Tuesday, May 23, 2023

A man in his early 40s with chest pain a "normal ECG" by computer algorithm. Should we avoid interrupting a physician to interpret his ECG?

Written by Pendell Meyers

A man in his early 40s experienced acute onset chest pain. The chest pain started about 24 hours ago, but there was no detailed information available about whether his pain had come and gone, or what prompted him to be evaluated 24 hours after onset. 

EMS arrived and recorded this ECG:

What do you think?
See same ECG below with computer automated interpretation, using the Glasgow ECG algorithm which apparently is used by many different providers and devices

Amazing that the computer calls this normal. Notice on the right side of the image how the algorithm correctly measures STE sufficient in V1 and V2 to meet STEMI criteria in a man older than age 40.
And yet it still says "normal". 

As most would agree, this ECG shows highly specific findings of anterolateral OMI, even with STEMI criteria in this case. Thus, this is obvious STEMI(+) OMI until proven otherwise.

If the Queen of Hearts AI program had been used instead, this is what the paramedic would have seen:

Many authors advocate that a computer algorithm "normal" or "otherwise normal" reading has such a low rate of important findings that would change management, such that they do not think it is worth interrupting a physician to have the physician overread the ECG. See discussion and links at end of case for more info on this debate.

But the paramedic and the ED physician in this case did not subscribe to this idea. Instead, they reviewed the ECG immediately regardless of the computer interpretation of "normal."

And they of course activated the cath lab immediately, where he was found to have acute thrombotic occlusion (TIMI 0) of the proximal LAD, as well as embolic D1 occlusion. LCX and RCA were described as "normal" in the cath report. LAD and D1 were stented, but flow unfortunately could not be well restored despite efforts (they list the post intervention TIMI flow still as 0). They did mention some faint collaterals from the RCA to the LAD.

After the angiogram, the first high sensitivity troponin I resulted at 7,259 ng/L (99% URL for men is 20 ng/L for this assay). The next four troponins all returned at "greater than 25,000 ng/L", which is simply the lab's upper reporting limit at this institution.

Here is his ECG hours after the interventions:
There is interval improvement, deflation of STE and HATWs, with terminal T wave inversions. Despite the poor flow noted post-PCI on angiogram, the tissue perfusion has improved according to the ECG.

Next day ECG. QS waves in V2 and aVL. Ongoing T wave inversion in V2, I, aVL.

Echo showed hypokinesis of the mid anterior, mid intravascular septum, and apical anterior walls. EF 40%.

Learning Points:

1. You cannot fully trust conventional computer ECG algorithms to find even STEMI(+) OMI, as in this case

2. You can trust them even less to interpret STEMI(-) OMI findings on ECG. 

It is far too premature to say that paramedics and physicians should not be bothered to interpret ECGs labelled as "normal" or "otherwise normal" by the computer algorithm. How many unnecessary interruptions would you accept to find one case like this? by being interrupted???  For me, the number is really, really high. Far higher than what has been studied so far. Perhaps in the future with better automated ECG interpretation, as also seen in this case, it could be different.

Here are 38 cases of "Normal ECG by Computer Algorithm," the vast majority show subtler STEMI (-) OMI on the ECG.

Here are some of them:

A middle aged female with "heartburn" and a "normal ECG" per the computer

Computer says "normal," troponin undetectable.


Emergent cardiac outcomes in patients with normal electrocardiograms in the emergency department. Am J Emerg Med. 2022 Jan;51:384-387. doi: 10.1016/j.ajem.2021.11.023. Epub 2021 Nov 17.

We wrote this editorial on the above paper (full text).  Reference: 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 May;55:180-182. doi: 10.1016/j.ajem.2022.03.022. Epub 2022 Mar 17. PMID: 35361516.


This study looked at less than 1000 cases, which is not nearly enough (see below for analysis) and they used cardiologists as the gold standard (a very poor gold standard), NOT presence or absence of Occlusion MI (which we have done in all of our ECG studies, and must be ascertained by 1) TIMI 0/2 flow on angiogram or 2) culprit + TIMI 3 flow and very high troponin. 

So this study is worthless and must be ignored. 

I have here 38 cases of "Computer Normal" ECGs which were critically abnormal and the vast majority are missed acute coronary occlusions (Missed Acute OMI) and most were recognized by the physician.

We wrote this Editorial in the Journal of Electrocardiology in 2019.  Litell JM, Meyers HP, Smith SW. Emergency physicians should be shown all triage ECGs, even those with a computer interpretation of “Normal.” J Electrocardiol [Internet] 2019;54:79–81. Available from: http://dx.doi.org/10.1016/j.jelectrocard.2019.03.003


"To illustrate the limitations imposed by sample size, recent data from our institution reveal that we identify approximately 225 type I myocardial infarctions (MI) in a typical year. These include about 60 occlusion MI (OMI) with clear ST segment elevation (none of which would be called “Normal” by the computer) and about 165 Non-STEMI. Of the Non-STEMI in our cohort, about 25% will actually have acute coronary occlusion.  While most of these roughly 40 NSTEMI occlusions would be read by the computer algorithm as abnormal in some way (typically nonspecific ST segment or T wave abnormality), they would not be labelled STEMI.  We might conservatively estimate that 5 of these 40 acute OMIs without diagnostic ST segment elevation would be erroneously read by the computer as “Normal.” That is five OMIs per year misread by the computer algorithm as normal. In that same year we collect approximately 24,000 ECGs in our ED, of which 20% are called “Normal” by the computer. Taken together, these data suggest that out of 5000 “Normal” ECGs in a given year, about 5 (0.1%) will actually be acute OMIs that have been misinterpreted by a computer algorithm as completely normal. A sample size of 855 has no chance of generating a meaningful conclusion about the reliability of computer “Normal”.   In fairness, it is not certain that an average emergency physician will catch these few false normals, but they will absolutely go missed if the physician never sees them.We would prefer to be interrupted."

