In this ECG Cases blog we review 10 patients with a variety of presentations. Can you spot the computer interpretation error and prevent the misdiagnosis?

Written by Jesse McLaren; Peer Reviewed and edited by Anton Helman. April 2022

Can you spot the computer interpretation error and prevent the misdiagnosis?

Case 1: 75yo with weakness. HR 35, BP 140

Case 2: 80yo with an unwitnessed fall, normal vitals. These two ECGs were taken 30 seconds apart, one measuring the QTc at 616 (requiring immediate treatment to prevent polymorphic VT), the other at 493. Which is correct, and why is the other incorrect?

Case 3: 30 year old previously healthy with shortness of breath and syncope. R18 sat 96% HR 110 BP 140/100

Case 4: 80 year with old syncope while sitting. Normal vitals

Case 5. 50yo with resolved palpitations, normal vitals

Case 6: 50yo previously healthy with exertional chest pain. Went to GP then sent to ED. ECG from GP’s office and on arrival in ED where pain had lessened

Case 7: 50yo with throat pain, normal vitals

Case 8: 60yo with chest pain, normal vitals

Case 9: 50yo with chest pain, normal vitals. Old then new ECG

Case 10: 60yo with two weeks of exertional chest pain, currently pain free with normal vitals. Old then new ECG:

ECG Computer interpretation errors

Computer interpretation of the ECG (CIE) has been called a double-edged sword: when correct, it increases physician accuracy, but when incorrect it increases errors.[1,2] This is especially problematic in the emergency department, where computer accuracy drops as clinical significance increases—with common errors for arrhythmias and ischemia. [3] We also have to be skeptical of the final interpretation recorded on the ECG, because overreading cardiologists not involved in the patient’s care are more likely to accept the computer interpretation and as a result have lower accuracy.[4] As a review summarized: “computer-based analysis of the ECG may lead to erroneous diagnosis with useless, inappropriate, or even dangerous care of the patient…Knowledge of the strengths and limitations of CIE is a prerequisite for avoiding blind trust in software interpretations”[5] The strength of computers is they can measure individual values according to specific parameters–heart rate, intervals, axis, amplitude, millimeters of ST elevation. But the limitation is that computers don’t actually “interpret”, they simply label the individual abnormalities they measure. But the parameters and paradigm on which they are based are limited, they can’t compare to old or serial ECGs, and they don’t consider the patient. It’s up to us to systematically analyze the ECG, integrate the different parameters, go beyond the false dichotomy of STEMI criteria, arrive at a final interpretation, and apply it in clinical context. Knowing what errors computers make, and why, can also help improve our own interpretation.

Heart rate/rhythm: The first source of computer error is rhythm interpretation. As one study found, the computer accuracy was 95% for sinus rhythm but only 53.5% for non-sinus rhythm (including atrial fibrillation/flutter, ectopic atrial rhythm, junctional rhythm, paced rhythms and ventricular rhythm).[6] This is because small amplitude deflections between QRS complexes can include P waves, T waves, U waves, flutter waves, pacemaker spikes or artifact. This can lead to overdiagnosing atrial fibrillation (for example sinus arrhythmia or PACs), underdiagnosing atrial flutter (calling it sinus tachycardia), or double counting the heart rate if T waves are peaked or hyperacute. The best leads to identify sinus rhythm are lead II (upright P wave, because depolarization from the sinus node in the upper right atrium travels down/left towards lead II) or V1 (biphasic P wave from sinus activity that depolarizes the right atrium towards V1 and left the left atrium away from V1).

Electrical conduction: computers are usually accurate in measuring the intervals of electrical conduction—the PR, QRS and QT. But QT interval measurements are not always reliable [7]. Computers also can’t tell why intervals are abnormal or integrate multiple abnormalities: for example, prolonged PR and QRS intervals and peaked T waves from life-threatening hyperkalemia might be labeled ‘first degree AV block” and “non-specific intraventricular delay.”

Axis: computers are usually accurate in identifying axis deviation, but there are still limitations. A new S wave in lead I from PE with acute RV strain won’t be recognized unless it exceeds the amplitude of the R wave to become fully right axis deviated. And while computers can identify left axis deviation they will not always indicate the etiology—which can lead to missing a left anterior fascicular block, which could be relevant in the context of syncope.

R-wave progression: analyzing the R-wave progression in the horizontal plane is just as important as the axis in the frontal plane, yet computer interpretation doesn’t seem to look for this. While computers often identify right bundle branch block, they will often miss other causes of early R wave progression—from left sided WPW to posterior MI. And while they can identify anterior Q waves from old anterior MI, computers will often miss other causes of late R wave progression like loss of R waves from acute MI.

Tall/small voltages: voltage amplitude affects the ST segment and its interpretation. Large voltage complexes with proportional ST elevation from LVH or BER can be falsely labeled “STEMI”[8], or computers can miss signs of Occlusion MI in the context of LVH.

