Readers respond to the growing use of ambient AI in doctor's offices.
You're reading The Checkup With Dr. Wen, a newsletter on how to navigate medical and public health challenges. Click here to get the full newsletter in your inbox, including answers to reader questions and a summary of new scientific research.
This week, I wrote about how doctors are using artificial intelligence to help document their visits with patients. Many clinicians wrote in to say they are just as excited about this transformative technology as I am.
"I'm not a Luddite, but I would hardly describe myself as a technophile," wrote Randy, a family physician from North Carolina. "I wasn't sold on the idea of an AI scribe until partners in my practice began raving about it. It's remarkable. I can focus completely on my patients, look them in the eye and have a real conversation instead of staring at my computer clicking boxes. It takes me maybe 30 minutes, maximum an hour, to finish my notes instead of two to three hours."
Sandy, an internist in Michigan, is in a two-physician marriage. He and his wife used to spend hours after work completing their documentation. They now finish sooner and have more time with their children. Diane, a rheumatologist in Maryland, was going to take early retirement partly because she was burned out from all the paperwork. She is reconsidering because her workflow has improved so much.
"You wrote about physicians having this tool to help their workload," wrote Jeneen in D.C. "What about nurses? Is there something on the horizon to help us?"
To help answer this question, I spoke with Christopher Sharp, a physician and chief medical information officer at Stanford University Medical Center. He said there are many efforts underway to assist with nursing documentation, such as automating the flow sheet, which nurses use to keep track of a patient's vital signs, pain levels and other key data. "Nurses, just like other clinicians, spend a lot of time discussing with a patient then transcribing elements into their documentation. So we think there's a huge opportunity for nurse efficiency as well."
Sharp identified three other areas where generative AI can help health workers. The first is interpreting laboratory tests. Patients often learn of their test results through electronic medical record notifications. Certain numbers might be flagged as abnormal, and until the patient can communicate with their physician, they are in the dark about those warnings' significance.
AI tools can draft language that helps clinicians explain lab results to patients. The technology can also put results in the context of patients' medical history, such as what medications they are taking and whether they have diabetes. A doctor still must check that language, but it could provide patients with faster information and reduce clinicians' workload, Sharp said.
The second is explaining billing. "Patients write to us and say, 'Why was this charged? I have a co-pay. Why didn't my co-pay cover this?'" Sharp said, so his hospital is testing AI to respond to these questions. Again, a human must check all responses before sending them to patients.
The third is help with messaging. This was something Kristin, a preventive cardiologist from Virginia, brought up: "My hospital has integrated AI into patient messages we receive through electronic health records," she wrote. "I've found the AI drafts to be consistently very good. It goes into more detail than I would normally write, and it saves me a lot of time. (I estimate as much as an hour a day.)"
Sharp said that during the pandemic, many more patients started emailing their providers with questions they previously would have asked at appointments. This "created a burden on our teams to respond appropriately and with empathy and accuracy," he said. His health system is among many that have started using AI to draft physician responses.
The technology has improved so much that it can now pull in patient data in responding to queries. For instance, Sharp explained, "when the patient says, 'In my last visit, my doctor told me I need to change a medication, but I can't remember what that is,' the draft will naturally surface that information."
This saves time, but it also alleviates what Sharp refers to as cognitive burden. "We all know how hard it is to write an important email," he said. "Now imagine doing that dozens and dozens of time a day." He told me Stanford's internal research shows that the AI-generated responses tend to be longer and more empathetic than what clinicians would write.
Anthony from Maryland is still unsure about this technology. "I'm retired from clinical practice, but my son is a doctor who raves about AI improving his work-life balance," he wrote. "Here's my worry: When administrators see that doctors are more efficient, won't they just make them see more patients? Could this be a temporary benefit for doctors that ends up making it worse for them in the long run?"
I thought this was a great point, and so did Sharp. His answer, though, surprised me: "Let me be a little controversial and say, if we can see more patients, we probably should," he said. "In my practice, I would love to see more patients in my day if I can do so in a way where I can be wholly present and meet their needs completely and not burn myself out."
And herein lies the promise, and peril, of AI in health care. As Anthony accurately identified, a solution to make care more efficient could be misused and worsen clinician burnout. Or it could increase access to care while helping providers alleviate clerical burdens. The outcome depends not only on advances in technology but also on how we as a society choose to deploy them.