Linguistic Effects of Ambient AI on Clinical Documentation: A Matched Pre-Post Study

Area of Science:

• Clinical Informatics
• Natural Language Processing
• Health Informatics

Background:

• Ambient intelligence systems are increasingly adopted for clinical documentation.
• Quantifying linguistic changes in notes generated by these AI systems is crucial.

Purpose of the Study:

• To analyze linguistic differences in clinical notes before and after implementing ambient AI documentation systems.
• To compare the impact of different ambient AI vendors (Nuance DAX and Abridge) on note structure and language.

Main Methods:

• A matched pre-post analysis of 6,026 outpatient clinical notes was performed.
• Linguistic features including syntactic complexity, lexical ambiguity, coherence, and readability were evaluated.
• Automated analyses were validated through manual review of 50 paired notes.

Main Results:

• Ambient documentation resulted in longer History of Present Illness (HPI) notes with increased syntactic complexity and discourse coherence.
• Lexical ambiguity decreased, and language-model perplexity was lower in ambient HPI notes.
• Assessment and Plan (A&P) sections showed smaller, more heterogeneous linguistic changes.
• Effects were vendor-dependent and section-specific, with HPI showing more significant standardization.

Conclusions:

• Ambient AI documentation systems systematically alter clinical note linguistics, leading to more elaborated and predictable narratives.
• The impact varies by note section (HPI vs. A&P) and AI vendor.
• Future evaluations should consider broader impacts beyond efficiency, such as communication and cognitive load.