In her first year of medical school, Diana Cejas discovered a lump in her neck. She visited the student medical center, where doctors told her it was likely benign. The lump kept growing. She returned. She was reassured: just a large lymph node. Then, after a grueling 36-hour residency shift, the pain became so severe she couldn't sleep. The following day, after she begged for attention, a doctor finally ordered a CT scan. She pulled up the results herself on the hospital's computer system. On the screen: a large mass in her neck. It was cancer. Even as she was learning to diagnose others, she had been unable to get an accurate diagnosis for herself.
Cejas was a victim of what Alexandra Sifferlin, in her new book The Elusive Body: Patients, Doctors, and the Diagnosis Crisis, calls "multiple kinds of diagnostic error—both incorrect and delayed." These errors are far more common than most people realize: an estimated 5 percent of Americans—roughly 13 million people—experience a diagnostic error each year. A 2023 study found that more than 750,000 Americans are permanently disabled or killed annually as a direct result of misdiagnosis.
Diagnosis is, as Sifferlin writes, "the most important piece of medical information a person can receive." Against that backdrop, the health-care system's failure to treat diagnostic error as a serious, structural problem is both mystifying and deeply troubling.
In 2015, a landmark report from the National Academies of Science, Engineering, and Medicine concluded that most people will be incorrectly diagnosed at least once in their lifetime. The report was designed to galvanize reform, much as a groundbreaking 1999 analysis of medical errors pushed hospitals to reduce mistakes in surgery and other procedural areas. Yet more than a decade later, little has changed. Sifferlin writes that no major U.S. health system tracks diagnostic error systematically. When patient-safety researcher Hardeep Singh attempted to launch a national diagnostic-error-measurement program, only nine hospital systems agreed to participate.
You might expect AI—with its enormous processing capacity—to offer a solution. But it is unlikely to do so without a fundamental rethinking of how medicine engages with patients. Misdiagnosis, The Elusive Body makes clear, is not a periodic glitch but the product of several interrelated and deeply embedded flaws—ones that are, surprisingly, rooted in medicine's own evolution over the past century and a half.
[Read: Hypochondria never dies]
Chief among these weaknesses is medicine's fixation on metrics and testing, and its assumption that patients' bodies will behave in broadly predictable ways. The rise of germ theory in the 19th century recast disease as a set of specific, identifiable entities with distinct symptoms and biological markers. New technologies—X-rays, laboratory blood work—gave physicians the tools to detect them. This was an extraordinary advance, extending lives and eventually giving rise to evidence-based medicine. But it also turned testing into the centerpiece of diagnosis. The consequence: doctors have become less attuned to what patients describe than to what tests reveal.
This problem is compounded by medicine's reluctance to examine its own errors, and by the enduring myth of the infallible physician who simply knows. As Sifferlin documents, medical schools largely fail to teach doctors to recognize their own cognitive vulnerabilities; one study found that more than half have no curriculum addressing the frequency with which physicians make diagnostic mistakes. A 2020 survey to which Singh contributed found that "clinicians were significantly less comfortable discussing diagnostic issues than other medical errors"—likely because acknowledging such errors strikes at the heart of professional identity and, in a litigious environment, carries legal risk. With no meaningful feedback loops—patients rarely return to tell a doctor You got it wrong—the system rarely confronts its own fallibility.
Adding to the pressure, physicians are increasingly squeezed by the demands of hospital administrators and profit-driven insurers who reward speed over thoroughness: 15-minute appointments, denied claims, and bureaucratic burdens that erode time with patients. Under these conditions, critical decisions are made in circumstances that all but invite mistakes. One study at Johns Hopkins found that first-year medical interns spent just 13 percent of their time in patients' rooms.
The physical examination—once the cornerstone of diagnosis—is in serious decline, replaced almost reflexively by a battery of tests. Yet a 2011 study suggested that at least 80 percent of diagnoses could be reached through basic clinical skills alone. Reporting on a bedside-medicine program at Johns Hopkins, Sifferlin describes a retired physician with heart disease who served as a teaching patient. The intern assigned to him moved immediately toward ordering lab tests, an EKG, and an echocardiogram. Her supervisor intervened: a careful physical exam—checking his pulse, listening to his chest—combined with a review of his medical history, yielded the diagnosis on the spot, with 99 percent certainty, without the delays that come from reflexive test ordering.
Popular culture, shaped by decades of medical drama and 20th-century faith in expertise, tends to frame diagnostic puzzles as hunts for rare diseases—a zebra where you should expect a horse. But Sifferlin reveals a paradox: the problem may be most pervasive precisely with common, treatable illnesses—because that is where the volume is greatest. The larger crisis is embodied by patients like Cejas: people with manageable conditions that were missed simply because no one listened carefully enough to reconsider. As Sifferlin reports, "Between 65 and 80 percent of diagnostic errors can be attributed to breakdowns between the doctor and patient during clinical interactions."
