Vanderbilt Health researchers plan to build an artificial intelligence agent inside their electronic health record system to reduce how long patients with Alzheimer's disease wait before receiving their first infusion of anti-amyloid therapy.
The 18-month, $600,000 project targets the referral from primary care or geriatrics to neurology, the stage where the team believes avoidable delay accumulates. The grant comes from Eli Lilly and Company.
That funding source belongs near the top rather than in a footnote. Lilly manufactures donanemab, one of the anti-amyloid therapies patients would reach faster if the project succeeds. Principal investigator You Chen also leads an ongoing $1 million Lilly-funded project on gaps in obesity care, so this is a continuing relationship rather than a single grant. Speeding patients toward a sponsor's product category is a documented benefit to the sponsor, and readers should hold that alongside the patient benefit the project also describes.
Nothing has been built. Nothing has been validated. This is an announced project with a stated plan, and the distance between a plan and a working clinical tool is where most health AI projects end.
The Delay It Targets Is Real
The clinical problem the project describes is well documented and independent of who is paying for the solution.
Anti-amyloid monoclonal antibodies are approved to slow early cognitive decline in patients with amyloid-confirmed Alzheimer's disease. Qualifying for one is not simple. A patient typically moves through primary care recognition of cognitive concerns, referral to neurology, specialist evaluation, cognitive testing, brain imaging, amyloid confirmation, MRI to check for findings that would make treatment unsafe, and insurance authorization.
Each of those steps has a queue. Weeks pass, and in a progressive disease where the drugs are indicated early, time spent in the pathway is time the treatment window narrows.
Amalia Peterson, an assistant professor of neurology and a co-principal investigator, described the problem from the specialist's end, saying that "by the time a patient reaches our clinic, the clock has often been running for weeks." Co-principal investigator Sean Huang, a geriatrician, described the other end, noting that the pathway often begins in primary care or geriatrics where cognitive concerns are first recognized and the next steps can be difficult to navigate.
What an EHR-Embedded Agent Would Do
The proposed function is narrower than the phrase AI agent suggests, and the narrowness is a point in its favor.
As a clinician enters a referral for cognitive concern, the agent is intended to summarize the relevant chart information, flag missing details that commonly slow evaluation, and recommend whether a case should be routed as priority or standard. Clinicians would retain the ability to accept, edit, or override every recommendation. It is a workflow tool operating on information already in the record.
The team plans first to map where delays accumulate, using AI to reconstruct care timelines from an existing Vanderbilt Health cohort of more than 5,300 patients. They will then identify root causes with input from clinicians and operational staff, and finally deploy and evaluate the triage agent in a pilot. Success will be measured by reduction in time from diagnosis to first infusion.
The project is designed for transferability to other health systems, and the proposal contemplates two further agents along the same pathway: one to analyze MRI images and generate safety reports for radiologist review, and one to compile therapy authorization packets and track the insurance process. Chen, an associate professor of biomedical informatics who leads the project, said the goal is to reduce avoidable delays across the dementia care pathway.
The Questions Worth Asking of Any Tool Like This
Because nothing has been built, the useful contribution now is specifying what would need to be true for it to be trustworthy.
Time to infusion is a process measure rather than a health outcome. A tool can succeed on that metric while producing no benefit in cognition or function, and evaluation should eventually address whether patients who move faster do better.
Selection is the harder question. A triage agent recommends who gets prioritized. If it draws on documentation patterns that differ by race, language, insurance status or clinic type, it can widen access gaps while appearing to improve efficiency. The clinician override provision is a real safeguard, but override rates themselves can vary by patient group. Whether the pilot examines performance across subgroups is worth asking.
False positives carry a real cost. Surfacing patients who then undergo amyloid imaging, MRI and specialist evaluation without qualifying imposes expense and the distress of an incomplete workup.
Independent evaluation matters given the funding arrangement. A pilot designed and assessed by the team receiving the grant is a weaker evidence base than external replication, and the transferability goal makes that more important rather than less.
What Families Facing This Pathway Can Do Now
Practical steps exist regardless of whether this tool works.
Raise cognitive concerns with a primary care clinician early and specifically, describing what changed and when. Vague reports of forgetfulness are handled differently than a concrete account of a person struggling with tasks they previously managed.
Ask directly whether the patient may be a candidate for anti-amyloid therapy and what the evaluation requires. Asking for the list up front lets a family track what has been completed rather than waiting between appointments.
Ask what the insurance requirements are early, since authorization is a common delay point and documentation gathered late has to be gathered anyway.
Understand that these therapies are not appropriate for everyone. They require amyloid confirmation, carry risk of brain swelling and small bleeds requiring MRI monitoring, and are indicated in early disease. A patient who does not qualify has other needs that specialist evaluation can address, including safety planning, caregiver support and treatment of reversible contributors.
MedicalDaily will report whether the tool is built, piloted, and evaluated. This article is general information and is not medical advice.
Frequently Asked Questions
What is Vanderbilt building? An AI triage agent embedded in its electronic health record to speed patients from referral to first Alzheimer's infusion.
Has it been built? No. It is an announced 18-month project. Nothing has been developed or validated.
Who is funding it? Eli Lilly and Company, which manufactures one of the anti-amyloid therapies involved. The project lead also holds a separate $1 million Lilly grant.
What would the agent do? Summarize chart information at referral, flag missing details, and recommend priority or standard routing, with clinician override retained.
How will success be measured? By reduction in time from diagnosis to first infusion.
What are the risks of such a tool? Selection bias across patient groups, false positives leading to unnecessary workups, and a process measure standing in for a health outcome.
What can families do now? Raise cognitive concerns early and specifically, and ask what the full evaluation and insurance requirements involve.
