In case you missed the news, last week saw the launch of the smartest medical investigation tool in the world. Or, if not that, then at least last week’s biggest health claim in the world.
Ada, a smart phone app designed in Europe but yet to be launched there, is being tested on New Zealanders and Australians first. The makers suggest it is more than just an app (not a literal claim – it is indeed an app), and instead describe their product as a virtual companion.
The underlying concept is clever. This app is the first to tackle health diagnosis on such a grand scale, and its market penetration will be closely watched by Google, Facebook and perhaps startup companies hoping to access a portion of the trillions spent by US medical insurers.
Everything about its design is big, from the team involved (said to include a hundred doctors) to its capacity to eventually integrate with medical video consultations. It even aims to build a picture of the user’s medical history over time, which will presumably adjust the weightings given to each branch of its diagnostic algorithm.
The RACGP and AMA have both issued warnings about diagnostic apps, pointing out the dangers of relying on their results. The app’s marketers have tried to pre-empt this criticism by building in a function where the app can send a copy of its output to the user’s chosen GP.
Technology enthusiasts may be tempted to dismiss RACGP concerns as doctors trying to keep robots off their turf, but the warnings are salient. The issue isn’t as simple as getting the diagnosis right or wrong.
In algorithmic approaches, the risks of both underdiagnosis and overdiagnosis are high. The former can lead to false reassurance with dangerous consequences for the individual, but the latter is the bigger risk at a population level.
In the same way as phone triage systems tend to err on the safe side by referring to the Emergency Department, the makers of any algorithm know that the lower they set the bar, the less likely they are to run into trouble.
The risk of overdiagnosis will be tempered if the thousands of users attend GPs, but this is not free: it involves financial, time and opportunity costs. And the capacity to diagnose and treat always outstrips the availability of evidence to sort the useful from the useless or harmful.
The accuracy of the Ada app remains to be seen. But diagnostic tools, even if accurate, can cause harm via their flow-on effects.
An analogy would be the superbly accurate images of spinal MRI, which turn out not to be helpful for the average person with low back pain. Once the ‘diagnosis’ exists (virtually every spinal MRI gets one), there is pressure on both patient and doctor to intervene with a treatment, even if it lacks evidence.
People who work in IT tend to envisage a perfect system, where no harm results merely from adding extra information into the algorithm. In real health systems, however, these harms happen every day, and increasingly so. For example, see this BMJ article on overdiagnosis or last week’s NEJM article describing a world-wide thyroid cancer epidemic which mainly exists not because we are better at diagnosing a disease that might kill us, but better at diagnosing a disease which would never have harmed us in our lifetime, yet results in surgery or radiation therapy for those diagnosed.
Soon, the average punter will have access to cheap, whole-of-genome tests, which will be able to print out a list of predispositions written into your DNA. Again, when the capacity to ‘diagnose’ becomes button-click simple, we have to think carefully about how and why we asked the question.
Not that this alone means we shouldn’t use diagnostic smart phone apps, but it sounds a note of caution for those who assume their use will necessarily improve health outcomes. Developers who rush to follow in Ada’s footsteps will pay more heed to their marketing and revenue wins rather than weighing up health benefits versus harms: that is someone else’s job.
I have no doubt diagnostic apps will also result in benefits, and we will no doubt soon be reading about life-saving anecdotes and things the doctor missed but the app picked up on the way through to the keeper.
Now they have arrived, the apps are here to stay. This isn’t a bad thing, but it will require plenty of work from within the health system to ensure users reap a net benefit.
For me, it’s a matter of avoiding the anecdotal hype and keeping an eye on the big picture.