Generative AI in preanesthetic consultations: Effects on efficiency, documentation workload, quality, and...

Clinicians spend over 30% of their workday on electronic health records, reducing patient interaction and contributing to burnout. Preanesthetic consultations demand particularly detailed documentation, making them ideal for generative artificial intelligence (AI)-driven support.

This randomized simulation study evaluated a generative AI application based on a large language model (LLM) designed to automate documentation during preanesthetic consultations. We assessed its effects on consultation efficiency, clinician workload, physician-patient interaction, documentation quality, and user experience.

Thirty anesthesiologists at University Hospital Zurich each conducted two standardized consultations with the same simulated patient, once using the AI tool Isaac (Saipient AG, Zurich) and once with conventional manual documentation. Case order was randomized. The primary outcome was consultation duration. Secondary outcomes included visual attention (eye-tracking), human-computer interaction metrics, subjective workload (NASA-TLX), documentation quality (PDQI-9), self-assessed consultation quality, and workflow preferences.

AI-assisted documentation reduced consultation duration by an average of 252 s (-18%, p < 0.0001), screen fixation (-78%, p = 0.0002), refixations (-73%, p < 0.0001), keyboard input (-87%, p < 0.0001), and mouse clicks (-19%, p = 0.01). Clinicians reported a trend toward lower workload (-16%, p = 0.07) and better patient engagement (median rating 87 vs. 69). However, external raters judged documentation quality to be higher for manual reports (+4 PDQI-9 points; p = 0.004), and clinicians expressed less confidence in AI-generated formatting. Still, 60% preferred AI assistance overall.

LLM-based generative AI-supported documentation significantly improved efficiency and user experience in simulated preanesthetic consultations. While real-world use will require physicians to review and approve AI-generated drafts to ensure documentation quality, the structured outputs may still help reduce typing effort and screen interaction time, although the overall time savings may be smaller in clinical practice due to this additional review step.