Can GPT-3.5 generate and code discharge summaries?
While GPT-3.5 alone given our prompt setting is unsuitable for ICD-10 coding, it supports data augmentation for training neural models. Augmentation positively affects generation code families but mainly benefits codes with existing examples. Augmentation reduces out-of-family errors. Documents generated by GPT-3.5 state prompted concepts correctly but lack variety, and authenticity in narratives.
Authors
- Michael Holder (ORCID: https://orcid.org/0000-0001-9828-9153)
- Hang Dong (ORCID: https://orcid.org/0000-0001-6828-6891)
- Alexandra Birch (ORCID: https://orcid.org/0000-0002-9022-3405)
- Beatrice Alex (ORCID: https://orcid.org/0000-0002-7279-1476)
- Siddharth Basetti (ORCID: https://orcid.org/0000-0002-7639-1859)
- Luke Daines (ORCID: https://orcid.org/0000-0003-0564-4000)
- Aryo Pradipta Gema (ORCID: https://orcid.org/0009-0007-1163-3531)
- Matúš Falis (ORCID: https://orcid.org/0009-0006-7649-6251)
- Rose Penfold (ORCID: https://orcid.org/0000-0001-7023-7108)
Institutions
- National Health Service (GB)
- University of Exeter (GB)
- NHS Highland (GB)
- University of Edinburgh (GB)
Publication Details
- Journal
- Journal of the American Medical Informatics Association
- Published
- 2024-09-13
- DOI
- https://doi.org/10.1093/jamia/ocae132
- Citations
- 29
- Primary Topic
- Machine Learning in Healthcare
- Type
- article
- Field-Weighted Citation Impact
- 9.73
Funders
- Wellcome Trust
- UK Research and Innovation
- National Institute for Health and Care Research
- University of Oxford
- Engineering and Physical Sciences Research Council