Why are we still interpreting cardiac rhythms?

[MMiz]

Whenever I see a cardiac strip posted online, in this community or elsewhere, I note how different the interpretations are by so many qualified people.

With AI, deep learning models, and so much technology, why aren't we better leveraging technology to assess cardiac rhythms? Why in 2021 are paramedics still interpreting cardiac rhythms?

 

[NPO]

Whenever I see a cardiac strip posted online, in this community or elsewhere, I note how different the interpretations are by so many qualified people.

With AI, deep learning models, and so much technology, why aren't we better leveraging technology to assess cardiac rhythms? Why in 2021 are paramedics still interpreting cardiac rhythms?

Because the algorithms are wrong enough of the time that we can't trust them 100%.

 

[EpiEMS]

With AI, deep learning models, and so much technology, why aren't we better leveraging technology to assess cardiac rhythms? Why in 2021 are paramedics still interpreting cardiac rhythms?

I love this question! I don't know the answer and I am so not qualified to interpret rhythms, but I have been involved in my share of AI and automation work, so I can lend a little insight.

In a financial services context, there is an immense amount of scrutiny that has commenced in recent years around automated decisionmaking tools, such as trading models and credit decision tools. Part of this is because models - particularly without skilled oversight - not infrequently run into issues like overfitting to the training dataset, errors in use (e.g., applying a model to the wrong circumstances), or an inability to adjust to changing conditions (for example, a model designed based on a certain market environment that doesn't "kill" itself when conditions change).

In short, models aren't perfect -- they are best used as a guide or subject to skilled oversight, even if they've been tested and retested.

So, what does this mean for cardiac rhythm analysis?

Well, I think it's about the general lack of concordance in expert reads of ECGs, for one. Combine that with the problems of technology application, as well as legal and ethical concerns about machine diagnosis (namely, failure to diagnose & who is accountable) given the tech issues themselves, I am not so surprised that there is a need for an art as well as a technical solution.

 

[EpiEMS]

Because the algorithms are wrong enough of the time that we can't trust them 100%.

And to that point, I don't think we can trust them even close enough to 100%, ethically speaking, to make a clinical decision.

 

[CALEMT]

Whenever I see a cardiac strip posted online, in this community or elsewhere, I note how different the interpretations are by so many qualified people.

With AI, deep learning models, and so much technology, why aren't we better leveraging technology to assess cardiac rhythms? Why in 2021 are paramedics still interpreting cardiac rhythms?

If you have the knowledge to change your car oil, why pay someone to do it?

I've seen LP15's interpret a regular sinus rhythm as a STEMI. My X series is pretty consistent but I still interpret every 12 lead I run. Why? Because I have the knowledge to do so.

 

[EpiEMS]

If you have the knowledge to change your car oil, why pay someone to do it?

I've seen LP15's interpret a regular sinus rhythm as a STEMI. My X series is pretty consistent but I still interpret every 12 lead I run. Why? Because I have the knowledge to do so.

I agree with your second point, but I am not so convinced about the first. If there is less knowledge required, or at least if there can be cognitive offload to a machine (contingent on it being accurate...), I think that is an advantage system wide.

 

[Aprz]

I totally thought you were gonna go somewhere different with this. I thought you were gonna ask why do we interpret when so many people say "It doesn't matter. What matters is if it is sloooow or fast? Are they stable or unstable?" Lol.

 

[mgr22]

I think there are too many patient and equipment variables to trust machine algorithms more than we do now. In my opinion, human oversight will need to accompany automated rhythm interpretation indefinitely -- sort of like pilots overseeing glass cockpits.

 

[CALEMT]

I agree with your second point, but I am not so convinced about the first. If there is less knowledge required, or at least if there can be cognitive offload to a machine (contingent on it being accurate...), I think that is an advantage system wide.

I'm a firefighter/paramedic its a miracle that I know what analogy means, executing a good analogy on the other hand...

Interestingly enough to the topic at hand I have a pt today C/O difficulty breathing with associated chest pressure. On my monitor the pt was anywhere from 60-150 on the rate. There was artifact on my monitor (Zoll X series) so I was having trouble finding P waves. Got a brief picture saw that it was in fact a sinus rhythm, just irregular AF. I didn't run a 12 lead becuase the pt was sitting on the tailboard of the engine. In the back of the ambulance on a different monitor (LP15) the 12 lead was reading inferior MI. Lead 2 on the 12 showed depression, on the monitor lead 2 wasn't showing depression. My X series also showed no depression on lead 2. My and the transporting medic's interpretation of the 12 was sinus arrthymia. There was no elevation anywhere.

I guess what I'm getting at is why solely rely on a technological based algorithm that is wrong more times that it is right? I don't view it as a crutch, more as a last ditch tool for when I get those 12 leads that are littered with FLB's (funny looking beats) where I have no idea what the hell is going on. At least I can look at the top of the 12 and it more than likely gets my train of though going in a different direction.

