Engineers develop wearable heart attack detection technology
medicalxpress.com74 points by PaulHoule 7 days ago
74 points by PaulHoule 7 days ago
How do you get a decent ECG with a wearable that has a single point of contact with the skin. If you need to touch a contact on the device to get a two lead equivalent ECG then how can this be real time monitoring?
Real-time monitoring and continuous monitoring are different goals. The use case for something like this is probably a wearer who is experiencing heart palpitations, and has the time and clarity of thought to actually find the correct app on their device and start recording.
Continuous monitoring is extremely challenging still because ECG data needs to be sampled at a relatively high frequency (~200 Hz) to accurately identify the QRS complex in the waveform. That uses a lot of power, and the batteries we have still aren't good enough to support those types of demands. 200 * 60 * 60 = 720,000 samples per hour to collect and process.
It's possible that algorithmic approaches may be able to reduce the sampling frequency required. Power-constraints were the main issue when I studied this topic 10 years ago during my master's degree. I had looked into non-frequency domain techniques (such as empirical mode analysis/Hilbert-Huang transform) as a possible way to reduce sampling frequency and thus power consumption.
https://github.com/AshwinSundar/Empirical-Mode-Decomposition...
There are audio recorders like TileRec by Attodigit that can record for 30 hours on a single charge and voice is recorded at 16khz and compressed live to mp3 and is super tiny at 0.5z. I’m thoroughly unconvinced that power or complexity of handling a 200hz signal is the bottleneck. 200hz is nothing probably even to a low end microcontroller these days.
Fitbit supports detection of Afib and has FDA approval. It’s not entirely out of the realm of possibility that one day it could also receive approval to detect heart attacks. It probably already has the data that might lead a savvy user to infer it just by looking at elevated resting heart rate, SpO2 differences, and irregular heart rhythm alerts.
My Fitbit record showing heart rate spikes to 125 was one of the things that I noticed that suggested I was having a heart attack in 2020 (I was).
Hm, why is 125 a heart attack? Isn't that a fairly normal heart rate for exertion? Or were you just at rest?
GP could have been having a non STEMI heart attack.
I was, actually, but I don’t know how that’s connected? This was something like a ten minute spike in the middle of the night, a couple of times.
The sympathetic nervous system activates, increasing heart rate and the body is trying to compensate for the compromised heart muscle. Then, there is the possible adrenaline surge because you are actually in pain. If the NSTEMI causes left ventricular dysfunction, the heart compensates with a faster rate to maintain blood flow.
I went through this with a relative just two weeks ago and learned this from the cardiologists
This is out of my specific expertise, but the AFE1291 does 8k 24 bit samples per second and uses 495 μW when sampling. That's enough to run full rate for 10 straight days on an apple watch battery (1.189 Wh was the first google result).
Integrated frontends have had big impacts on efficiency and improving batteries have had basically the same increase in specific energy. I would be shocked if power was a limiting factor.
https://www.ti.com/lit/ds/symlink/ads1291.pdf?ts=17469713549...
i would imagine you could take an ekg for 1-2 seconds every 30 seconds and run it through a much more simple model to determine if it should take a longer sample, no?
rather than reducing the frequency of the sampling, dynamically adjust the duty cycle of when sampling is happening?
this is probably a dumb suggestion, it seems pretty obvious. for example the apple watch doesn’t do o2 monitoring continuously, just for some fraction of the time.
do you need to sample every second to detect heart attacks? don’t they continue to show up on an ekg for more than 30 seconds?
> apple watch doesn’t do o2 monitoring continuously, just for some fraction of the time
Making it effectively useless? Unless the fraction of the time is multiple times per minute? E.g. in sleep apnea it's not uncommon for some desaturation to occur, triggering an arousal and deeper breaths, restoring saturation, only for the cycle to repeat 2 minutes later.
My Garmin has a similarly useless feature. I have no idea what the supposed benefit is. Maybe they hope that if they sample multiple nights they can detect some desaturation anyway and can get the user in for polysomnography? Might be worth it.
I have mild sleep apnea and my Apple Watch is the reason I figured it out. If you wear it every night, you will catch it. I have on average 1-2 events an hour. Even with that few, the Apple Watch would catch the O2 drop enough times over a month that it worried me to see my O2 drop under 90%
Impressive that it caught it. Did the apple watch catch 1-2 events per hour? Or was that a sleep study? Sleep apnea isn't usually "officially" diagnosed until you hit >5/hour, 1-2/hour isn't even "mild", it's basically "normal".
There’s a strong history of useful signals from single lead ecgs.
Detection of ecg anomalies(especially episodic ones with intermittent recording) was the subject of the physionet cardiac computing challenge almost 10years ago[0].
It’s amazing how far machine learning has come. I know teach a version of this challenge as a one day in class activity in my department’s physiology class. They actually get to train multiple models on a gpu cluster (and compare that to trying to train models on their laptops).
