Evaluation of an AI Scribe Tool in the Emergency Department: A Single-Arm Observational Study.
Generative artificial intelligence (AI) is reshaping the way clinicians record their clinical notes. AI-scribe systems leverage generative AI capabilities to transcribe clinical encounters into draft clinical notes. In this study, we assessed clinician uptake and estimated modelled documentation time savings for an AI-scribe system in an emergency department (ED).
ED physicians and trainees were provided access to an AI-scribe for 5 weeks. Data from the first week were excluded. The transcript of each presentation, the initial AI-generated clinical note and the final EMR clinical notes were used to calculate time to finalise AI-assisted notes.
Forty ED consultants and 23 trainees accessed the AI-scribe. Over the study period, nine consultants (22.5%) and 11 registrars (48%) used the system. The AI-scribe was used for 248 ED presentations, including 185 (74.6%) by trainees and 63 (25.4%) by consultants. The system generated 44,489 words. Following clinician review, 18,140 words were added and 2274 were deleted prior to submitting the final clinical notes. For a clinician with an average typing speed, use of the AI-scribe resulted in a time saving of 7.1 h of documentation. This was reduced to 4.9 h for rapid typer clinicians. Overall, the initial AI-generated notes were modified on 1143 occasions. The most frequently revised section was the history of presenting illness (23.3%) followed by the management plan (22.9%).
Uptake of AI-scribe was higher among trainees than consultants, and the platform achieved substantial time savings. Future studies are required to quantify real-time productivity gains over longer periods.
ED physicians and trainees were provided access to an AI-scribe for 5 weeks. Data from the first week were excluded. The transcript of each presentation, the initial AI-generated clinical note and the final EMR clinical notes were used to calculate time to finalise AI-assisted notes.
Forty ED consultants and 23 trainees accessed the AI-scribe. Over the study period, nine consultants (22.5%) and 11 registrars (48%) used the system. The AI-scribe was used for 248 ED presentations, including 185 (74.6%) by trainees and 63 (25.4%) by consultants. The system generated 44,489 words. Following clinician review, 18,140 words were added and 2274 were deleted prior to submitting the final clinical notes. For a clinician with an average typing speed, use of the AI-scribe resulted in a time saving of 7.1 h of documentation. This was reduced to 4.9 h for rapid typer clinicians. Overall, the initial AI-generated notes were modified on 1143 occasions. The most frequently revised section was the history of presenting illness (23.3%) followed by the management plan (22.9%).
Uptake of AI-scribe was higher among trainees than consultants, and the platform achieved substantial time savings. Future studies are required to quantify real-time productivity gains over longer periods.
Authors
Akhlaghi Akhlaghi, Freeman Freeman, Sun Sun, Nie Nie, Ding Ding, Chen Chen, Pham Pham, Morrissey Morrissey, Karro Karro
View on Pubmed