Treat AFib ‘Diagnosed' by Smartwatch?

Medscape

06-05-2025

This transcript has been edited for clarity.
Welcome to Impact Factor , your weekly dose of commentary on a new medical study. I'm Dr F. Perry Wilson from the Yale School of Medicine.
A 77-year-old man with some cholesterol issues, but otherwise healthy, feels a fluttering sensation in his chest, along with some shortness of breath. It's disquieting but not painful. It persists for an hour or so and then seems to go away on its own. The next day it happens again, and his wife finally convinces him to call his doctor. He goes into the office where a 12-lead EKG is placed and the diagnosis is made clear: atrial fibrillation (AFib), the most common arrhythmia in the world.
That is how AFib used to be diagnosed: a symptom, a doctor's visit, an EKG. It is data from people diagnosed in that way that gave us all our guidelines about how to treat AFib — rate control with beta-blockers, anticoagulants, and so on.
But that's not really how AFib is getting diagnosed anymore. Now, that whole process is boiled down to a smartwatch notification.
Whether we realize it or not, we are in a whole new world of medical diagnosis. The era of waiting for a symptom to develop and having a doctor order a diagnostic test is rapidly coming to an end. We have multiple consumer-facing products that give real information about health status, from blood pressure to blood sugar. We have direct-to-consumer testing ranging from genetics to cancer screening.
We have machine learning models that can look at your prior data and make predictions about your future health without you ever setting foot in a doctor's office.
Is that a good thing? Let's take the AFib example. Your smartwatch tells you it thinks you have AFib. You are having no symptoms whatsoever. We call this 'subclinical AFib.' What emotions are you feeling? Anxiety over a new medical problem? Relief that it was caught? Are you glad you had the watch on or is ignorance bliss?
To remove this from the realm of emotion, I will make the argument that detection of a disease is only important insofar as it can be treated. And I'll go one step further to say that early detection of a disease is only important insofar as early treatment provides better outcomes than late treatment.
So, how do we treat AFib?
Well, in the old 'visit your doctor' days, we knew just what to do. We calculated your risk for stroke based on something like the classic CHA 2 DS 2 -VASc score, and if the stroke risk was high enough — usually 2 or higher — we put you on a blood thinner.
But blood thinners aren't exactly gummy vitamins. The data are quite clear that, among people with AFib detected the old-fashioned way, blood thinners reduce the risk for stroke. But they come at a cost, and that cost is bleeding.
Risk/benefit. As ever. For always. This is why we estimate stroke risk before prescribing blood thinners; if it is very low, the benefit is not worth the risk.
But what about people whose AFib is detected in the new way, the smartwatch cohort? What about their risk?
We can actually answer this question now, thanks to two randomized trials — the ARTESIA and NOAH-AFNET 6 studies — that specifically evaluated whether anticoagulation was useful in people with what they call 'device detected subclinical atrial fibrillation.' This would include smartwatch types but also people with other devices, like pacemakers, that might be able to detect AFib incidentally.
The studies differed in their primary results. ARTESIA showed that the use of a blood thinner reduced the risk for stroke but increased the risk for major bleeding. NOAH-AFNET showed that the use of blood thinner had minimal effect on the risk for stroke but increased the risk for major bleeding.
Benefit and risk.
Putting the data together, the benefit in terms of reducing stroke seems rather small, and the risk for bleeding is clear and real, but, how do you compare those things? What's worse — a stroke or bleeding? Or, to make the question even harder, how many bleeding episodes are equal to one stroke?
Finally, this week we got a really interesting approach to answer that question, thanks to this study appearing in JAMA Network Open, which integrates not only the two big trials of blood thinners for subclinical AFib, but also the risk for stroke with the risk for bleeding. Here's how it worked.
The study was done using a computer. I know — all studies are done with computers. But here I mean literally. The authors simulated 20,000 individuals in a computer. All of them were basically the average of patients in the NOAH and ARTESIA trials — around 77 years old — and had the same baseline risk for stroke, bleeding, and death from other causes as the control group in those trials, which is to say about 1% for strokes and bleeds and 4.3% per year for death from other causes. Half of the virtual humans with subclinical AFib would get blood thinners and half wouldn't.
The researchers also assumed that the effect of treating these people would be the average effect seen in those trials — to wit, reducing the risk for stroke by 32% and increasing the risk for bleeding by 62%.
With that, they could run these 20,000 little silicon beings through a web of potential changes in health status — this is called a Markov chain — all with various probabilities.
Basically, they tell the computer to skip forward a month at a time. Everyone starts out healthy. After a month, most people are still healthy; strokes and bleeding and death are rare events, after all. But some people experience one of those outcomes. And then time leaps forward again, and some more people experience those outcomes, and maybe some of the people from the past month get better and some people develop overt atrial fibrillation and go on the old-fashioned "see your doctor" pathway. It's all driven by probability.
And they didn't just capture whether the people would have a stroke or a bleed or not; they captured how bad it would be in terms of quality of life. This is how you start to compare apples and oranges. A stroke can be devastating to quality of life. But bleeding can be as well, particularly bleeding in the brain. With every cycle of the simulation, we know how these 20,000 people are doing not just in terms of whether they are alive, but in terms of their quality of life.
With that set up, you can ask a really straightforward question: If I take a healthy person with subclinical AFib, will their quality of life be better over the next 10 years if I treat them with a blood thinner or if I don't?
With 20,000 patients iterating in silicon, over the 10-year virtual time scale, you'd see 1076 strokes if you didn't treat with a blood thinner and 843 strokes if you did. That's good. But treatment would net you an additional 453 major bleeding events. That's bad. Treatment would lead to 55 fewer deaths. That's good.
But for an individual person, what is the answer? With treatment, do you live longer and with better quality of life?
Well, over 10 years, treating someone whose smartwatch tells them they have AFib with a blood thinner will increase their lifespan — by 9 days. It will also increase their quality-adjusted life — by 9 days.
Nine days over 10 years of treatment. A net benefit, but, come on. Nine days?
Now, I know what you're thinking. We don't just blindly treat everyone, even when they are diagnosed with AFib the old-fashioned way; we treat them based on their stroke risk. Well, the authors could give these simulated patients any stroke risk they wanted. At the high end of risk, treatment with a blood thinner improved quality-adjusted life duration by 11 days.
It is very hard for me to look at this data and conclude that we should be treating subclinical AFib with these drugs. Of course, this is one of those 'talk to your doctor' situations. The risk/benefit paradigm requires some personal reflection about risk tolerance as well as thinking through the implications of a stroke and the implications of a major bleed. In a larger sense, this study is emblematic of a problem we'll face in this new direct-to-consumer medical economy. Every medical decision is based on the risk vs the benefit of treatment, but our understanding of risk comes from a dying paradigm of diagnostic care that occurs in the doctor's office. People diagnosing themselves almost certainly have lower risk than those who make it to the clinic, which changes the treatment calculus substantially.
That said, for me at least, there is something disconcerting about knowing that something is going on with my heart and doing nothing about it, even if that is the rational choice. I feel like it would nag at me. The AFib alert on your smartwatch is a genie that can't really be put back in the bottle. People may want to think about turning that functionality off. As a great computer once said, 'The only winning move is not to play.'