This article originally appeared on VICE US.
So many minor inconveniences can now be solved via phone app, it makes sense that basic healthcare is on its way to becoming one of them. In the past few years, apps that purport to do things like manage diabetes, use period tracking data to warn users about PCOS, and identify potentially cancerous skin lesions have cropped up. But a new review published Monday in the BMJ raises serious concerns about that third category.
The review examined studies that have previously looked into the efficacy and reliability of six available apps that use artificial intelligence to determine the cancer risk of skin lesions. The studies included the review examine the apps’ efficacy, and are used by the apps as a testament to their validity. These apps scan user-submitted photos and give an immediate analysis of whether the lesion is of low, moderate, or high risk of skin cancer. Jon Deeks, lead study author and professor of biostatistics at the University of Birmingham in the UK, told VICE that most of the studies reviewed were done by app developers themselves (sus), with a small number of independent studies.
That’s only the first major issue; Deeks said there were “major limitations” in all the reviewed studies. Upon review, researchers found many of the studies didn’t examine actual user-submitted photos of skin lesions, but photos taken by clinicians, so the photo quality was likely much higher than if a regular person took it themselves (as is the intended use of these apps). On top of using biased images, study sizes were also found to be quite small. Based on the review’s findings, Deeks said none of the available apps are currently reliable enough to be used for their marketed purpose of reducing doctor visits, and helping with early diagnosis of skin cancer.
In spite of this, one market leader—SkinVision, an app available throughout the UK—vaguely claims on its website to be “clinically proven,” and advertises that users will “no longer have to visit the doctor unnecessarily.” Deeks said SkinVision currently sells itself as being able to identify nine out of 10 potential cancers, but that seems to be an overestimate, given the issues with the reviewed studies. “Whether [SkinVision] is reliable enough is the leading question,” Deeks said. “Our concern is the users of these apps haven’t been told there’s an error rate at all; they may be missing cancer, or delaying getting help when they need it.”
None of the skin cancer screening apps are currently available for use in the United States, and none of them have received FDA approval (similar apps that help guide users through home checks are available in the US, but don’t determine whether lesions are high or low risk). But Deeks mentioned that two of the apps in his review found to be insufficient (SkinVision and SkinScan) currently have a CE marking, or an EU trademark verifying their health efficacy. “We see a trademark on an app, and we trust it does what it says,” Deeks said. “That’s an issue that needs to be resolved.”
The biggest issue is that an app might deem a mole as being low-risk, when it is, in fact, at high risk of becoming cancer. If caught early, skin cancer is highly treatable, and delays in diagnosis can be fatal. Knowing this, it feels sinister than any app would market itself as being anything close to properly effective, when the only studies to that effect are slim at best, and dubious and unethical at worst.
With improvements to the AI, Deeks believes it could be possible for an app to help people in the general public spot potentially cancerous lesions. “The idea that we could use our phones to judge whether or not we need medical assessment is really powerful, because there are a lot of people who delay going to see the doctor; the idea of an app is brilliant,” Deeks said. For now, though, none of the available options are sufficient; an app, believe it or not, isn’t yet able to identify skin cancer.
Follow Hannah Smothers on Twitter.