The questions a doctor asks a patient and their family are particularly important for a diagnosis in Alzheimer’s disease and related dementias, but the demand for care is rapidly outpacing the supply of clinicians, and the time to truly listen keeps shrinking. My colleagues and I have been testing a possible way to protect that conversation: a voice-based artificial intelligence “agent” that interviews patients before they ever see a clinician, so the clinician can start the visit with the story already in hand.
After spending an hour talking with our voice-based artificial intelligence interviewer, one of our study participants said it was the first time she felt her symptoms were fully heard. To me, as a neurologist, that was a gut punch. In my own clinic at Emory University in Atlanta, I sometimes have to cut patients off and redirect just to cover everything in the time we have. I’ve heard seasoned clinicians lament that they still sometimes realize at the end of a busy day when writing their notes that they forgot to ask an important question. Everybody is human, and sometimes the right question just doesn’t get asked.
Sir William Osler, often called the father of modern medicine, put it simply: “Listen to your patient; they are telling you the diagnosis.” That is still the first step in diagnosing most medical conditions.
Over one hundred years after Osler’s famous words, large language models reached the public in late 2022 through ChatGPT to enable fluid, adaptive conversations with machines, which raised the question: Could they be tasked with asking these questions before a visit? This eventually led me to Jinho Choi, an expert in natural language processing at Emory University. Together, with funding from the American Brain Foundation, the Alzheimer’s Association, and Eisai, we built a voice-based conversational “agent” that people experiencing cognitive decline can talk with before they see a neurologist.
Real physician voices, twice a week
Free, and one click to unsubscribe.
The symptoms that change a diagnosis or a care plan in Alzheimer’s disease and related dementias can be nuanced, and patients and families rarely bring them up on their own; the clinician has to ask. What counts as a “memory problem” differs from one person to the next, so we have to draw out specific examples to understand what is really happening. Skilled interviewing is part of the art of medicine: following the answer, knowing the adjacent question, and still ensuring all essential questions are covered in the time we have.
A thorough interview is what makes an early diagnosis possible, and an early diagnosis can do a lot. It improves quality of care, determines eligibility for new therapies that can slow the disease, and opens the door to research trials working toward a cure. It also connects families to programs early on that allow patients to remain in their homes longer and avoid costly emergency room visits.
Interviewing expertise is earned over thousands of interviews, and most of it stays locked inside individual doctors’ heads. One study of roughly 100,000 clinical note templates, which are often personal checklists doctors build to remind themselves what to ask, found that only 17 percent were ever used by more than one person. That is a lot of hard-won knowledge that never gets shared.
This initially led me to create note templates with subspecialist-designed questions and test them with neurology residents. While they improved documentation of quality measures, there were too many questions to cover in time-limited visits. We explored medical staff asking them before the visit, but no one had the bandwidth (health care workforce shortages run well beyond clinicians).
The agent works over any ordinary telephone (no app, smartphone, or digital know-how required). That matters, because there is often less access to care for people with dementia who are rural and have less digital literacy. The agent interviews the patient, and separately interviews a family member or care partner, whose observations are often essential in dementia. It asks questions drawn from a panel of dementia specialists, and it probes when an answer is unclear.
To our surprise, most patients in our pilot study enjoyed it. The agent never rushes them, and 89 percent reported being satisfied. Participants typically spent about an hour with the agent, about twice as long as the specialist interviews we recorded. The agent also covered more of the symptoms specialists consider important than time-pressed clinicians were able to.
The agent does not diagnose. It takes the history and produces an organized summary for the clinician to verify with the patient and their family. The clinician remains the decision-maker. We are preparing for a randomized clinical trial to rigorously test whether this holds up in practice, because technology like this should earn its place in medicine with evidence.
I’m particularly excited about the improvements that can be made over time. When one clinician learns a better question, that knowledge usually stays with that one clinician. But there is only one agent, and every improvement reaches every patient it interviews, whether they live near a major academic medical center or hours from the nearest neurologist. One day, it could reach patients in any language.
My hope is that with the agent’s interviews in hand at the start of a visit, I’ll have the gift of time: time to build a better relationship, to steer care toward what my patients value, and to be confident that no meaningful part of their story goes unheard.
Machines helped create medicine’s time crunch. Used wisely, they may help give that time back.
This essay was developed with editorial support from Andrew Smith, with funding support from the American Brain Foundation.
Drew Breithaupt, MD, is a cognitive neurologist and assistant professor at Emory University School of Medicine, where he conducts research as part of Emory’s Goizueta Brain Health Institute.
The American Brain Foundation funds research across the spectrum of the brain and nervous system because it views that research as the best hope for improving treatments, advancing prevention strategies, and discovering cures. The foundation’s work rests on the understanding that brain diseases, disorders, and injuries are interconnected, so a cure for one will lead to cures for many.
The American Brain Foundation shares updates on Facebook, X, Instagram, and YouTube.
