AI-Assisted Medical Documentation in a Multilingual Swiss Health Care System: Proof-of-Concept Study.

Area of Science:

• Medical informatics
• Clinical documentation
• Artificial intelligence in healthcare

Background:

• Physician administrative burden impacts patient care time.
• AI tools like speech recognition and LLMs offer solutions but lack multilingual performance data.
• Multilingual environments, like Switzerland, pose documentation challenges for non-native speakers.

Purpose of the Study:

• To compare AI-assisted and traditional clinical documentation workflows.
• To evaluate efficiency and quality in a linguistically diverse Swiss hospital.
• To assess performance for both native and non-native German-speaking physicians.

Main Methods:

• Four workflows tested: traditional dictation, speech recognition, AI transcription/processing, and AI ambient dictation.
• Two physicians (native and non-native German speakers) documented simulated patient encounters.
• Efficiency measured by physician time; quality assessed by LLM scoring (PDQI-9).

Main Results:

• AI-assisted ambient dictation (workflow 4) yielded the shortest documentation times.
• Workflow 4 was significantly faster than speech recognition (workflow 2) for both physicians.
• For non-native speakers, workflow 4 showed a non-significant time improvement over traditional dictation (P=.08).

Conclusions:

• AI-assisted documentation offers potential time savings, particularly for native speakers.
• AI tools may help alleviate linguistic challenges for non-native speaking physicians.
• LLM-based quality scoring lacks reliability; human evaluation is crucial for validation.