Atrial fibrillation and optical heart rate sensors
CNBC is reporting that Apple is expected to announce a study on detecting atrial fibrillation using Apple Watch’s heart rate sensor tomorrow, following UCSF and Cardiogram’s May study showing 97% accuracy at detecting atrial fibrillation using Apple Watch and DeepHeart, Fitbit’s announcement of its own AF study, and early work detecting AF with the Samsung Simband.
For everybody who’s new to this topic, here’s a brief rundown on what atrial fibrillation is, why it matters, and why so many entities seem to be starting with AF as “low hanging fruit” for wearables in healthcare.
What is atrial fibrillation and why does it matter?
Atrial fibrillation is the most common abnormal heart rhythm and causes 1 in 4 strokes. For some people, AF feels like a three-legged washing machine in their chest, but other people feel nothing at all—and that’s the problem. In the GLORIA-AF study, those with asymptomatic AF were more than twice as likely to have a stroke. By detecting atrial fibrillation early, we can prevent strokes.
For more detail, see this video from Dr. Greg Marcus, UC San Francisco, who’s an investigator on our study:
How is atrial fibrillation usually detected? How is using an optical heart rate sensor, like Apple Watch’s, different?
AF is typically detected using an ECG (electrocardiogram), such as a Holter Monitor or Zio Patch, which the patient wears from 24 hours to 2 weeks. The optical heart rate sensor on Apple Watch, Fitbit, and Android Wear devices are all based on PPG (photoplethysmography). They measure pulse waves that travel from your heart down your limbs—one pulse wave per heart beat.
Detecting AF with consumer-grade wearables presents both challenges and opportunities. The opportunity is that you can monitor for months or even years, not just weeks, since people like to wear watches and hate wearing medical devices. That means you can catch a higher percentage of atrial fibrillation—in the CRYSTAL-AF study, for example, it took a median of 84 days of monitoring to find the first AF episode. You can’t do that with a Holter monitor.
The challenges are twofold: first, PPG provides less information than ECG—for example, there are no P waves. Second, your heart rate reflects everything that happens in your life, leading to huge amounts of variability due to stress, sleep, and physical activity:
As a result, most simple, “off-the-shelf” algorithms work well in a clinical setting, but fail in the more complex ambulatory setting.
What’s been proven by these studies so far? What’s a realistic timeline for real deployment?
What Apple and Fitbit are announcing are research partnerships. Medicine is an evidence-based field, and you have to walk (prove it works in a clinical setting) before you can run (deploy it to real users).
The typical timeline for a research partnership is roughly two years, broken into phases:
- Setup: 3–6 months to design the study, receive IRB approval, and begin recruiting patients.
- Recruitment: most studies take 6–12 months to recruit enough patients to achieve statistically significant results.
- Analysis: Data analysis is variable, but it’s reasonable to budget 3–6 months.
- Publication: often work will first be presented at a conference (~6 month timeline), then a medical journal (6–9 months).
To give an example timeline, Cardiogram’s study with UCSF Cardiology launched in March 2016, the initial architecture of DeepHeart was presented at the NIPS Machine Learning for Healthcare workshop in 2016, the first clinical results were presented at Heart Rhythm Society in May 2017, and a medical journal publication is in process. Rigorous clinical research requires patience.
Hey Cardiogram! Are Apple and Fitbit going to eat your lunch?
Why, thank you for asking. But we were expecting this. Our mission is to reinvent preventive medicine with consumer wearables; while atrial fibrillation is a logical place to begin, it’s just the start. Over the last year, we’ve developed DeepHeart’s capabilities to detect several major health conditions—ailments that affect 1 in 4 people—using consumer wearables alone. We’ve already submitted study results, and you’ll see some news from us in the coming months as those are released.
It’s also worth remembering that although Apple and Fitbit are #1 and #3 wearable manufacturers, they represent, respectively, only 15% and 12% of wearable sales. In fact, the wearables market is actually fragmenting as it goes mainstream:
We think there’s going to be a strong need for a cross-platform “operating system”—an entity that ensures that no matter which wearable you own, you can get equal access to the future of medical care.
It’s our goal to become that entity. If you’re an engineer, designer, payer relations specialist, or clinician and looking for meaningful work, please email me: brandon@cardiogr.am.
What if I have questions?
Feel free to email me, too. I’m brandon@cardiogr.am.
Additional References
[1] “Screening for atrial fibrillation” (2017). Freedman et al.
[2] “Economic Burden of Undiagnosed Nonvalvular Atrial Fibrillation in the United States” (2015) Turakhia et el.