 

[Carlos Danger]

Interesting question. As much as the conversation so far has revolved around artifact and examples of poor interpretation by current machines, I have little doubt that software could be developed that far exceeds the ability of most individuals to accurately interpret rhythms or MI criteria.

Like so many other things, it probably just comes down to how much $$ you are willing to spend.

Edit: this isn’t working on my phone. I meant to reply directly to the OP and somehow ended up replacing CALEMT’s text.

 

[wtferick]

My main concern is why do Paramedics transport elevated trop levels to non cardiac receiving hospitals? Lol

 

[DrParasite]

Because the algorithms are wrong enough of the time that we can't trust them 100%.

Why are they wrong? is not the critera for a stemi the same on the east coast as on the west coast ( as per https://ecgweekly.com/classic-stemi-criteria/
Fourth Universal Definition of MI (Thygesen et al., 2018):

New or presumed new, ST-segment elevation (STE) ≥ 1.0 mm (measured at the J-point in 2 contiguous leads) is required in all leads (except V2, V3, V3R, V4R, V7-V9))? ditto a LBBB? benign early repolarization? and a slew of other things? I mean, all paramedics learn from the same book right? Afib is afib is afib right? and it's different from aflutter via objective criteria.

If I give a rhythm strip to 3 cardiologists are they going to have the same diagnoses? isn't interpreting an ECG an objective activity? if the monitor says STEMI, what is it triggering off of? Artifact can screw stuff up, but if you are putting two monitors on a person 5 minutes apart, and they are showing different things, how accurate is the $15,000 monitor (assuming no weird medical conditions)?

And is this really accurate? https://www.tiktok.com/@drglaucomflecken/video/6934839183609433349

While I understand having human oversight (and agree it's needed), and not treating strictly off a machine, shouldn't the machines be able to objectively weird stuff with a heart rhythm?

 

[EpiEMS]

If I give a rhythm strip to 3 cardiologists are they going to have the same diagnoses? isn't interpreting an ECG an objective activity?

I don't think they will, at least, a significant fraction will be off. For example, in one study from 2013, it was found that:

One hundred twenty‐four physicians interpreted a total of 4392 ECGs. Among all physicians, interreader agreement (kappa) for ECG interpretation was 0.33, reflecting poor agreement. The sensitivity to identify “true” STEMIs was 65% (95% CI: 63 to 67) and the specificity was 79% (95% CI: 77 to 81).

This is specifically for ECGs without clinical information in the setting of ST-elevation, and these were all real ECGs where a patient was cathed.

While I understand having human oversight (and agree it's needed), and not treating strictly off a machine, shouldn't the machines be able to objectively weird stuff with a heart rhythm?

One would think so, but I have seen enough literature to be convinced that the machine is not 100% trustworthy, at least not yet to the point where it can be relied on blindly.

 

[DrParasite]

One would think so, but I have seen enough literature to be convinced that the machine is not 100% trustworthy, at least not yet to the point where it can be relied on blindly.

Is the machine not 100% trustworthy because the objective criteria aren't as objective as we would like them to be, or because the cardiologist who are interpreting can't agree on what constitutes a STEMI (using your examples among MDs that there was poor agreement over what constituted a STEMI)?

 

[ffemt8978]

Is the machine not 100% trustworthy because the objective criteria aren't as objective as we would like them to be, or because the cardiologist who are interpreting can't agree on what constitutes a STEMI (using your examples among MDs that there was poor agreement over what constituted a STEMI)?

Does it really matter in deciding if the machine is capable of being trusted to interpret the rhythms 100% accurately, or is it more of a reflection of our inability to accurately interpret what we see?

 

[NPO]

Is the machine not 100% trustworthy because the objective criteria aren't as objective as we would like them to be, or because the cardiologist who are interpreting can't agree on what constitutes a STEMI (using your examples among MDs that there was poor agreement over what constituted a STEMI)?

It's several reasons. The largest factor in my experience is artifact. They don't seem to factor artifact at all.

No Life Pack 15, my patient is not atrial paced irregularly at 300 times per minute. She's just got Parkinson's.

 

[Aprz]

Have they done any studies with EPs? I feel like cardiologists do not study ECGs as much as EPs and emergency physicians or physicians in general would not be good to study agreement and accuracy of interpreting ECGs. We always bring up cardiologists because it would make sense that a cardiologists would know ECGs well, but I don't think it is actually their specialty. From what I've seen EPs tend to agree with each other, are more specific, and have methods to confirm rhythms.

 

[BobBarker]

My main concern is why do Paramedics transport elevated trop levels to non cardiac receiving hospitals? Lol

I can't tell you the amount of times my father's trops were high from missing dialysis and fluid overload. He never went to a cardiac center to fix it.

 

[wtferick]

I can't tell you the amount of times my father's trops were high from missing dialysis and fluid overload. He never went to a cardiac center to fix it.

My apologies. I should have been more specific. Turns out many Paramedics have no clue what troponin is.

 

[NPO]

My apologies. I should have been more specific. Turns out many Paramedics have no clue what troponin is.

Nor do we have the ability to check it in the field (usually).