HPB SmartNotes: The impact of artificial intelligence on surgeon workload in the outpatient office

In this study, we aimed to evaluate the feasibility, linguistic accuracy, and coherence of medical notes generated by the integration of an automatic speech recognition system (ASR) and a generative pre-trained transformer (GPT) in an outpatient surgical setting and to assess their impact on documentation time. This prospective exploratory study included 20 adults who visited their first outpatient clinic due to hepatobiliary or pancreatic conditions. Consultations were audio-recorded, transcribed using Whisper-1 (OpenAI), and converted into structured Subjective, Objective, Assessment and Plan (SOAP) clinical notes using ChatGPT 3.5 (OpenAI), implemented through Python 3.11.6. The transcriptions were manually reviewed against gold-standard references, and the ASR performance was quantified using standard metrics from the Jiwer library. Descriptive statistics were computed, and the ChatGPT note quality was evaluated using a standardized checklist, before assessing for statistical significance. The median patient age was 47 years (65 % female). The average interview duration was 15.98 min. Whisper transcribed audio in a median of 59.3 s, and ChatGPT generated SOAP notes in 5.7 s. ChatGPT produced satisfactory results in 85 % of cases, with a significant performance. The ASR system demonstrated acceptable transcription quality. The integration of Whisper and ChatGPT drastically reduced the documentation time compared with manual electronic health record (EHR) entry. Despite some limitations, these results highlight the promise of artificial intelligence (AI)-powered documentation tools in outpatient practice. Further research with larger samples and direct comparisons with traditional methods is warranted to confirm these findings and address the ethical, legal, and implementation challenges associated with AI in clinical environments. • AI reduced EHR documentation time from 15.9 to 1.08 min. • Whisper showed low error rates (WER 0.07, CER 0.04) in Spanish transcriptions. • ChatGPT generated adequate SOAP notes in 85 % of consultations. • Integration improved workflow efficiency and surgeon-patient interaction.