AI diagnoses well, doctors weigh treatment

Andrew Parsons, an associate professor of medicine at the University of Virginia, wrote in a June 1 essay for The Conversation that large language models are becoming “remarkably good” at diagnosis, but that treatment planning remains a different task. His argument was that diagnosis rewards pattern recognition, while management reasoning asks doctors to weigh risks, patient preferences, follow-up constraints and competing clinical goals. That distinction is now showing up in how health systems and researchers are deploying AI: not as an autonomous decider of care, but as a tool for triage, referral and early risk detection. ### If AI can diagnose well, where does it still fall short? Andrew Parsons wrote that making the right diagnosis is only part of a doctor’s job, and that clinicians still hold an edge in deciding what to do next for a specific patient. He described treatment planning as a process that includes uncertainty, trade-offs and context that may not be fully captured in a prompt or a chart. (medicalxpress.com) An April 2026 study cited in later coverage found OpenAI’s o1 model reached 78% accuracy on complex diagnostic cases published in the New England Journal of Medicine, outperforming experienced doctors on that narrow task. But Parsons argued that strong diagnostic performance does not automatically translate into sound management decisions, especially when care depends on patient goals, timing, side effects or access to follow-up. (medicalxpress.com) ### Why is Alzheimer’s referral work a better near-term fit for AI? Vanderbilt Health said on August 5 that Eli Lilly gave its researchers an 18-month, $600,000 grant to build an AI triage agent inside the electronic health record for patients with Alzheimer’s disease. The tool is designed to speed the path from a primary care or geriatrics referral to neurology, where delays can stretch for weeks as patients move through chart review, imaging and insurance authorization. (seattlepi.com) The Vanderbilt system will summarize chart information, flag missing details that commonly slow evaluation and suggest whether a referral should be routed as priority or standard. Clinicians will be able to accept, edit or override every recommendation, according to Vanderbilt and Newswise. ### What does the sleep-EEG “brain age” work actually show? JAMA Network Open published a study in March that pooled five community-based cohorts and found that a higher sleep EEG-based Brain Age Index was associated with higher incident dementia risk. (news.vumc.org) The analysis reported that every 10-year increase in the index was linked to about a 39% higher risk of dementia, independent of chronological age and other established risk factors. (newswise.com) The study covered 7,105 adults and more than 1,000 incident dementia cases across follow-up periods ranging from 3.6 to 16.9 years, according to the journal summary and follow-up coverage. The researchers described the measure as a noninvasive digital marker that still needs further evaluation for predictive use in community settings. ### Does that mean AI is ready to make treatment decisions on its own? (jamanetwork.com) The Vanderbilt project does not hand treatment decisions to software; it targets referral preparation and queueing inside an existing clinical pathway. That design matches Parsons’ argument that AI is strongest when the task is narrow, structured and measurable. The sleep-EEG work also stops short of telling doctors what therapy to choose. (jamanetwork.com) It identifies a risk signal that could help with screening or earlier evaluation, but the published study and follow-up reports said the approach still needs validation across broader populations and implementation work before routine clinical use. ### What should readers watch next? Vanderbilt Health said the Alzheimer’s referral project is intended to be transferable to other health systems after it is built inside its own electronic record environment. (newswise.com) The next visible milestones are development and deployment of the triage agent during the 18-month grant period, with Eli Lilly and Vanderbilt as named participants. (news.vumc.org) (jamanetwork.com)