Conventional computer ECG algorithm misses another potentially deadly abnormality. How does AI perform?

Written by Pendell Meyers

A woman in her 70s with with multiple medical comorbidities began to have episodes of weakness and pre-syncope off and on for several days, then suffered three episodes of syncope within a few hours. She was brought to an Emergency Department for evaluation. On arrival she was awake, alert, and asymptomatic, with normal vital signs except for heart rate in the low 50s.

Here is her ECG:

What do you think?

Here are the computer measurements:

Which one is a potentially life-threatening error?

The QT interval is grossly underestimated. This ECG shows sinus bradycardia with massively long QT (or QU?) interval, at over 600 msec. Especially when present in the setting of bradycardia and syncopal episodes, this is very worrisome for high risk of lethal dysrhythmias including polymorphic ventricular fibrillation (termed Torsades when in the setting of long QT).

Unfortunately, the EM physician somehow did not notice this extreme QT prolongation, and simply agreed with the computer interpretation that there were no significant abnormalities on the ECG. 

The patient was sent home!

She then had several more syncopal episodes the next day, and returned to a different Emergency Department.

Here is her ECG at that time:

Here are the computer measurements:

QT 382 msec!?!?

Incredible how bad the traditional computer algorithms can be at this important measurement. 

It is again over 600 msec. 

If you had used the PM Cardio app, see what you would have found.  Below we applied it to the first ECG at the top

The PMCardio app is more than the Queen of Hearts for diagnosis of OMI.  It also diagnoses 38 different ECG abnormalities such as hypertrophies, bundle branch blocks, AV blocks, dysrhythmias, etc.  And it was trained to do so using artificial intelligence (AI), a deep convolutional neural network.

The conventional algorithm had measured the QT at 374ms, correcting for rate to 356ms.  The PMcardio app states 513ms and 493, respectively, and gives a diagnosis of suspected long QT syndrome.  While a long QT does not translate directly to long QT syndrome, the message is clear.

This time, the physician did not trust the computer and noticed the extremely long QT. 

All electrolytes were within normal limits (including magnesium and potassium 4.2 mEq/L). Medications were reviewed, and the patient was found to be on amiodarone (for atrial fibrillation, but had been in sinus rhythm for a few months since a recent atrial fibrillation ablation). 

No other QT prolonging medications or conditions were found, and it was attributed to the amiodarone, which was discontinued.

Here is her ECG a few days later:

QT normalizing.

She had no more episodes of syncope, and no episodes of dysrhythmias on cardiac telemetry during hospitalization. She was discharged home.

===================================

MY Comment, by KEN GRAUER, MD (10/24/2024):

===================================

Important post by Dr. Meyers — that highlights an ECG finding that clearly should not be missed. That said — I suggest an additional perspective below.

• For clarity in Figure-1 — I've reproduced and labeled today's initial ECG.

Figure-1: I've labeled the initial ECG in today's case.

Regarding the “Miss” by the Computer ...

• Although the standard ECG computer algorithm completely missed the diagnosis of the markedly prolonged QTc (really the “QU” ) interval in today’s case — this is not “the fault” of the computer. Instead — the fault lies squarely with the clinician who unknowingly “trusted” the computer measurement.
• Standardized computer ECG interpretation programs have their pros and cons. (For "My Take" regarding optimal use of computerized ECG Reports — Check out my ECG Podcast #3 in the July 24, 2024 post in Dr. Smith's ECG Blog). Computerized ECG programs usually are accurate for assessing ECG intervals IF the onset and offset of the interval is readily readily definable. 
• I found today’s initial ECG intriguing as one of the most difficult ECGs I have yet encountered for defining the limits of the T waves and U waves on this tracing. Since I (as an expert human interpreter) have trouble defining the end point of T waves and U waves on the tracing in Figure-1 — we should not expect the computer to be any better at defining these T wave and U wave end points. In a word — You can not use the computer “measurement” of the QT interval in today’s ECG — and clinicians need to appreciate this important point.

The general “rule” for assessing ECG interval prolongation (including prolongation the PR interval; QRS duration; and the QTc) — is to survey all 12 leads — and then to select that lead (or leads) in which you can clearly define the onset and offset of the interval you are looking at in which the interval is longest.

• NOTE: Unless you regularly include assessment of intervals (PR-QRS-QTc) in your systematic ECG interpretation — potentially life-threatening errors as seen in today's case will occur. My way of ensuring that I do not overlook any abnormal intervals — is to automatically insert "Intervals" immediately after my assessment of Rate and Rhythm for every 12-lead ECG I encounter.

In today's case — it should be immediately apparent that measurement of the QTc is no simple matter.

• Although the QTc might not seem overly prolonged on survey of the long lead V1 rhythm strip in Figure-1 — simultaneous survey of the neighboring long lead II and lead V5 rhythm strips suggest otherwise. In fact — there appear to be 3 distinct "parts" to the baseline in the long lead II (that I've labeled "a" - "b" - "c" ). This is not artifact — and it should alert the clinician to look closer. 
• Review of the other 9 leads on this tracing should make it apparent that defining the limits of the QTc is not obvious. We will not be able to trust the computer QTc measurement in Figure-1.
• Looking closer — I thought a distinct "dip" is seen defining the the end of the T wave of beat #8 in lead V5 (RED arrow in this lead).
• This suggests that the end of the U wave is marked by the BLUE arrow in lead V4. Vertical red and blue lines show where these end points appear in other simultaneously recorded leads. As per Dr. Meyers — the QTc (Q-U) is over 600 msec, and markedly prolonged!

Learning Points:

Today's patient was an older woman with a complex medical history (including recent ablation of AFib) — who was on multiple medications including Amiodarone. Her symptoms included weakness and a number of syncopal episodes.

• The above history should more than merit the need for hospital admission until clarification of the cause of her syncopal episodes — Yet this patient was initially sent home.
• Despite strong suggestion on ECG #1 of hypokalemia and/or hypomagnesemia — these electrolyte levels were apparently normal. Another cause of the patient's syncopal episodes must be sought.
• Given the above history — assessment of the QTc interval is a critical part of the ED evaluation. But given the lack of clear limits for T waves and U waves on today's ECGs — computer measurements for the QTc can not be trusted.
• Manual assessment of the QTc (Q-U) interval clearly shows marked prolongation that needs to be recognized.