One thing we reinforce in the lesson is human vs. computer “interpretation”. They/clinicians can look at ecgs and make some sense of them. An LTSM is worse than random chance/a medical student. However moving to the frequency domain makes the LTSM more accurate than cardiologists, but neither they nor clinicians can “see” afib ina spectrograph. It’s a great way to talk about algorithmic versus human reasoning and illustrate that to students.
That then gets reinforced with other case studies of the ying and yang of human and machine decision making throughout our curriculum- like alpha fold working great until you ask it about a structure in the absence of oxygen, because that’s not in its training data.
> There’s a strong history of useful signals from single lead ecgs.
But to be clear, a single lead ECG requires two electrodes at a minimum and commonly a third as ground. So a single lead ECG will have minimum two cables attached to electrodes on the patient. The placement depends on which lead (eg lead I, lead II, etc) but there's always two minimum.
How do you handle the whole legal elements without turning your app utterly useless? If a ekg sensoradhesive goes loose in the hospital its a alarm event but not a code. So a human has to look and evaluate the data for plausibility and alarm?
Article seems very light on details. Is this trying to detect ECG markers of heart attacks (like ST segment issues)? Is it somehow detecting troponin in the blood stream? How? And how are they going to prevent false positives if this is indeed a wrist-based device as I imagine it will be?
It seems to be using ECG, the (correct) springer link is https://link.springer.com/chapter/10.1007/978-3-031-82377-0_...
Can't read the full article. Abstract mentions 92% accuracy. That could be abysmal depending on how it's calculated? Correctly identifying 92% of heart attacks and missing 8% might be pretty good. But reporting false positives 8% of the time would be awful.
I suspect it's probably worse than that in reality. From a quick search on state of the art ECG results (the full system of leads attached all over your torso) it looks like around 90 percent specificity (True negative rate) and under 50 percent sensitivity (true positive rate). So it's only pretty good at ruling out heart attacks, but still misses them sometimes. But is pretty bad when it comes to false alarms. I think they use it along with multiple other tests and consideration of symptoms in triage at the hospital.
Even 'return false' would best 92% accuracy.
Yes, global afib prevalence is about 1-2%, so return false would be 98-99% accurate, which is why accuracy is not a metric used to assess medical diagnostics.
However, the sensitivity of return false is 0%, which renders it useless (and why sensitivity and specificity are used in this context).
This kind of technology has been technologically viable for decades. The fact that we’re only now seeing prototypes, not mass adoption, is an indictment of the legal framework around medical devices.
The FDA classifies these devices as high-risk because they might give a false result but completely ignores the guaranteed harm of not having them at all. It’s a system that punishes action and rewards delay.
What is burdensome about the regulations? They are here: https://www.fda.gov/medical-devices/guidance-documents-medic...
They allow a summary report. They don't require a clinical human trial. They barely care if you follow the FCC and safety requirements for electronics in general. This does not look burdensome.
This is for arrhythmia detectors which are under a special class 2 restriction, not under the standard class 2 restrictions. Standard class 2 restrictions are more stringent. Secondly, even with arrhythmia detectors, clinical studies will be needed with any new design.
See the "clinical studies" section here:
https://www.fda.gov/medical-devices/guidance-documents-medic...
"designs dissimilar from designs previously cleared under a 510(k), such as those incorporating significant new features or algorithm techniques new technology, i.e., technology different from that used in legally marketed arrhythmia detectors indications for use dissimilar from arrhythmia detector and alarm devices of arrhythmia detectors of the same type."
And those studies aren’t cheap. A Phase II cardiovascular feasibility trial runs roughly US$7 million. If you go pivotal (Phase III), you’re easily in the $11.5 million–$53 million range. Source: https://www.sofpromed.com/what-is-the-cost-of-a-clinical-tri...
We absolutely cannot create a hacker culture of medical devices with these kinds of costs, and in the long run, not having a hacker culture is going to massively slow innovation, and with it, the opportunity to reduce the cost of healthcare and improve the quality.
> Standard class 2 restrictions are more stringent.
Nope! Special class II is all the general class II requirements plus the special rules: https://www.fda.gov/medical-devices/overview-device-regulati...
Specifically in this case I chose this classification because the only difference is the alarm. Without an alarm or alert, you're normal Class II. The alert bumps you up to the special rules. A wearable device would have an alert.
> See the "clinical studies" section
I see where it says "clinical studies will not be needed for most arrhythmia detector and alarm devices [...] FDA will always consider alternatives to clinical testing when the proposed alternatives are supported by an adequate scientific rationale."
They recommend ANSI guidelines and standards for algorithms to measure ST length and for the electrodes. They don't even require following them. The clinical requirements are for all of the crazies who want to do crazy things like sell mood rings for heart attacks. It is very unlikely they would require a premarket clinical trial(NOTE that they allow many products to register as investigative!) for something like an apple watch.
