Professor Terry McCormack, GP and professor of primary care cardiovascular medicine, considers how to handle an unforeseen computerised ECG interpretation
The case
A rather anxious 23-year-old man attends with palpitations. A careful history suggests simple ectopic beats, perhaps exacerbated by his anxiety. There are no red flags in the history, he is a non-smoker and has no family history of cardiac disease. There are no other cardiological features in the story and he exercises regularly without problem. Cardiovascular examination is normal.
The computerised ECG interpretation confirms occasional atrial ectopics, but to your surprise, also states ‘Infarct, age undetermined’.
How accurate are computerised interpretations of ECGs?
ECG interpretation software was designed to avoid false negatives. Therefore, the most reliable finding is ‘Normal ECG’.
If you ask, ‘How accurate are human interpretations?’, the answer in a relatively old study from 1985 would be that 82% of computer and 64% of physician interpretations were accurate.1
The more recent SAFE study, conducted by the University of Birmingham and published in 2007, looked specifically at the diagnosis of atrial fibrillation and compared GP, practice nurse and computer interpretation. The computer outperformed the GPs and practice nurses, but the best results came from combining human and computer skills. This resulted in a sensitivity of 92% and specificity of 91% in detecting atrial fibrillation.2
Ultimately, every clinician should build up their own skills in ECG reading, rather than relying on the computer interpretation. We need to ask the same questions every time:
- What is the rate, rhythm and axis?
- Is there a P-wave?
- What is the PR, QRS and QT interval?
- What does the ST segment look like and is there inappropriate T wave inversion?
Consider the ECG interpretation as a second opinion, not the definitive diagnosis.
The development of artificial intelligence may improve computing to the point that it will be 98% reliable, which would be better than many cardiologists.
How should we manage this scenario in the short term?
It is extremely unlikely, but not impossible, for a 23-year-old man to have a myocardial infarction (MI).
Contextual information is crucial when considering an unexpected computerised interpretation such as this. If the patient had been older, had other risk factors or had a history of symptoms suggesting MI, this would be far more likely to represent a true reading.
This patient is not describing chest pain. The suggestion from the ECG is an old, silent infarct – extremely rare in this age group.
My first action would be to read the ECG myself, ignoring the computer interpretation. I would look for pathological Q-waves, particularly in leads III and aVF, and also evidence of ST elevation. As with VAR checks in football and rugby, there would have to be a ‘clear and obvious’ issue, because my level of suspicion would be extremely low. Assuming this is all negative, my next step would be to investigate possible causes of his palpitations.
To give me time to discuss this with colleagues, I would suggest some blood tests, such as full blood count and thyroid stimulating hormone.
What should be our longer-term management?
Discussing a puzzling case with colleagues can be invaluable. I would propose using a regular clinical meeting to get the input and advice from other GPs in the practice. Options for management could include sending the ECG to the local cardiology department for advice and guidance.
One tricky question is whether or not to tell the patient the computer’s interpretation. Doing so might only worsen the patient’s existing anxiety. However, in this age of open access to records you do leave yourself open to criticism if you do not show complete transparency.
My preference would be to get the backup of a specialist interpretation and then explain to the patient that the ECG interpretive software is not without fault.
I sought the views of Glasgow cardiologist Professor Adrian Brady, who said: ‘The likelihood is that the ECG showed Q waves in III and aVF, caused by a minor irregularity in the cardiac axis. The “old school” method of overcoming this is to repeat the ECG on deep inspiration. The heart descends and rotates within the chest and the cardiac axis changes slightly. This is often enough to restore the ECG to “Normal”. I would always confirm this with an echocardiogram, which in all likelihood would be normal.’
Although I have open access for echocardiograms in suspected heart failure, I would need the backup of the cardiology advice to order one in these circumstances.
Assuming the blood tests are normal, I now need to consider the patient’s anxiety. He needs to avoid excessive caffeine intake and steer clear of cocaine. Anxiety-coping mechanisms are available online, as well as via our local IAPT service. We also have a primary care mental health worker.
In terms of his palpitations/ectopics, there are technological advances that can monitor heart health. One such advance is the Apple Watch, which has been shown to successfully detect atrial fibrillation.3 However, this is clearly a risky suggestion for an anxious young man.
References
- Jakobsson A et al. Does a computer‐based ECG‐recorder interpret electrocardiograms more efficiently than physicians? Clin Physiol. 1985 Oct;5(5):417-23. Link
- Mant J et al. Accuracy of diagnosing atrial fibrillation on electrocardiogram by primary care practitioners and interpretative diagnostic software: analysis of data from screening for atrial fibrillation in the elderly (SAFE) trial. BMJ. 2007 Aug 23;335(7616):380. Link
- Perez M et al. Large-scale assessment of a smartwatch to identify atrial fibrillation. NEJM. 2019 Nov 14;381(20):1909-17. Link
Dr Terry McCormack is a GP in Whitby and Honorary professor of primary care cardiovascular medicine at Hull York Medical School
I am surprised that the first response to the unexpected interpretation was not to check for the reason the computer reported this : we should all be familiar with the acute and later ECG signs of MI, and should check whether any of these are present on this trace. We must identify the signs the computer has used to make the diagnosis, and confirm if they are, or are not, actually present.
The commonest incorrect diagnoses from computer and doctors are AF and SR, mostly due to incorrect interpretation of movements and mains artefacts!
I remember one patient who had a continuous paper 3-lead strip during a sudden collapse and resuscitation attempt on the ward. The computer reported just about every diagnosis in it’s dictionery, and ultimately the last one (asystole) was actually correct, but none of the others were. I looked at the strip from the beginning and immediately spotted a typical pattern for a massive pulmonary embolism, although this was the ONE diagnosis that the computer had not reported. I also spotted several other effects, some of which were reported, mainly due to various aspects of the resuscitation attempt, rather than the diagnosis reported.
a fit, non smoking, pregnant nurse came to see me with “a pulse of over 140”. they – i.e. all her classmates – had been doing ecg’s on each other for their training. the cardiology registrar had reviewed them all – and hers was normal. however, it showed very obvious WPW. she had ablation and has been fine ever since.
just saying.
even monkeys fall off trees !!
Our surgery machines frequently report a long QT interval. I’ve got used to rechecking it manually now, and the machine is almost never correct.
We have to do a lot of ECG;s on mental health patients on anti-psychotics to check QTc and often the computer report shows a prolonged QTc based on the older Bazett formula. The newer formula’s like Framingham are much more accurate and you can often change the setting on your ECG program. If not use MD calculator which has all options and you will be shocked by the difference in the corrected QT interval measurements: https://www.mdcalc.com/calc/48/corrected-qt-interval-qtc