Low responsiveness of machine learning models to critical or deteriorating health conditions
Using generated test cases, we find that statistical machine-learning models trained solely from patient data are grossly insufficient and have many dangerous blind spots. Most of the ML models tested fail to respond adequately to critically ill patients. How to incorporate medical knowledge into clinical machine learning models is an important future research direction.
Authors
- Xinwei Deng (ORCID: https://orcid.org/0000-0002-1560-2405)
- Tanmoy Sarkar Pias (ORCID: https://orcid.org/0000-0002-7325-9844)
- Danfeng Yao (ORCID: https://orcid.org/0000-0001-8969-2792)
- Charles B. Nemeroff (ORCID: https://orcid.org/0000-0001-7867-1160)
- Sharmin Afrose (ORCID: https://orcid.org/0000-0003-1835-773X)
- Ipsita Hamid Trisha
- Moon Das Tuli
Institutions
- Oak Ridge National Laboratory (US)
- University of Arizona (US)
- Banner - University Medical Center Tucson (US)
- Virginia Tech (US)
- The University of Texas at Austin (US)
- Dhaka Medical College and Hospital (BD)
Publication Details
- Journal
- Communications Medicine
- Published
- 2025-03-11
- DOI
- https://doi.org/10.1038/s43856-025-00775-0
- Citations
- 6
- Primary Topic
- Machine Learning in Healthcare
- Type
- article
- Field-Weighted Citation Impact
- 14.89