ST/T changes: Computers will often identify obvious STEMIs but fail at identifying subtle occlusions, which contributes to delayed reperfusion. In a study of Code STEMIs, 44.6% of initial ECGs were not labeled “STEMI” and these were associated with prolonged ECG-to-Activation time; the greatest limitation was not the computer failing to apply STEMI criteria but STEMI criteria themselves, as a quarter of initial ECGs were “STEMI negative” but diagnostic of Occlusion MI.[9] The computer algorithms and the STEMI paradigm on which they are based can’t differentiate between different causes of ST elevation (eg LAD occlusion, early repolarization and pericarditis), occlusion in the presence of pre-existing ST changes (eg LBBB, LV aneurysm), or occlusion which doesn’t meet STEMI criteria. As a result computers can apply the label of “normal” to ECGs diagnostic of Occlusion MI.[10] Computers can also fail at identifying reperfusion T wave inversion, which can be crucial to identifying patients at risk for re-occlusion.

So how should we use the computer interpretation? According to ECG expert Dr. Grauer, this depends on your level of ECG interpretation: “For the expert interpreter: review the computer report either before or after evaluation of the ECG itself. Minimize time devoted to determination of heart rate, intevals and axis (since the computer is very accurate for these parameters). Consider more careful evaluation if the rhythm is not sinus, or if the ECG is interpreted by the computer as abnormal. Overread each computer statement. Place a check mark next to those that are accurate. Delete, modify, or add to incorrect statements…The most important step for the non-expert is to first cover up the computerized report. It is otherwise all too easy to be biased by what the computer says. Used in this way, comparing one’s own interpretation with what the computer says optimally incorporates potential benefit from any discrepancy in interpretation that may exist.”

Back to the cases

Case 1: atrial flutter mislabeled as sinus bradycardia with inferior ischemia

  • Heart rate/rhythm: atrial flutter (fully upright atrial activity in V1 from flutter waves, negative flutter waves in II) with slow ventricular response
  • Electrical conduction: normal QRS and QT
  • Axis: borderline right
  • R-wave: anterior Q waves from old infarct
  • Tall/small voltages: normal voltages
  • ST/T: no ST/T waves changes – inferior flutter waves misinterpreted as ischemia

Impression: atrial flutter with slow ventricular response. Admitted for pacemaker

Case 2: QT mislabeled as long because of prominent U wave

  • H: normal sinus rhythm with PAC
  • E: normal PR and QRS; by inspection the QT is less than 500 (2.5 large boxes), but there’s a prominent U wave in V2 that misleads the computer
  • A: normal axis
  • R: normal R wave progression
  • T: normal voltages
  • S: inferolateral ST depression

Impression: QT is less than 500. ST depression with small T waves and prominent U waves can be seen with hypokalemia but the potassium and other electrolytes were normal, and these repolarization changes were old.

Case 3: Classic constellation of ECG signs for PE, missed by the computer

  • H: sinus tach
  • E: normal conduction
  • A: borderline right axis with prominent S wave in I
  • R: delayed R wave progression
  • T: normal voltages
  • S: antero-inferior T wave inversion

Impression: the computer is technically correct–there is sinus tach, normal axis and T wave abnormalities–but this is not very helpful. Sometimes computers will label this “T wave abnormalities consider anterior ischemia” and “T wave abnormalities consider inferior ischemia”, which more accurately identifies the individual abnormalities but does not integrate them. The combination of sinus tachycardia, antero-inferior T wave inversion and a S wave in I is classic for RV strain, which in the context of acute SOB/syncope is PE until proven otherwise. CT chest: saddle PE.

Case 4: bifascicular block with borderline first-degree heart block, labeled “RBBB with left axis”

  • H: sinus bradycardia
  • E: borderline 1st degree AV block + left anterior fascicular block + RBBB
  • A: left axis from LAFB
  • R: Tall R in V1 from RBBB, ongoing S wave in lateral leads from LAFB
  • T: borderline LVH criteria in context of LAFB
  • S: no ST/T changes

Impression: technically the computer is correct. But the PR interval is at the upper limit of normal, the left axis deviation is from LAFB, and the combination of 1st degree AV block + LAFB + RBBB + syncope = admission for pacemaker.

Case 5. missed early R wave progression from WPW

  • H: normal sinus rhythm
  • E: short PR with delta wave and slurred QRS; normal QT
  • A: normal axis
  • R: early R wave progression from left sided WPW
  • T: normal voltages
  • S: mild anterior ST depression, appropriately discordant to abnormal QRS

Impression: WPW

Case 6: loss of precordial R waves and fragmented QRS from acute LAD occlusion, labeled ‘normal’

  • H: borderline sinus tach
  • E: normal conduction
  • A: physiological left axis
  • R: loss of precordial R waves and fragmented QRS complexes in V4-5 and inferiorly
  • T: normal voltages
  • S: first ECG has mild ST elevation V2-3 and hyperacute T waves V2-5, reduced in the second ECG

Impression: LAD occlusion with loss of anterior R waves, resolution of ST elevation and hyperacute T waves but no reperfusion T wave inversion. This was missed and the patient observed with serial troponins. First Trop I was elevated at 75, and after repeat was 500 the patient was started on antiplatelets and referred to cardiology. Repeat ECG showed persisting upright T waves from ongoing occlusion

Patient admitted as “NSTEMI”. When repeat trop increased to 5000 the patient was treated with nitro. Next day angiogram revealed 99% mid LAD occlusion with TIMI 1-2 flow. Door-to-balloon time was 22 hours. Peak trop was 18,000 and EF reduced to 45% with mid anterior and apical regional wall abnormality. Discharge ECG had anterior Q waves with reperfusion T wave inversion.

