STAT+: A geriatrician explains why AI for older adults deserves careful scrutiny

James Deardorff, a geriatrician and assistant professor in the division of geriatrics at the University of California San Francisco, published a commentary in JAMA Network Open this month on a large analysis of Epic's proprietary end-of-life prediction model, arguing that an accurate algorithm can still contribute to poor outcomes depending on how its output is used. Deardorff and his co-author wrote that using a patient's one-year mortality prediction to prompt an open-ended conversation about goals of care carries few downsides, but using it to inform a higher-stakes decision such as transplant priority could have profound impacts. Deardorff, who has developed several prediction models for older adults covering mortality and the need for nursing home care, says clinicians must be aware of an algorithm's performance in subgroups like older patients and of how to responsibly use its output.

Across medicine, doctors are increasingly using artificial intelligence models to predict patient risk, from sepsis to falls to death. Often embedded directly into electronic health records, these predictions can be easy to incorporate into care — and their performance can be easily taken for granted. James Deardorff, a geriatrician and assistant professor in the division of geriatrics at the University of California San Francisco, has developed several models that aim to make predictions for older adults, from mortality to the need for nursing home care. In geriatrics, he says, it’s important for clinicians to be aware of both the performance of an algorithm — including in subgroups like older patients — and how to responsibly use its output. This month, Deardorff penned a commentary https://jamanetwork.com/journals/jamanetworkopen/fullarticle/2853923?resultClick=3 on a large analysis https://jamanetwork.com/journals/jamanetworkopen/fullarticle/2853919 of Epic’s proprietary end-of-life prediction model published in JAMA Network Open, highlighting how a model can contribute to a poor outcome even if it’s accurate. If a patient’s one-year mortality prediction is used to prompt an open-ended conversation about the goals of care, he and his co-author wrote, there are few downsides. But if it’s used to inform a higher-stakes decision like transplant priority, the impacts could be profound. This article is exclusive to STAT+ subscribers Unlock this article — and get additional analysis of the technologies disrupting health care — by subscribing to STAT+. Already have an account? Log in https://www.statnews.com/login/ View All Plans https://www.statnews.com/stat-plus/