Denver patients turn to AI for medical advice

For months, Ben chased a diagnosis through Google Gemini and ChatGPT.

“You have meningitis.” “It could be vertigo.” “This looks like multiple sclerosis.” 

Panicked by what he was seeing from artificial intelligence, Ben booked appointments with cardiologists, ear, nose and throat specialists and balance specialists. None of them found anything unusual. When Ben finally saw a neurologist, he was told what no AI search had told him during his scramble for answers: It was anxiety.

“The only place that diagnosis ever came up was with friends and family that have dealt with the same kind of symptoms,” Ben said. “And the neurologist was just able to kind of tie that together.”

For his wife, Alexa, AI was more useful. During a high-risk twin pregnancy, she used ChatGPT to prepare for appointments, ask about symptoms and understand what treatments or injections she might need before giving birth.

“What I like about using ChatGPT is that it seems like it does a more thorough internet search for me,” Alexa said. “So yes, it does save time by having to read a million articles. Like it at least seems like it scans all of the information and then uses all of that information to kind of condense into an answer for you.”

Their different experiences capture a tension doctors say they are now navigating in exam rooms: AI can help some patients ask better questions, but it can also send others into unnecessary panic. A Tebra survey of 500 health care professionals and 1,000 Americans found that eight in 10 Americans believe AI has the potential to improve care quality, reduce costs and expand access to health care.

Dr. Daniel Kortsch, a practicing physician and double board-certified clinical informatics specialist, is associate chief of artificial intelligence and digital health at Denver Health. After spending years integrating tools such as Epic into clinical care, Kortsch is now on the front lines of how AI is changing conversations between doctors and patients.

“AI is really good with big data, and it’s really good with synthesizing things and finding summaries and stuff like that,” Kortsch said. “But the free versions and just like easy Google search, both of them kind of fall when you talk about analysis. And so it’s the analysis component that often is the most difficult to search online, either using AI or not. Analysis is a real challenge.”

For the Denver-area couple, AI became a tool for shaping conversations with doctors. Alexa used it “every day” to seek advice on her pregnancy and what to expect later.

When the couple wondered about her odds of developing preeclampsia before 28 weeks, they did not wait to ask her doctor. They asked ChatGPT first and brought the information they found into conversations at the hospital.

“When we’re brainstorming with the doctors, especially about the babies and stuff, or with her pregnancy, like we would bring up other points that could have been, you know, on the horizon of possibilities,” Ben said. “And honestly, I feel like 90% of the time, like medical staff and doctors would kind of shoot that down.”

But when Ben used Google Gemini to figure out what was happening with his own health, the AI-generated possibilities pointed him away from anxiety and toward more frightening conditions.

“When you Google things and you use ChatGPT, they tell you the worst of everything,” Ben said. “It kind of takes you down a darker path, just because I feel like it’s so broad. Like so many symptoms can be so many different things.”

His experience shows what can happen when patients ask AI to do too much. For Alexa, the technology proved most useful in a narrower role: translating dense, jargon-heavy medical charts and records into language she could understand.

Patients are not the only ones using AI to navigate medical conversations. Doctors at Denver Health are also using AI tools to assist with patient interactions, documentation and clinical decision-making.

According to Kortsch, Denver Health currently has more than 40 generative AI tools, with many of them being deployed through Epic Systems, a major electronic health record software vendor. Kortsch’s job is to enable those tools within Denver Health and give doctors the opportunity to use them if they choose to.

One model Denver Health uses creates a digital twin of a patient, using characteristics such as age, gender, conditions and medications to compare that patient with a database of more than 300 million patients nationwide. The tool can help doctors predict how a patient might respond to a treatment decision, such as starting or skipping a medication, based on outcomes from similar cases.

Kortsch said models like that were “literally science fiction” four years ago.

Another AI tool clinicians use is ambient AI, which listens to conversations between doctors and patients before summarizing what was discussed during the appointment. The tool can reduce the time clinicians spend writing notes during or after appointments.

“Ambient AI is one of those tools that is just really, really good and improves provider efficiency and reduces burnout,” Kortsch said. “And we recently studied over 16,000 different medical encounters and found that patient satisfaction also significantly increases when providers use ambient AI in their encounter because it allows them just to end up looking their patient in the eyes, and I think that’s really exciting.”

While AI tools are moving quickly, Kortsch said the models are “not perfect” and still “require a significant amount of human oversight.” He said he is “constantly” spotting errors in some of the AI models he uses and would be worried if people used them to replace clinical thinking.

The most promising uses, he said, are not about letting AI diagnose patients on its own. They are about helping patients and doctors have clearer conversations.

Ben and Alexa said they are comfortable with doctors using AI to analyze data and predict trends as long as their “team of doctors with years of experience” is not relying on it to diagnose them.

As Denver Health and other hospitals continue to test AI tools, the question is not only whether the technology will keep advancing. It is whether patients and doctors can use it without losing the human connection at the center of health care.

“I think, at the patient’s perspective, their ability to come to their provider and not overwhelm them with 10,000 points of data, but instead provide them with an analysis that is founded in a coherent, safe analysis that is free from the bias that sometimes exists in some of our data sets right now, I really look forward to the day when that can happen,” Kortsch said. 

Alexa said that balance is what matters most.

“The human aspect of health care is very important, so AI should never replace that,” she said. “But I think they can work together.”