James Deardorff, a geriatrician at the University of California San Francisco, urges clinicians to carefully scrutinize artificial intelligence tools used for older adults. In a September 2026 JAMA Network Open commentary, Deardorff warns that while Epic's proprietary end-of-life prediction model can prompt useful care conversations, relying on such algorithms for high-stakes decisions like transplant priority carries profound risks to patient outcomes.
AI prediction models
- ▪James Deardorff at the University of California San Francisco has developed multiple artificial intelligence models designed to predict health outcomes for older adults.
- ▪Medical professionals are increasingly utilizing artificial intelligence models to predict patient risks, including sepsis, falls, and mortality.
Geriatric patient populations
- ▪James Deardorff, a geriatrician at the University of California San Francisco, developed models to predict mortality and the need for nursing home care among older adults.
- ▪James Deardorff stated that clinicians caring for older patients must understand how algorithms perform within specific age subgroups and how to use their outputs responsibly.
Epic mortality prediction algorithm
- ▪James Deardorff's commentary in JAMA Network Open highlighted that Epic's proprietary end-of-life prediction model could lead to poor patient outcomes even if the algorithm is accurate.
- ▪In September 2026, James Deardorff published a commentary in JAMA Network Open analyzing Epic's proprietary end-of-life prediction model.
Clinical decision-making risks
- ▪Using a patient's one-year mortality prediction for high-stakes clinical decisions, such as determining organ transplant priority, can result in profound negative impacts, according to James Deardorff.
- ▪Using a patient's one-year mortality prediction to guide open-ended conversations about care goals carries minimal downside, according to James Deardorff.
EHR-embedded AI tools
- ▪Artificial intelligence prediction models are frequently integrated directly into electronic health records, making them easy for clinicians to incorporate into daily care.
- ▪The ease of integrating artificial intelligence models into electronic health records can lead clinicians to take the performance of these algorithms for granted.
Story comments
Loading comments…