Medical device regulations are a paper tiger. The effort in development is the concern- you can't register until the development is paid for, so if you're rejected then you've wasted all that money. Registering could be zero effort and that would not change. Especially for companies like apple, it is much more convincing that they are more concerned with patient lawsuits than problems with the FDA.
You're right, but it is actually not the general class 2 that electrocardiographs are under, it's Class II (performance standards).
And yes, the FDA will require pre-market clinical data. The FDA required a clinical study for the Apple Watch ECG:
https://www.accessdata.fda.gov/cdrh_docs/reviews/DEN180044.p...
Apple also ran the Apple Heart Study which had 419,000 subjects to validate the algorithm:
https://www.nejm.org/doi/full/10.1056/NEJMoa1901183
Median cost for a U.S. pivotal cardiovascular‑device trial: ≈ $19 million:
https://jamanetwork.com/journals/jamanetworkopen/fullarticle...
You're right that a company like Apple could afford it. However, the type of small company or small group that would form the base of a hacker culture cannot afford these kinds of clinical studies just for an experimental product. We need more experimental products being used in the market to get a faster rate of innovation.
"NOTE that they allow many products to register as investigative!"
Receiving the Investigational Device Exemption does NOT allow the device to be commercially sold or mass-distributed to the general public. The IDE just lets the device be used in human clinical trials. That's it.
There’s no “might” here, in a device advertised as “92% accuracy”. This device _will_ give false results, almost certainly an overwhelmingly number of false results relative to the actual heart attacks it detects. And those false results have real, guaranteed harm and cost as well.
Both false positives and false negatives are harmful. False positives will send people to the hospital for no reason and divert resources from people with real emergencies - not to mention leaving them with a large ER bill to pay. False negatives will result in people with actual heart attacks dismissing their symptoms and dying.
The counter is that people can build their own informed intuition about whether these things are helpful, particularly in coordination with their doctor and experience of using it. And a product is easier to improve and will improve more quickly if the company can easily bring updates and find investment that improve quality to product they have revenue for.
I think the FDA safety margin for things like this should be more “this has no actual obvious harm to use, it has a plausible mechanism of action to help + isn’t fraudulently measuring what it claims to measure and its science backed.
Something like this hits all the targets already.
You’re basically saying "92% accurate" devices will cause harm, so let’s ban them outright. Even if the failure was death, we should allow it. It's a free country. We let consenting adults do all sorts of dangerous stuff — base jumps, do weird diets, drink alcohol, smoke, etc — yet somehow medical device regulation treats personal risk tolerance as if it’s off-limits.
But the failure scenario is not death. It's an unnecessary hospital visit. And you STILL want it banned. If someone wants early heart-attack warnings at 92% accuracy, that should be their call.
As for costs, a few thousand false alarms might clog ERs and bill wallets, but big deal. That’s the price of getting these things in the wild and learning what actually works. More wearables in real world situations equals massive real-world data, faster device improvement, economies of scale, and ultimately, over the long run, millions of prevented heart attacks — plus orders-of-magnitude lower costs down the road. I’d take a handful of extra ER visits any day if it speeds up innovation and saves lives.
>is an indictment of the legal framework around medical devices.
Well medical devices aside, the legal framework around anything, including business, manufacture etc. is more impeding while failing to address things like environmental destruction/pollution which causes real harm. ( notice, that I did not say climate change, a separate subject).
It all makes sense when one sees it either though the lens of either corruption of more likely human stupidity - where a bunch of rules give people the comfort of being protected.
I agree the legal restrictions (and liability) should be looser. But other countries are pretty bad about this too in my experience. It's more of a medical establishment monopoly thing it seems.
Also you can already buy home ECG devices for a couple hundred bucks. Not sure if there is some history of being banned in the past or whatever, but otherwise I'd guess the main problem is just a lack of much interest in the market.
Is the ECG for a heart attack the same for women and men?
The symptoms aren’t.
The most common heart attack symptoms are the same for men and women. The less common, less diagnostic symptoms are more common in women. Over time the medical opinion has swung back and forth (eg thinking women had less pain, until it was recognized that many men have heart attacks without pain) but the current consensus is that the important symptoms are pretty much the same between men and women. Women may be more likely to experience nausea during a heart attack, but you can't use nausea to diagnose a heart attack and men get nauseated a similar amount. These symptoms also don't take into account body types, which have a big impact on the type of attack and symptoms.
Men and women have slightly different ECGs and a doctor can usually tell your gender from an ECG. The appearance of a heart attack will be more similar than a normal heartbeat. Gender differences have a much smaller impact on an ECG than things like body mass and blood pressure. Overworked hearts will look more like overworked hearts.
Awesome