Geriatrician: Accurate AI Can Still Harm Older Patients — SkimNews
Get the Health newsletter
Daily health & science — research, biotech, public health, the studies worth knowing. Free.
- JAMA Network Open published a commentary this month by James Deardorff, a geriatrician and assistant professor in UCSF's division of geriatrics, on a large analysis of Epic's proprietary end-of-life prediction model.
- Deardorff and his co-author argued that an accurate predictive model can still contribute to a poor patient outcome depending on how clinicians apply its output.
- The commentary distinguished low-stakes uses of a one-year mortality prediction — like prompting an open-ended goals-of-care conversation, with few downsides — from higher-stakes uses such as informing transplant priority, which Deardorff and his co-author described as carrying profound impacts.
- Deardorff has developed multiple AI models targeting older adults, including predictions for mortality and the need for nursing home care.
- AI risk-prediction tools for sepsis, falls, and death are increasingly embedded in electronic health records, and Deardorff says clinicians should scrutinize both their performance — including in subgroups like older patients — and how to responsibly use their outputs.
Why it matters: As AI mortality predictions get baked into electronic health records, Deardorff's commentary draws a sharp line between using a one-year mortality score to start a goals-of-care conversation — few downsides — and using the same score to determine transplant priority, which Deardorff and his co-author describe as carrying profound impacts for older patients.
Ask SkimNews
