a clinician chooses the next question, follows a thread, notices the symptom the patient almost left out, that is diagnostic reasoning happening in real time. The history is not the paperwork that precedes the clinical thinking. It is the clinical thinking.

The signal is in the story the patient tells, and in the skill of the clinician who draws it out. That is not a soft claim. It is measurable, and it has been measured for fifty years.

There is a quiet acknowledgment of this in how these tools are regulated. In Europe, an AI scribe intended to support clinical care carries a medical device intended purpose, whether it lands in Class I or Class IIa. Capturing the clinical encounter is not treated as clerical work under the law. It is the acquisition of clinical information, in the same family as a thermometer acquiring temperature or a stethoscope acquiring a heart sound. The anamnesis is a diagnostic acquisition tool, indispensable and irreplaceable, and the regulation already recognizes that taking the history is a clinical act. The strange part is that the products built on top of that recognition still behave as if it were only paperwork.

A great start, and only a start

First, credit where it is due. The industry has done remarkable work, and we count ourselves in it. AI scribes are genuinely relieving physician burnout, giving clinicians their evenings back, letting them look at the patient instead of the screen. That is not a small thing. Beyond the efficiency, these tools have made clinicians smile again, and anyone who has watched a doctor stop dreading their notes knows how real that is. This was a hard problem and the field solved a large part of it.

My take is simple: this was the first mile, not the whole marathon.

The first great wave of clinical AI has organized itself almost entirely around one use of the anamnesis: writing it down. The ambient scribe listens to the encounter and produces a note, with the codes attached for reimbursement. That is the whole product, and it is a large and growing industry. The pitch is documentation relief and cleaner billing, and both are real needs, well met.

But look at where the value has gone so far. This technology has already unlocked real gains, cleaner documentation, and a whole set of revenue cycle management wins that health systems genuinely needed. That is value well extracted. What has not been built yet is the diagnostic layer. The same anamnesis that feeds the note and the billing carries the intelligence that makes the history worth 80% of diagnoses, and that intelligence is right there, ready to power diagnostic solutions too. Documentation and revenue were the first things this data delivered. Diagnosis is the next, and it runs on the exact same signal.

A question for the diagnostics industry

There is a strange silence worth pointing at. The scribe companies are, at their core, documentation companies. And the medical diagnostics companies, the ones building genuine diagnostic intelligence, have largely stayed out of the scribe race entirely.

Why?

The anamnesis is the fuel their tools run on. A diagnostic model is only as good as the clinical context it is given, and the richest source of that context in the entire encounter is the history. An imaging model reading a scan, or a model interpreting a lab panel, is working in isolation from the very information that gives a finding its meaning. The same shadow on an image or value on a panel points in different directions depending on what the patient said. The history is precisely the context that would make those models more accurate.

Radiologists and pathologists have always known this. They routinely ask for clinical history before signing out a study, because a scan or a slide is often ambiguous without it. The request for clinical correlation is not a formality. It is an admission that the finding cannot be fully interpreted without the anamnesis. If the human experts need the history to read the image, so does the model built to assist them.

If the diagnostics industry understood the history the way the studies describe it, they would be racing to capture it, because it is the input that makes everything downstream better. That they have left the field to documentation vendors suggests the whole industry has accepted the same false split: documentation over here, diagnosis over there, and the anamnesis quietly miscategorized as belonging only to the first.

It belongs to both. It always did.

Bringing the clinical act back together

We believe in integration and in purpose. So we built the opposite of the split.

Keikku binds the anamnesis, the structured note, and the rest of the encounter to diagnostic intelligence. It starts as software alone, so any clinician can bring documentation and diagnostic intelligence together from day one, on the devices they already have. And in our Complete subscription, that same clinical intelligence binds to our own hardware, an FDA-cleared medical device that captures the conversation with the highest ambient performance, and captures something no other ambient device can, the physiology of the body through auscultatory sound. The anamnesis and the physical exam, documentation and diagnosis, brought together into one clinical intelligence rather than scattered across separate tools. One captured clinical act, because that is how the clinician always experienced it and how medicine always worked, until the software decided otherwise.

The interview has been the backbone of medicine for centuries. The task now is not to reduce it to a transcript for a billing system. It is to carry it whole, its record and its reasoning intact, into the AI revolution medicine is finally having. Anamnesis was never about documentation. It is time our tools remembered that.