Yet beyond these preventable failures, there are more zebras than physicians are often willing to acknowledge. Perhaps the most dangerous mistake a doctor can make is not believing they know everything—but acting as if everything is already known. Millions of people today live with poorly understood chronic illnesses—myalgic encephalomyelitis/chronic fatigue syndrome, autoimmune diseases, long COVID—that don't fit neatly into established diagnostic categories or appear on standard tests in their early stages. Sifferlin draws on the work of anthropologist Abigail Dumes, who has argued that the ascendance of evidence-based medicine had the unintended effect of pushing these conditions outside "the trappings of medical legibility."
[Read: The making of a diagnostic mind]
The ability to see inside the body, in other words, also taught medicine to overlook patients whose illnesses aren't yet well studied. Gulf War veterans, for example, were told their symptoms were stress-induced—until researchers demonstrated that nerve-gas exposure had caused genuine biological damage. Long-COVID patients still routinely have their symptoms attributed to anxiety, despite rapidly accumulating evidence of the disease's physical pathology. In such cases, uncertainty curdles into dismissiveness. A system that cannot yet quantify what is wrong concludes that nothing is wrong—that it's all in your head—even as doctors are careful to add that what patients are feeling is, of course, real.
In some medical-humanities circles, it has become fashionable to question whether diagnosis even matters. At a talk I gave about searching for a diagnosis as a patient living at what I called "the edge of medical knowledge," a professor asked: "What is the point of a diagnosis, anyway?" I understood the question. A diagnosis is not a destination—it's a waypoint on a complicated journey. But as someone who spent more than a decade without one, I also know that receiving a diagnosis is transformative. In our bureaucratized medical system, it unlocks access to treatment, insurance coverage, clinical-trial eligibility, and—perhaps most essentially—recognition: a formal validation that your illness is real. A woman quoted in Sifferlin's book put it plainly: "I just want permission to be ill." The more urgent question, then, is not Why do we care so much about diagnosis? but Why isn't the system doing more to get it right?
Sifferlin's central argument is that the answer isn't primarily technological. The Undiagnosed Diseases Program at the National Institutes of Health succeeds at identifying rare and novel conditions not simply because its clinicians can sequence genomes—though they do—but because of the time devoted to each case and the cross-disciplinary collaboration that results when specialists are encouraged to think beyond their silos. Most physicians, however, are trained in a system that makes this kind of practice nearly impossible.
In How Doctors Think, medical-humanities scholar Kathryn Montgomery defines clinical judgment as "an intellectual capacity carefully cultivated through the rigors of a long apprenticeship spent dealing with radical uncertainty." Yet Brian Garibaldi, who directs the Center for Bedside Medicine at Northwestern University, tells Sifferlin that some medical students complete their entire training without ever being supervised while examining a real patient. Sifferlin offers a telling example: an intern who aces an ultrasound assessment but neglects to check a patient's basic reflexes—because he never noticed the reflex hammer sitting in the exam room.
Many observers have placed their hopes in artificial intelligence, which can assimilate vast amounts of information, detect patterns, and render rapid judgments. AI does offer real benefits: by transcribing appointments, it frees physicians from their keyboards and can help them give patients more direct attention. Research by UC Berkeley's Ziad Obermeyer and others has also shown that AI can help clinicians overcome unconscious bias. But The Elusive Body argues—persuasively—that AI will not be a panacea on its own. Obermeyer's own work demonstrates that AI replicates existing biases just as readily as it corrects them. Technology, after all, is deployed according to human values. The values our medical system currently rewards—speed, volume, cost reduction—are precisely the values that produced the diagnostic crisis we now face.
Today, Diana Cejas practices pediatric neurology. Her ordeal transformed her as a physician. "Sometimes it's all I feel like I do, listen to patients. I just let them talk," she tells Sifferlin. Her experience of misdiagnosis left her in a painfully paradoxical position: "I want my patients to trust me. At the same time, I have a hard time trusting physicians."
The word diagnosis derives from the Greek for "knowledge." To diagnose is not merely a medical act but an epistemological and moral one—a bridge between the world of the well and the world of the sick. It requires a fundamental willingness to bear witness: I believe your body is telling us something, even if I don't know what it is. The crisis Sifferlin documents—and that the reviewer has experienced firsthand as a patient—is that the system has made this foundational act of medicine, one practiced for thousands of years, exceedingly difficult to perform.
The Elusive Body insists that things don't have to be this way. The solutions already exist; what they demand is a reorientation toward uncertainty, attention, and time—and the political will to act on it. In an era of federal retreat from science, and at a moment when AI risks making every clinical encounter still more impersonal, that person-to-person reorientation feels not just overdue, but urgent.
