Kevin Lomangino is the managing editor of HealthNewsReview.org. He tweets as @KLomangino.
A study generated worldwide news coverage last week for documenting an advance in artificial intelligence (AI) applied to health care. It showed that a “deep learning convolutional neural network” could be trained to differentiate cancerous melanomas from benign moles more accurately than an international group of dermatologists.
Many stories, like this U.S. News & World Report piece, suggested that AI “may serve physicians involved in skin cancer screening as an aid in their decision whether to biopsy a lesion or not.”
But none of the stories I looked at paused to ask, Is finding more melanoma definitely a good thing?
That was the question that immediately came to mind for Ade Adamson, MD, an assistant professor in the Department of Dermatology at the University of North Carolina at Chapel Hill. He points out that the number of melanoma diagnoses among white Americans has increased some four to six times over the past 40 years, and yet deaths from melanoma have stayed constant during that period.
This suggests that many of the melanomas being detected today might be better off not being found. These cancers are so slow-growing that they would never cause a problem, he said.
The U.S. Preventive Services Task Force cites overdiagnosis as one of the reasons it doesn’t endorse routine skin cancer screening. The task force concludes that “the current evidence is insufficient and that the balance of benefit and harms of visual skin examination by a clinician to screen for skin cancer in asymptomatic adults cannot be determined.”
And yet doctors are certainly finding and treating more of these cancers every year and creating an ever-growing number of melanoma patients. Adamson believes AI-based evaluations could make the problem worse by increasing the availability of screening outside of doctors’ offices.
“Overdiagnosis in melanoma is one of the most under-discussed problems in dermatology,” he said. “Actually, some call it the ‘third rail’ of dermatology, so many don’t even mention it. It was never discussed when I was training.”
David Elpern, MD, a dermatologist in private practice in Williamstown, MA, agrees that overdiagnosis is a big problem in dermatology and sees technology as contributing to it. He spoke to the Lown Institute conference recently about changes in dermatology practice that promote unnecessary and wasteful care.
Elpern sent me this graph from the National Cancer Institute which documents the disconnect between melanoma incidence (the rising green line which indicates the number of people diagnosed with melanoma each year) and melanoma deaths (the flat dark line at the bottom of the graph indicating no change in the mortality rate).
He says there are many factors driving the increase in melanoma diagnoses.
“Pathologists are calling things melanoma that they never called melanoma 30 years ago,” Elpern said.
Elpern also points to tools like the dermatoscope, which is used to evaluate suspicious moles. He says they can pick up potentially cancerous changes before they might be detected with the naked eye, which is not always a good thing.
“The dermatoscope and mole scans can pick up very early things, and then the patient is labeled as having a melanoma and they get shunted into the system,” he said. He believes screening “creates a lot of health anxiety on the part of patients” despite the lack of proof that it’s improving outcomes.
Most of the stories that I looked at, including these pieces from Men’s Health, Newsweek and The Guardian, did little more than re-write a European Society for Medical Oncology news release about the study. But HealthDay dug deeper, quoting an independent dermatologist who questioned whether the study represented real-world conditions.
That’s an issue that was also raised by David Swanson, MD, an associate professor of dermatology at the Mayo Clinic in Scottsdale, AZ. Swanson, who has consulted for a company that wants to make automated skin cancer screening more widely available through pharmacies, says the study included certain types of moles that had already been pre-sorted by humans, and that the machine was only asked to make a choice between “melanoma” or “benign.”
“The test that they did was a simple one, it’s simply a binary choice and it was very artificial,” Swanson said. “There are all sorts of things that grow on human bodies in the real world, and that study hasn’t been done yet.”