Photo: Andrew Harnik/Getty Images
Dr. Mehmet Oz has a plan to fix American health care. Use artificial intelligence, which will save costs and provide superior care, he told staff at the Centers for Medicaid & Medicare Services last April. He said a diabetes diagnosis would cost $2 an hour with AI, while a human would charge $100 for the same service — oh, and people might prefer an AI avatar to the mortal alternative. A robot can “do ultrasounds on pregnant women” and provide “digitized insights” so physicians never have to view the images themselves, he added this year. Robert F. Kennedy Jr. celebrated an “AI revolution” in front of Congress and has promoted it forcefully at Health and Human Services, though the technology hallucinated citations in a prominent report last May. While both officials are busy embracing AI, the White House has forced out thousands of staff at HHS and the Centers for Disease Control and delayed millions of dollars in funding for the National Institutes of Health, hindering cancer research. Maybe AI can summarize health records at federal agencies and hospitals. Or it will introduce more uncertainty and error into a health-care system about to implode.
Medicine is unpredictable by nature. When a person falls ill in the United States, she can’t get a diagnosis or treatment unless she can pay for it. She needs time away from work and family to make an appointment. She needs to find the right doctor, who can provide the right answers and the right solutions, or try to get better on her own. She may have to fight her insurance company to cover blood tests or imaging. Her doctor could ease her symptoms, but if they have an underlying cause, she’ll forgo more effective care until there’s an answer. Each barrier can delay an explanation, and with it, treatment that will improve or even save her life. Without a diagnosis, she may be “excluded from nearly every opportunity for treatment and community,” Alexandra Sifferlin writes in her new book, The Elusive Body: Patients, Doctors, and the Diagnosis Crisis.
The crisis of Sifferlin’s title spans “belated, inaccurate, or missed” diagnoses, which strain systems public and private. Although she completed her reporting before Trump won reelection, our current context lends urgency to The Elusive Body, which presents a complex and careful portrait of a poorly understood phenomenon. The second Trump administration inherited a problem, though officials don’t perceive it as such; Kennedy isn’t talking about the diagnosis crisis when he complains of a “chronic-disease epidemic.” His language is looser, conspiratorial. Something is making us all sick. Rather than address the system itself, along with its structural flaws, he dodges the subject altogether. There’s an element of magical thinking here — shiny new tech will improve all our lives — but the foundation is cynical. MAHA loves money. Good health is a personal responsibility and a product one buys. Kennedy and Oz will replace doctors with chatbots before they will question the profit motive in health care.
Although it’s difficult to quantify the problem, Sifferlin, an editor at the New York Times, provides a few figures. Roughly 5 percent of adults in the U.S. will experience diagnostic error, according to some reports, and as many as 3 million live with undiagnosed conditions. Researchers at Johns Hopkins University “estimated that some 795,000 Americans become permanently disabled or die each year because dangerous diseases are misdiagnosed,” she adds. There are many causes, such as racial bias and a for-profit model that leaves millions behind. Others may be less obvious: Appointments are getting shorter, specialists don’t always communicate with one another about the state of a patient’s health, hospitals aren’t hiring enough staff. One doctor tells Sifferlin about a lung-cancer diagnosis “that should have been made in weeks” but “stretched out over nine to 15 months” while X-ray readings got delayed because “there was no radiologist available in the hospital overnight.”
Yet the crisis is far from intractable. The Undiagnosed Diseases Network at the National Institutes of Health treats the most complex cases in medicine, using federal funds now under threat from the Trump White House. In Sifferlin’s work, the clinics are a rare haven in a corporatized, broken-down system, connecting sick people and their families to research, specialists, and, sometimes, answers. Outside the UDN, care is intermittent and patients must navigate a health-care scheme that doesn’t always take their needs or medical reality into account. If a primary-care doctor can spend only 15 minutes on each patient, it’s hard for either party to forge a working relationship — let alone a diagnostic journey. A physical exam can be critical to diagnosis; so can a patient’s demeanor. Remove that human relationship, as Oz suggested, and patients will have one fewer resource at their disposal, one fewer reason to trust the medical Establishment.
