Propagation of Interpreter Errors by Ambient AI Scribes: Study Using Simulated Clinical Encounters

Area of Science:

• Medical Informatics
• Clinical Linguistics
• Artificial Intelligence in Healthcare

Background:

• Interpreter errors can impact patient safety and care quality in multilingual clinical settings.
• Ambient AI scribes are increasingly used to automate clinical documentation.
• The accuracy of AI scribes in processing interpreter-mediated communication is not well understood.

Purpose of the Study:

• To evaluate the fidelity of ambient AI scribes in capturing clinical information during interpreter-mediated encounters.
• To identify patterns of AI scribe propagation of interpreter errors based on speaker role and error type.

Main Methods:

• Simulated clinical encounters were conducted in English and Spanish, involving professional interpreters.
• Ambient AI scribe technology was used to transcribe and document the encounters.
• Transcripts and clinical notes were analyzed for the presence and type of interpreter errors and their propagation by the AI scribe.

Main Results:

• Ambient AI scribes propagated interpreter errors into the generated clinical notes.
• The patterns of error propagation varied depending on the speaker role (e.g., clinician, patient, interpreter) and the specific type of error.
• Specific error types, such as omissions or additions, were more likely to be replicated by the AI scribe.

Conclusions:

• Ambient AI scribes may inadvertently perpetuate interpreter errors in clinical documentation.
• Further research is crucial to assess and improve the performance of AI scribes in multilingual healthcare environments.
• Ensuring the accuracy of AI-generated notes in interpreter-mediated care is essential for patient safety and effective communication.