Computer often fails to diagnose atrial fibrillation in ventricular paced rhythm, and that can be catastrophic

Case 1.

An elderly patient presented with a massive hemiplegic ischemic stroke of 24 hours duration.  CT stroke series showed a middle cerebral artery thrombus.

He had an ECG recorded:

Case 1, ECG 1:

Computer: “Electronic Atrial Pacemaker.  Electronic Ventricular Pacemaker”
No other interpretation.
What do you think?

Answer: There is no atrial pacing spike.  This is atrial fibrillation (not diagnosed by the computer) but with a regular ventricular rhythm because the ventricle is regularly paced by an artificial pacemaker (ventricular paced rhythm, VPR).  This may easily deceive both the computer and the overreading physician because one of the primary means of identifying atrial fibrillation is by an irregularly irregular heart rate.

Case 2

This patient presented on the same day: She had a history of atrial fibrillation, CHF, COPD, HTN, CVA anticoagulated on Coumadin who presents for evaluation of shortness of breath.

Case 2 ECG (ECG 2):

Computer: “electronic ventricular pacemaker”
What do you think?

There is no mention of sinus rhythm, or any atrial rhythm diagnosis, even though sinus rhythm is obvious.  There is also, of course, VPR and PVCs.

Case 1, later in the day (ECG 3):

Now there is an irregular rhythm and so it is obviously atrial fib.  Because the AV node is conducting rapidly enough, the ventricle no longer needs to be paced, and there is an irregular rhythm.
There are also CNS T-waves (which is due to stress cardiomyopathy, common in acute severe stroke, and verified by multiple wall motion abnormalities on formal echo).

There were previous ECGs available for case 1 (the stroke patient):

Case 1, first previous ECG (ECG 4)

Computer and physician overread are:
“Electronic Atrial Pacer”
“Electronic Ventricular Pacer”
What do you think?

Again, there are no atrial pacer spikes.  There is probably underlying atrial fibrillation but neither the computer nor the physician diagnosed it.  The physician seemed to believe that if the computer detected it, it must be there. But this may not be an accurate assumption.

Another second previous ECG (ECG 5):

Computer and Physician overread:
“Ventricular Paced Rhythm.  Artifact makes interpretation difficult.”

There are many little spikes.  These are artifact, and have nothing to do with a cardiac pacemaker. Sometimes you can see similar artifact from central nervous system devices.   Thus, ECG 5 also has atrial fibrillation that went undiagnosed.


Thus there are many failures of atrial rhythm diagnosis here, by both the computer and the overreading physician, because of the presence of VPR:

ECG 1 has atrial fibrillation (undiagnosed), not atrial pacing
ECG 2 has sinus rhythm, which is not mentioned.
Only ECG 3 has atrial fibrillation correctly diagnosed because there are no ventricular pacing spikes and the rhythm is irregular.
ECG 4 has atrial fibrillation misdiagnosed as an atrial pacer
ECG 5 has undiagnosed atrial fibrillation

Thus, for Case 1, 2 previous ECGs had failed to diagnose atrial fibrillation.  In one, the diagnosis was incorrect: “Electronic Atrial Pacemaker” diagnosed by both computer and overreading physician.  In the other case, no atrial diagnosis was given by either computer or physician.

Thus, the case 1 patient was never diagnosed with atrial fibrillation, even though it was present, and the atrial fibrillation ultimately resulted in catastrophic stroke.

This failure to diagnose atrial fib by the automated algorithm is very common.  In the setting of VPR, many conventional algorithms do not even attempt an atrial rhythm diagnosis.  And this leads physicians astray.  Below are a couple important articles on the topic.  There is a large literature on misdiagnosis of atrial fib by conventional computer algorithms.

Moreover, when a conventional algorithm does diagnose atrial fibrillation, it is an overdiagnosis in up to 20% of cases.  It is very common for these computer errors to go uncorrected by the overreading physician.

Learning Points:

1. All atrial rhythm diagnoses by the computer algorithm must be scrutinized.
2. The computer algorithm may not even attempt an atrial rhythm diagnosis in the presence of VPR.
3. Atrial fibrillation in the presence of VPR will usually have a regular rhythm, and thus be difficult to discern.
4.  Even a computer diagnosis of “atrial paced rhythm” may be a false positive.

Diagnostic performance of a computer-based ECG rhythm algorithm

https://www-sciencedirect-com.ezp2.lib.umn.edu/science/article/pii/S0022073605000488

Of the 4297 consecutive ECGs forming the basis of this report, 13.1% (565/4297) required revision of the computer-based rhythm interpretation. The most common errors were related to interpretive statements involving patients with pacemakers: of 343 ECGs with pacemaker activity comprising 8.0% of the study ECGs, 75.2% (258/343) required revision, so that 45.7% of all inaccurate rhythm statements in this population occurred in patients with pacemakers.The most common error in this subgroup was failure of the algorithm to detect evident underlying rhythms, such as sinus rhythm with dual chamber pacing or underlying atrial fibrillation, in 40.2% (138/343). Dual chamber pacing was an evident problem, with a specificity of 100% but a sensitivity of only 28.1% (73/260), most often caused by relatively minor mis-identification of these patients as simply ventricular paced. More important was complete failure to identify pacemaker activity in 10.2% of the paced patients (35/343). Because special diagnostic problems and special engineering solutions apply to pacemaker detection algorithms, these patients were eliminated from further data analysis, which focuses on the remaining 3954 consecutive unpaced ECGs in the group. Predominant physician-confirmed primary rhythms in this unpaced patient population include sinus rhythm (90.5%), atrial fibrillation (6.3%), atrial flutter (1.0%), and atrial tachycardia (0.9%).


Misdiagnosis of atrial fibrillation and its clinical consequences

https://www.sciencedirect.com/science/article/pii/S0002934304004784

PurposeComputer algorithms are often used for cardiac rhythm interpretation and are subsequently corrected by an overreading physician. The purpose of this study was to assess the incidence and clinical consequences of misdiagnosis of atrial fibrillation based on a 12-lead electrocardiogram (ECG).  MethodsWe retrieved 2298 ECGs with the computerized interpretation of atrial fibrillation from 1085 patients. The ECGs were reinterpreted to determine the accuracy of the interpretation. In patients in whom the interpretation was incorrect, we reviewed the medical records to assess the clinical consequences resulting from misdiagnosis.   ResultsWe found that 442 ECGs (19%) from 382 (35%) of the 1085 patients had been incorrectly interpreted as atrial fibrillation by the computer algorithm. In 92 patients (24%), the physician ordering the ECG had failed to correct the inaccurate interpretation, resulting in change in management and initiation of inappropriate treatment, including antiarrhythmic medications and anticoagulation in 39 patients (10%), as well as unnecessary additional diagnostic testing in 90 patients (24%). A final diagnosis of paroxysmal atrial fibrillation based on the initial incorrect interpretation of the ECGs was generated in 43 patients (11%).  Conclusion:  Incorrect computerized interpretation of atrial fibrillation, combined with the failure of the ordering physician to correct the erroneous interpretation, can result in the initiation of unnecessary, potentially harmful medical treatment as well as inappropriate use of medical resources. Greater efforts should be directed toward educating physicians about the electrocardiographic appearance of atrial dysrhythmias and in the recognition of confounding artifacts.

Powered by WPeMatico