AI trong tài liệu lâm sàng

Ambient scribes capture clinician-patient conversations and use speech recognition and language models to draft clinical documentation. A randomized trial of 238 outpatient physicians assigned users of DAX Copilot or Nabla, or usual care, and measured time-in-note; it found a modest reduction in documentation time for one intervention. Clinician reports identified inaccuracies including omissions, formatting problems and pronoun-resolution errors. SOAP is one common structure for notes: Subjective, Objective, Assessment and Plan. Product-specific coding features should be verified in current documentation; Microsoft’s Dragon Copilot support, for example, describes coding-specific prompts for ICD-10 and CPT. Findings from a named product, study and setting should not be generalized to all clinical documentation systems. A simulated-encounter study using four transcript-producing platforms across 14 encounters found that 19.5% (95% CI, 6.6%–28.8%) of transcript errors were transmitted into clinical notes; this result is bounded to that study design.