This transcript has been edited for clarity.
John M. Mandrola, MD: Hi, everyone. This is John Mandrola from theheart.org | Medscape Cardiology. I'm here at the American College of Cardiology (ACC) meeting in New Orleans with my friend, Dr Mintu Turakhia. He is a Stanford cardiologist and the co–principal investigator of the Apple Heart Study, which he just presented here to a packed audience. Mintu, I'm really happy to have you.
Mintu Turakhia, MD: John, thanks for having me back. It's always fun to talk with you.
What Is the Apple Heart Study?
Mandrola: What is and what isn't the Apple Heart Study? Is it an observational study, a screening study, or a trial?
Turakhia: Great question. The Apple Heart Study is a single-arm, open-label study of an intervention. We are not calling it an observational study because we did give an intervention, so it's not a Framingham-like study. We wanted to assess an algorithm that was very carefully developed and really understand the key parameters around it.
The aims were: What is the notification rate, or the burden? How well does the algorithm do? Is it safe? What is the yield of atrial fibrillation (AF) subsequently on a gold-standard monitoring test? This is a very different way of thinking than a traditional framework of a clinical trial for screening. We did not randomize; this was not a prescribed intervention of a therapy that a doctor or a public health service would do. We're not really looking at differential outcomes, the hard outcomes that are very important to people. Instead, we wanted to evaluate this algorithm and its performance and safety in the real world in a way that people are going to interact with the technology they have.
Mandrola: This is a special kind of study. Is it a new way of doing clinical science? There are 420,000 people. Talk about the pragmatic nature of this.
Turakhia: As much as I am an AF specialist and interested in the risk and outcomes and so many of the things you and I have spoken about before, the fact that there is a wide consumer base of these devices affords this unusual and incredible opportunity to think about pragmatic design. The other thing is scale: What do you need to do for scale regarding safety, study visits, and making sure there are no emergencies when participants are evaluated? To that end, we created this end-to-end virtual framework for the study. I think that is very exciting and we hope that that provides a lot of information for the clinical trial community.
Mandrola: There are nearly 420,000 people in the study, but when you analyze the outcomes you analyze it on a small number. Talk about that.
Turakhia: This itself is a key finding of the study. We enrolled 400,000-plus participants, and almost 25,000 were 65 or older. When you map that to US census data, overall that is about 1 in 600 US adults, which I think is a striking number and one that we are very proud of. The top finding is that the overall proportion that got notified of an irregular pulse was 0.5%, meaning that 99.5% did not get notified. Now you are dealing with about 2100 individuals who received a notification and you follow them.
There are a couple of important points here. The first is that the notification rate was low. That is very important. Second, as you tie this into thinking about trial design methodology—we have to back-calculate our power and the quantities and number of ECG patches we want—you have to make assumptions on engagement and overall notifications. Even with that, this is whittled down to a small cohort, and that is the cohort from which you derive the study visits and ECG patches.
Mandrola: Were the 0.5% of patients notified out of the total 420,000? Or out of the small, over-age-65 group?
Turakhia: That is the overall notification rate.
Mandrola: Tell us about the second endpoint.
Turakhia: That was a technical outcome as a primary endpoint. Perhaps it's worth explaining the algorithm. For clinicians, this is a very different way of thinking about AF detection. This is what I would call almost a probabilistic or opportunistic approach.
Let's start with what we know. With insertable cardiac monitors, ambulatory ECG, and pacemakers, we look for continuous recordings of discreet AF episodes. If you take away the ECG and look at the optical sensor on the Apple Watch, what we can measure is a tachogram. A tachogram is a series of pulse timings between each other. You want to find out whether one tachogram is enough. It turns out that it's probably not, based on what we know about how these work. The algorithm will scan periodically, when you are still, opportunistically at these times. If it sees one tachogram that meets irregularity criteria, then it starts sampling more frequently. If then five out of six meet irregularity criteria over a 48-hour period, a notification is flagged. That is very different from single discreet episodes in the way we've thought of things.
The reason we wanted performance endpoints around positive predictive value is that at the tachogram level, we fundamentally wanted to understand the predictive value of the building block that forms that five-out-of-six notification trigger. That is why that was the first co-primary endpoint. The more clinically impactful endpoint is, what did you see simultaneously on an ECG patch at the time it would have met notification criteria? We looked at that for a secondary outcome.
Mandrola: How did that come out?
Turakhia: The clinical notification concordance endpoint—this is the group that was sent the ECG patch after having a notification and then those who had simultaneous watch data, so this is all downstream of notification. We're now looking at the watch and ECG data. We have the data, we lined them up, and we looked to see when their annotation markers flagged for when the algorithm would have met notification criteria. Of those situations, the positive predictive value for AF was 84%, meaning that 84% of the time that it would have met notification criteria, simultaneously you had AF.
Mandrola: What do you think about that?
Turakhia: We accomplished what we set out to do, which is to find what the boundaries are in characterizing the algorithm and to give some clinical insight and context about what we can do with that information. We think that that does provide substantial information to inform a diagnosis of AF and it does what it's supposed to do.
We just finished collecting data on February 25. We have a lot of other things we want to look at. One of those things is to look at all other arrhythmias that could have occurred at the time.
How Will This Study Help Clinicians?
Mandrola: One of the things mentioned in the presentation was how this will help clinicians. I really wonder about that. How would this help clinicians? What should clinicians do when a patient has an irregular tachogram?
Turakhia: It's a great question. The goal here was to understand the performance characteristics of the algorithm, and so now that you have that, what do you do? I think you fundamentally have to stick with what we do as clinicians. Take a good history, do a physical exam, look at risk factors. But I can extend that further now and say, "At the time you received the notification, what was happening? What were you doing? How did you feel?" It adds a little bit of contextual information on top of the other risk factors. I think as a starting point, that is where we go.
Garnering of Mainstream Attention
Mandrola: There has been a lot of hype, a lot of mainstream media coverage about this. As a cardiologist and researcher in this field, what do you think about that?
Turakhia: We, as investigators, come in with equipoise and we are always circumspect with our findings. This is a situation where we defined and accomplished our goals of algorithm performance, notification burden, and safety. We recognize that we have a lot more to do in terms of how this gets integrated into health and that we need more clinical trials. It was great to see that there are now randomized trials exploring these digital health tools in AF disease management.
Mandrola: Can you give us any insight as to what is going on?
Turakhia: I'm not involved with them; I'm just referring to the trial mentioned in the panel discussion that followed our presentation. [Editor's note: The HEARTLINE randomized trial will assess the effectiveness of the Apple Watch and educational initiatives for early AF detection to reduce death, MI, and stroke in people 65 years or older.]
Mandrola: It's good that there will be more trials. Finally, what is the future? I know you said that more data are coming, but what can we look for?
Turakhia: With this amazing team we have, I'm just focused on unpacking all these data for this Apple Heart Study and learning a lot more and asking key questions. What I found valuable from having this opportunity to present at ACC with my co–principal investigator, Marco Perez, MD, is getting extraordinary feedback on the things we could look at. I think that is still where we are for step 1. We've had a fantastic collaboration with the sponsor. It's been very enriching for us and enriching in the way we think of how to do trials with technology and engineering.
Mandrola: Excellent. Thank you for being here.
Turakhia: It's my pleasure, always.
© 2019 WebMD, LLC
Any views expressed above are the author's own and do not necessarily reflect the views of WebMD or Medscape.
Cite this: Getting to the Core of the Apple Heart Study: ACC 2019 - Medscape - Mar 19, 2019.