Tai-Seale AI draft replies
2024
AI-Generated Draft Replies Integrated Into Health Records and Physicians' Electronic Communication
In 52 primary care physicians, generative-AI drafted replies to patient messages did not reduce physician time spent reading and replying to messages compared with delayed (waiting-list) activation and contemporary controls.
Result
Primary endpoint: Time spent reading and replying to messages, reply length, and likelihood to recommend
Read time +21.8% (95% CI 5.2% to 41.0%; P=.008); reply time −5.9% (95% CI −16.6% to 6.2%; P=.33); reply length +17.9% (95% CI 10.1% to 26.2%; P<.001). 25 physicians randomised to immediate and 27 to delayed activation.
- Subtopic
- Draft message replies
- Design
- Randomised waiting-list quality-improvement study, single academic health system
- Relevance
- Supporting: context rather than must-know.
- Source
- Tai-Seale M, et al. JAMA Netw Open. 2024;7(4):e246565.doi 10.1001/jamanetworkopen.2024.6565
Discussion & critique
An early, honest test of the most widely deployed generative-AI feature in electronic health records: drafts made physicians read for longer and write longer replies without saving reply time, although participants saw value in them. A small quality-improvement study run from June to August 2023 with a waiting-list randomisation and a non-randomised contemporary control, and no patient-experience outcome; the authors themselves call for rigorous trials.