Case 7: LVH with secondary repolarization abnormalities, labeled ‘STEMI’

  • H: sinus arrhythmia
  • E: normal conduction
  • A: normal axis
  • R: normal R wave progression
  • T: tall voltages from LVH
  • S: ST and T waves changes appropriately discordant and proportional to large QRS

Impression: LVH with secondary repolarization abnormalities. Negative workup

Case 8: missed LAD occlusion in the presence of LBBB

  • H: sinus arrhythmia
  • E: normal PR, LBBB
  • A: left axis from LBBB
  • R: poor W wave progression from LBBB
  • T: normal voltages
  • S: concordant STE I/aVL/V5-6 and excessive discordant STE V3-4 (STE/S>25%)

Impression: LBBB with anterolateral ST elevation from proximal LAD occlusion. Taken directly by EMS to cath lab: 100% proximal LAD occlusion, peak trop 50,000. Discharge ECG had resolution of changes

Case 9: LAD occlusion labeled ‘early repolarization’

  • H: normal sinus rhythm
  • E: normal conduction
  • A: normal axis
  • R: normal R wave progression
  • T: normal voltages
  • ST: I/aVL and V2-5 have new non-concave ST elevation and hyperacute T waves; V3 has terminal QRS distortion (i.e. it has lost its S wave, and not because of a J wave) which excludes early repolarization; and there’s T wave inversion in V6/III

Impression: proximal LAD occlusion. Cath lab activated: 100% proximal LAD occlusion. First trop I was 30ng/L (just above normal) and peak was 28,000. Discharge ECG had reappearance of S wave in V3 and reperfusion T wave inversion in the anterolateral leads

Case 10: RCA reperfusion labeled ‘normal’

  • H: normal sinus rhythm
  • E: normal conduction
  • A: normal axis
  • R: new early R wave progression with R>S in V2
  • T: normal voltage
  • S: new T wave inversion in III/aVF

Impression: inferoposterior OMI with reperfusion (analogous to inferior Wellens syndrome), identified at triage. First trop 100, referred to cardiology. Admitted as “NSTEMI” with peak trop 1,000. Angiogram 3 days later (fortunately without any re-occlusion) confirmed ECG: 95% RCA occlusion. Discharge ECG the same.

Take home points for beware computer interpretation errors

Heart rate/rhythm: beware non-sinus rhythms, look at leads II and V1 for atrial activity

Electrical conduction: inspect the QT and consider causes of multiple blocks

Axis: verify the axis and consider differentials

R-wave progression: look for early or late R wave progression and consider its causes

Tall/small voltages: assess voltages to contextualize ST segments

ST/T waves: learn the signs of occlusion and reperfusion

References for ECG Cases 30 Beware Computer Interpretation Errors

  1. Martinez-Losas P, Higueras J, Gomez-Polo JC. The computerized interpretation of the electrocardiogram: a double-edged sword? Enferm Clinica 2017 Mar;27(2):136-7
  2. Martinez-Losas P, Igueras J, Gomez-Polo CJ, et al. The influence of computerized interpretation of an electrocardiogram reading. Am J Emerg Med 2016;34(10):2031-2
  3. Snyder CS, Fenrich AL, Friedman RA, et al. The emergency department vs the computer: which is the better electrocardiographer? Pediatr Cardiol 2003;24:364-368
  4. Anh D, Krishnan S, Bogun F. Accuracy of electrocardiogram interpretation by cardiologists in the setting of incorrect computer analysis. J of Electrocardiol 2006;39:343-345
  5. Schlapfer J and Wellens HJ. Computer-interpreted electrocardiograms: benefits and limitations. JACC 2017 Aug 29;70(9):1183-1192
  6. Shah AP, Rubin SA. Errors in the computerized electrocardiogram interpretation of cardiac rhythm. J of Electrocardiol 2007 (40): 385-390
  7. Garg A, Lehmannn M, Prolonged QT interval diagnosis suppression by a widely used computerized ECG analysis system. Circ Arrhthm Electrophysiol
  8. Bosson N, Sanko S, Stickney R, et al. Causes of prehospital misinterpretations of ST elevation myocardial infarction. Prehosp Emerg Care 2017;21:283-290
  9. McLaren J, Kapoor M, Yi SL, Chartier LB. Using ECG-to-Activation time to assess emergency physicians’ diagnostic time for acute coronary occlusion. J of Emerg Med 2021;60(1):25-34
  10. Bracey A, Meyers HP, Smith SW. Emergency physicians should interpret every triage ECG, including those with a computer interpretation of ‘normal’. Am J Emerg Med 2022 Mar 17;S0735-6757(22)00180-2