Join the enews community - Terms
Filter by Categories

What’s up, doc? Robots are beating medics in making a correct diagnosis

Reading time: 3 minutes

You laypeople need to understand that a correct diagnosis isn’t all that important to the modern doctor. A wrong diagnosis that leads to pointless, even dangerous, treatment? The merest bagatelle. Missing the real disease, which results in your premature death? A trifle.

We’re grateful to a leading investigator for putting us right on diagnosis. There’s a lot more to medicine than getting a diagnosis right, he explains—although you might think it’s a good place to start.  

Admittedly, his remarks were made on April 1, or All Fool’s Day, and so perhaps we shouldn’t take his comments too seriously. But it’s more likely that he was on the defensive after ChatGPT-4, an artificial intelligence (AI) robot, beat the pants off a group of interns and physicians.

The bot was challenged to come up with the correct diagnoses of 20 different clinical cases, while 39 medics were each given one case. They were all given the triage data, the patient’s report of what was bothering them. Then they went through the system review phase, when the medic gets additional information from the patient. Those steps were followed by a physical exam and finally diagnostic testing and imaging.1

While the bot scored 10 out of 10 across all four phases, the physicians achieved a score of nine, and the residents lagged behind with just eight—which, in real life, would mean they arrived at a wrong diagnosis in 20 percent of cases, researchers from Beth Israel Deaconess Medical Center found.

And, indeed, this is happening in real life. In 2017, a survey by the National Patient Safety Foundation in the US discovered that one in four people had been given a wrong diagnosis at some time or knew someone who had. This number had risen from one in six in the 2005 survey.2

The extent of the problem could depend on who you ask. Another survey of Massachusetts residents found that 23 percent had been victims of medical error, and half of these were diagnostic mistakes.

Not surprisingly, a poll by Isabel Healthcare—a company that provides tools to help a clinician arrive at a correct diagnosis—discovered that 55 percent of patients feared a misdiagnosis more than any other aspect of their therapy.

Isabel Healthcare was created in 1999 after Isabel, the three-year-old daughter of the founder, was given a diagnosis that was so wrong, it almost proved fatal. Her doctors had failed to detect that her chicken pox had developed into necrotizing fasciitis and toxic shock syndrome. The result of this error was two months in the hospital, including a month in intensive care, struggling to survive from multiple organ failure and cardiac arrest.

Isabel’s tragic story isn’t an isolated one. According to Patient Claim Line, misdiagnosis is by far the biggest reason for a medical negligence payout, and the UK’s NHS sets aside £200 million ($250 million) every year to compensate patients or their surviving families.

Lung cancer is one of the diseases that doctors are most likely to miss, along with stroke—although they commonly fail to detect sepsis, too. These three plus pneumonia and venous thromboembolism (a blood clot in a vein) make up nearly 40 percent of misdiagnoses.

In the US every year, misdiagnoses result in around 371,000 deaths and 424,000 permanent disabilities, researchers from Johns Hopkins University have discovered. These include brain damage, blindness, loss of limbs and cancers that have spread.

Spinal abscess, an infection of the central nervous system, is misdiagnosed more than 60 percent of the time, the researchers discovered, while cases of stroke are also regularly misdiagnosed. About 950,000 people have a stroke each year in the US, and doctors miss about 18 percent of those, leading to about 94,000 cases of serious disability and death.3

Misdiagnosis is certainly a matter of concern, but the public shouldn’t lose its faith in modern medicine, said lead researcher David Newman-Toker. After all, he argues, when someone has trouble speaking and moving an arm, it’s easy to diagnose a stroke. But a stroke can also cause dizziness or headache, which can be symptoms of many other things.

True, but it doesn’t fool the ChatGPT AI bot, and perhaps patients would feel a whole lot easier if they knew a robot, and not a medic, was giving the diagnosis.

What do you think? Start a conversation over on the... WDDTY Community

1. JAMA Intern Med, 2024; doi: 10.1001/jamainternmed.2024.0295 2. NORC at University of Chicago and IHI/NPSF, Americans’ Experiences with Medical Errors and Views on Patient Safety (IHI and NORC at University of Chicago, 2017) 3. BMJ Qual Saf, 2024; 33(2): 109–120
Article Topics:
  • Recent Posts

  • Copyright © 1989 - 2024 WDDTY
    Publishing Registered Office Address: Hill Place House, 55a High Street Wimbledon, London SW19 5BA
    Skip to content