Monday, June 12, 2017

Should Emergency Physicians be interrupted by ECGs that are read as "Normal" by the computer?

Here was a comment on Facebook about this post:

"I was hoping this post would let me off the hook for all those "normal ekg" reading and but no, they can't replace us yet..."

Here was my response:

"Be happy that you can't be replaced -- yet! Though stressful, it is good to be needed. It will be more stressful when computers replace us and we're out of work. I am working with a company that writes deep neural network software for interpreting ECGs. We're only beginning and we're already better than conventional algorithms. It is a matter of time."

This was sent by:

Jacob Smith, DO
Emergency Medicine Resident
Ohio Health Doctors Hospital Emergency Residency

Christopher Lloyd, DO, FACEP
Director of Clinical Education, USACS Midwest

Case

A 30 year old patient presents to triage with chest pain.  An ECG is recorded and the computer reads it as "normal".

Do you want to see it?

Or would you rather not be bothered?







I'd rather be bothered.

Here it is:

Computer read: "Normal ECG"
Aren't you glad that you looked at it?
Otherwise the patient could wait in triage for hours.

Unknown algorithm

















Interpretation: Inferior ST elevation, with reciprocal ST depression in aVL.  This is diagnostic of inferior MI, though does not meet millimeter criteria for "STEMI."  There is also ST elevation in lateral precordial leads V5 and V6.

Years later, I sent it to the Queen:

The Queen of Hearts PM Cardio App is now available in the European Union (CE approved) the App Store and on Google Play.  For Americans, you need to wait for the FDA.  But in the meantime:

YOU HAVE THE OPPORTUNITY TO GET EARLY ACCESS TO THE PM Cardio AI BOT!!  (THE PM CARDIO OMI AI APP)

If you want this bot to help you make the early diagnosis of OMI and save your patient and his/her myocardium, you can sign up to get an early beta version of the bot here.  It is not yet available, but this is your way to get on the list.




Here is the history:

A 30 yo man presented complaining of severe chest pain.  He had a family history of early CAD and occasional drug and tobacco use.

The ECG was alarming to the ED physician who did indeed review it.  He was worried for inferior MI and ordered another, which was recorded 15 minutes later:
Now clearly and obviously diagnostic of inferior STEMI.  It is not subtle any more.

Interventional cardiology was consulted and patient was taken to the cath lab. He was found to have a 100% circumflex lesion for which a bare metal stent was placed.

Imagine if this patient had been at a busy triage and they trusted the computer interpretation.  He may have waited hours before being seen.

Comment

A recently published article (1) found that a computer-interpreted “normal” ECG has a 99% NPV (95% CI: 97-99) for a clinically significant finding. They suggest (but did not conclude) that immediate EKG review by a physician could be safely eliminated when the computer interpretation is normal.

Those confidence intervals are important.  They mean that, in reality, there is a good chance that a computer-interpreted "normal ECG" misses 3% of significant abnormalities.

This study was far too small (n = 222 "normal" ECGs) to conclude that ECGs do not need to be read by a human.

This is probably why the authors only "suggest" and do not "conclude" this.

According to 2 recent articles (2, 3), the computer misses one third of STEMI, but misses many more subtle MI that do not meet STEMI criteria.  Most of these will not be "normal" but will have nonspecific ST-T abnormalities, or other features which the computer recognizes as "abnormal" but does not diagnose.

However, subtle coronary occlusion may be completely missed by the computer and called "normal."  


This one was not even so subtle!!!

References:

1. Katie E. Hughes KE., Scott M. Lewis SM., Laurence Katz and Jonathan Jones  Safety of Computer Interpretation of Normal Triage Electrocardiograms (pages 120–124).  Academic Emergency Medicine 24(1):120-124. January 2017. 
 http://onlinelibrary.wiley.com/doi/10.1111/acem.13067/full


Results: A total of 855 triage ECGs were collected over 16 weeks. A total of 222 (26%) were interpreted by the computer as normal. The negative predictive value for a triage ECGs interpreted by the computer as “normal” was calculated to be 99% (95% confidence interval = 97% to 99%). Of the ECGs with a computer interpretation of normal ECG, 13 had an interpretation by an attending cardiologist other than normal. Two attending EPs reviewed these triage ECGs. One of the 13 ECGs was found to have clinical significance that would alter triage care by one of the EPs. The stated triage intervention was “bed immediately.”


2. Mawri S, Michaels A, Gibbs J, et al. The Comparison of Physician to Computer Interpreted Electrocardiograms on ST-elevation Myocardial Infarction Door-to-balloon Times. Critical Pathways in Cardiology 2016;15:22-5.

3. Garvey JL, Zegre-Hemsey J, Gregg RE, Studnek JR. Electrocardiographic diagnosis of ST segment elevation myocardial infarction: An evaluation of three automated interpretation algorithms Journal of Electrocardiology 2016;49:728-32.





1 comment:

DEAR READER: I have loved receiving your comments, but I am no longer able to moderate them. Since the vast majority are SPAM, I need to moderate them all. Therefore, comments will rarely be published any more. So Sorry.