[pragone]

Determining PVCs vs afib is very trivial when you actually look at the strip. However, if Apple is only analyzing rate regularity and not actually doing analysis on the waveforms themselves they would be very difficult to differentiate.

If you're having so many PVCs that the watch thinks you're in afib, you should probably still see a doctor. As the grandparent noted, tens of thousands of PVCs in a couple day period is very unusual and should probably be seen by a specialist.

[bayesian_horse]

No it's not. It's trivial to recognize a PVC if you see one, but it's a lot harder to rule out AFib.

To recognize a PVC at the frequency mentioned, you'd need to see much less than a minute of ECG data. You could see AFib in the same length of data, if the data is of good quality, but it can occur episodically, so the 24 Hour ECG is the right call.

I have only limited cardiological knowledge, especially regarding Humans, but at least in my model of cardiac pathology PVCs and AFib aren't mutually exclusive.

[pragone]

Sorry, I may not have been clear.

The point I was trying to make is that PVCs and afib are not mutually exclusive - they're just not related entities. The parent brought up sinus rhythm, and I wanted to make sure everyone was on the same page regarding the fact that PVCs and afib are very different things that can seem similar if the only data point you look at is regularity of the rhythm.

[bayesian_horse]

Yes, maybe just looking at RR intervals maybe. Otherwise AFib leads to irregular undulations, at about the amplitude of a p-wave, whereas PVCs are longer and weird looking QRS complexes.

AFib is a "supraventricular" Tachycardia, whereas PVCs are more of a ventricular Tachycardia. Though I'm not sure if AFib always leads to significant Tachycardia at all.

[Unklejoe]

The reason I brought up “normal sinus rhythm” and “atrial fibrillation” a few posts up is because those are two of the four possible results produced by the Apple Watch EKG app, the other two being “low and high heart rate” and “inconclusive”.

[Unklejoe]

Yeah, I’m currently going through the rounds of testing - just had the echocardiogram a week ago. Fortunately, that turned out normal. They gave me metoprolol, but that doesn’t seem to help. My potassium was kind of low, so they gave me a supplement for that, but it didn’t help either. If these PVCs keep up at the current rate, I suspect they will suggest ablation. Not really looking forward to that...

Anyway, I’m not sure exactly how Apple is analyzing the data. I’ve performed probably 15 EKGs with only one or two PVCs present, and it reported a normal sinus rhythm every time. It only reported afib once, and there were quite a few PVCs in that capture. I can email it to anyone interested.

Apparently, occasional (a few a day) PVCs are very common, so I imagine Apple had to encounter this situation during development.

[pragone]

So it's worthwhile to make sure we all understand that we're talking about two different things: atrial fibrillation and sinus rhythm are two options in the same bucket: rhythm. You can have PVCs with afib, you can have PVCs with sinus rhythm. You can't simultaneously be in afib and sinus rhythm.

I'm guessing Apple is just looking at rhythm - it would be the most reliable datapoint, as opposed to trying to measure QRS duration or analyzing for presence of P-waves, both of which differ depending on which lead you're looking at. The lead being looked at depends on which two points on the body are used to produce the tracing as well, which is not something Apple can guarantee, so perhaps that's where the explanation lies, but I have no evidence or data to back up that conclusion.

[brlewis]

To be clear, you're transitioning to a related topic, not arguing for "The classifier can't get better with only one or two contacts", right?

[pragone]

Right.

I would argue it is possible for better classification; I don't have a reason why apple chose not to do this (especially if they're capturing the data) other than the algorithms are not as guaranteed as we would like and they don't want to assume that kind of liability. (ECG printouts include the computer's interpretation; a standard way of reading them is to completely ignore whatever the computer says, because it is not infrequently completely wrong. And those are the professional grade machines with 12-leads capturing 10 different points on the body)

Edit; As I thought more about this, I realized that analyzing QRS duration or presence of p-waves can be lead dependent, which can change based on which parts of the body are being used to generate the lead.