Of course, AI is already here in some capacity, and medical professionals are deciding what to do with it. Sifferlin interviews a medical student who uses ChatGPT to polish his arguments and review cases, but there are risks. Generative AI “can help people be more efficient,” but “it can also hinder critical-thinking abilities as people off-load their problem-solving to machines,” she writes, citing a 2025 study. Sifferlin’s subject tries to feed the bot “open ended” questions, making himself rely on his human brain. (There’s no way to tell if his peers are that diligent.) Dr. Isaac Kohane, who is part of the UDN, has studied medical AI and tells Sifferlin that GPT-4 could be particularly beneficial because it can know “all the arcana around rare diseases.” Still, it also makes “bafflingly simple mistakes,” as Sifferlin points out, and no chatbot can reckon with the ethical considerations that are part of medical care. Such nuances may be lost in the drive to cut costs at hospitals and clinics. Beyond the use of AI, remote staffing is on the rise as hospitals outsource crucial work to professionals who aren’t on site. That saves money; it may also be dangerous for patients. A dental student in Connecticut recently died after he was overseen by an ICU “teledoctor.” He wasn’t visited by an on-site doctor once in the four hours after his transfer to the ICU, according to a lawsuit filed by his family.
Nobody’s happy with our health-care system except people with something to advertise. Doctors burn out and patients doomscroll, commiserate, retreat into their imaginations. The average medical drama is a fantasy of recognition. Dr. Robby and Dr. House and Dr. Grey will see the truth, crack the case, and fix a body up. Even if the patient dies, their loved ones will at least know why; a tragedy with a cause is easier to accept than a mystery. The Elusive Body is strongest, then, when Sifferlin wrestles with the fallibility of our human minds. She quotes researchers Maria Dahm and Carmel Crock, who write that diagnosticians “should understand how uncertainty manifests,” instead of pretending it does not exist. What’s more, they should “openly discuss it with other clinicians and patients to normalize its ubiquitous and inevitable part in the diagnostic process,” the researchers add. If patients feel they’re part of the process instead of an impediment, they may be more likely to trust their doctors.
The MAHA ethos can’t solve the diagnostic crisis or help a person heal. As enforced by Kennedy and Oz, the movement severs patients from their best chance at health. Our health-care system pits patients against doctors, billing departments, and insurance companies; it is merciless in its machinery. In this world, why shouldn’t we buy a supplement? A DNA kit? What additional harm can a chatbot do? But expertise has a function, whatever MAHA believes. I may believe I have chronic Lyme, which Sifferlin calls a “disputed or controversial” diagnosis, yet there’s no evidence the condition exists. I still need a doctor, not sycophantic AI or a quack, because something else could explain my symptoms.
There is no alternative, in the end, to human care. When I was diagnosed with hereditary spherocytosis, a rare red-blood-cell disorder, I had an answer, but I needed support and expertise to make sense of it. When I was diagnosed with type 2 diabetes years later, I needed education, medicine, and still more support. If I’d gotten the news from a chatbot instead of a person, I would have punched my screen — if it had given me the correct diagnosis at all. In one recent study, LLMs misdiagnosed over 80 percent of cases in early stages of the diagnostic process. And AI can operate only within the parameters we give it. If it is trained on data that is flawed because of racial bias, it will reinforce that bias until a computer scientist corrects it. Even if AI improves and becomes more useful to physicians and researchers, it will never replace the human relationships at the heart of medicine. Health is a public endeavor, which Sifferlin reinforces over and over. She follows the case of Louise Proctor, a Kentucky woman whose legs begin to lock up when she moves. Nobody knows how to help her, if they even believe her, but then the symptoms spread to her siblings. They land in the care of the Undiagnosed Diseases Network, where physicians and researchers discover a genetic mutation is calcifying the Proctors’ arteries. There isn’t a quick fix, but research goes on, looking at treatments, finding new connections. When Louise visits the NIH in 2022 and sees the physician who helped diagnose her, she’s excited. “He’s such a sweetheart,” she tells Sifferlin, and that makes her life a little easier.
Source link
