Comment nous évaluons le risque d'hallucination de l'IA clinique

L'histoire de preuve de Notat n'est pas un pourcentage de précision magique. C'est une méthodologie transparente : extraire les faits cliniques d'abord, générer les notes à partir de ces faits, montrer le contexte brut au clinicien, et examiner chaque résultat par rapport aux preuves.

Méthode d'évaluation

1. Build the reference facts

A reviewer reads the encounter material and creates a reference set of clinical facts: symptoms, negatives, medication changes, assessment, plan, safety-netting, and coding-relevant details.

2. Compare transcript-direct output

We generate a note from transcript-style input and mark unsupported statements, missing high-salience facts, incorrect attribution, and invented certainty.

3. Compare FactsContext output

We generate documentation from extracted facts and review whether each sentence is supported by a visible fact. The clinician-facing fact list is evaluated as part of the output, not hidden.

4. Record limits, not magic numbers

We do not publish fake accuracy percentages. Each evaluation note includes dataset scope, review date, known limitations, and examples of what the system still requires clinicians to verify.

Ce que les évaluateurs notent

Unsupported clinical assertion

Wrong medication, dose, frequency, or route

Missing red-flag negative or safety-net advice

Wrong diagnosis certainty: possible vs established

Wrong speaker attribution

Unsupported ICD-10 suggestion

Clinician-visible evidence for each key statement

Facts reusable for notes, codes, referrals, and patient instructions

Note interne datée

Note d'évaluation actuelle, 2026-07-01 : cette page documente la méthode et les exemples qualitatifs utilisés pour évaluer l'architecture FactsContext. Elle ne revendique pas de validation externe ni d'exactitude universelle. La prochaine étape de preuve devrait être un examen en aveugle, stratifié par spécialité, avec la taille de l'ensemble de données, l'accord entre évaluateurs et le taux de déclarations non étayées publiés ouvertement.

Extraction de faits vs génération directe à partir de la transcription

Medication change discussed twice

Risque de transcription directe

“Increase amlodipine to 10 mg and stop lisinopril.” The transcript contained a correction: the clinician first considered stopping lisinopril, then decided to continue it after reviewing renal function.

Sortie FactsContext

Facts: amlodipine increased to 10 mg daily; lisinopril continued; renal function normal; review in 6 weeks. Note generated from those facts only.

Pourquoi c'est important

Transcript-direct generation can smooth over corrections. FactsContext preserves the final decision as a discrete fact before writing.

Negative finding matters

Risque de transcription directe

“No neurological symptoms.” The actual encounter only documented no saddle anesthesia and no bladder symptoms; leg radiation was present.

Sortie FactsContext

Facts: left leg radiation to calf; SLR positive left; no saddle anesthesia; no bladder or bowel symptoms. Note keeps the negatives specific.

Pourquoi c'est important

Broad negative statements are risky. The fact list keeps the clinical context granular and reviewable.

Code suggestion requires evidence

Risque de transcription directe

Suggested J44.1 for COPD exacerbation without showing the symptom or treatment evidence.

Sortie FactsContext

Facts: increased breathlessness, purulent sputum, prednisolone burst, antibiotics started. Suggested J44.1 with those facts as evidence.

Pourquoi c'est important

The code is easier to verify because the reason for the suggestion is visible, not buried in prose.

Limites connues

Clinician review remains mandatory. Notat drafts; clinicians verify and sign.

The method reduces unsupported statements by architecture, but no clinical AI should claim zero hallucinations.

Small internal evaluations are useful for engineering direction, not a substitute for external clinical validation.

Specialty, language, audio quality, speaker overlap, and local coding rules can change performance.

Lisez la preuve, puis inspectez vos propres faits.

La méthode d'évaluation est simple parce que le produit est conçu pour être inspectable : les faits d'abord, la note ensuite, l'examen clinique toujours.

Essayer Notat gratuitement