My Comment by KEN GRAUER, MD (5/23/2023):
Today's case illustrates a series of important points that we continually emphasize by means of the cases we present on Dr. Smith's ECG Blog:
  • Point #1: The paramedics and ED physician in today's case did the right thing by immediate personal review of the initial ECG (and by not accepting the "normal" reading from the computer)Existing computerized ECG interpretation programs are (and have always been) by definition faulty! This is because until recently — computerized programs have been based on STEMI-criteria — which as we have shown, will miss an estimated 25-30% of acute coronary occlusions (See My Comment in the July 31, 2020 post in Dr. Smith's ECG Blog).

  • Point #2: Today's case illustrates how existing computerized ECG interpretation programs may sometimes miss even STEMI infarctions (!) — as happened in today's case, in which the initial ECG was labeled "normal" by the computer. I have on multiple occasions in this ECG Blog presented my perspective — in that I find existing computerized ECG interpretation programs useful because I know HOW to use them (See My Comment at the bottom of the page in the February 4, 2022 post of Dr. Smith's ECG Blog). But regardless of your opinion of current computerized interpretations — the KEY point is that you should never accept the computer reading as correct until YOU have verified its interpretation with your own eyes! And, if you disagree with what the computer says (because YOU think the ECG suggests an acute process) — YOUR opinion is the one that needs to be followed!

  • Point #3: Help is on the way! As we have repeatedly shown (evidence today's post — among many others) — the new "Queen of Hearts" (QoH) AI app has demonstrated amazing sensitivity and specificity for recognizing acute coronary occlusions (including both those with or without satisfying STEMI criteria) — so there very soon should be increasingly more widespread use of this superb resource to assist in recognition of acute OMIs (See My Comment at the bottom of the page in the March 31, 2023 post of Dr. Smith's ECG Blog)
  • To Emphasize: As good as the QoH app is for distinguishing between acute OMIs vs non-OMIs — My recommendation will remain for all clinicians to personally review each ECG. That said — clinician accuracy will certainly improve for early recognition of acute OMIs as use of QoH interpretation increases.

  • Point #4: Once again — the History is a KEY component for optimal ECG interpretation. Specifically, the history in today's case was, "chest pain that started 24 hours earlier — without information about whether the pain came and went, or persisted — nor about what prompted evaluation 24 hours after onset". Awareness of the extended duration of symptoms in today's case (suggesting that symptoms probably were not consistently intense — or we would have expected the patient to present sooner than 24 hours) — prepares us for: i) The established Q wave in lead V2; andii) The overall modest amount of ST elevation despite total occlusion on cath of the proximal LAD and 1st Diagonal (See below for further discussion of these findings).

  • Point #5: The initial ECG in today's case is worthy of a detailed look as to why Dr. Meyers said it shows "highly specific findings of anterolateral OMI". I review these specific ECG findings in Figure-1 — in which I've reproduced and labeled this initial ECG.

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

The Initial ECG in Today's Case:
The rhythm in ECG #1 is sinus at ~65/minute (The rhythm is actually completely regular — with the first 3 beats being repeated in each simultaneously-recorded set of 3 leads). Intervals (PR, QRS, QTc) and the frontal plane axis are normal. There is no chamber enlargement.
  • The most concerning lead is lead V2. There is a Q wave in lead V2 that should not be there (RED arrow) — especially since there has been loss of R wave from the small-but-definitely-present initial positive deflection of the QRS complex in lead V1 (ie, BLUE arrow in this lead highlighting the initial r wave in V1)
  • As per computer calculation — there are 2 mm of J-point ST elevation in lead V2, with initial straightening of the ST segment takeoff (GREEN line in this lead). Although some gently upsloping ST elevation is commonly seen as a normal finding in lead V2 — the amount of ST elevation seen in ECG #1 seems excessive, especially in association with the abnormal Q wave, ST segment straightening, and overly wide base of the T wave in this lead. In this patient with 24 hours of chest pain — I interpreted the ST-T wave in lead V2 as hyperacute.

  • In support that there is hyperacute ST elevation in lead V2 — is the more than 1 mm of ST elevation with surprisingly tall positive T wave in lead V1. In the absence of LVH or bundle branch block — this is rarely a normal finding (especially in a patient with new chest pain and other suspicious ECG findings).
  • In addition — there is T wave "imbalance", in that the T wave in lead V1 is taller than the T wave in lead V6 (ie, Comparison of relative T wave amplitude within the dotted RED vs BLUE rectangles in Figure-1).

PEARL: When T waves in each of the chest leads are upright (as they are in ECG #1) — the T wave in lead V1 is usually not taller than the T wave in lead V6This "imbalance of precordial T waves" is not seen very often — and in the “right” clinical setting, has been associated with acute OMI (See Manno et al: JACC 1:1213, 1983 — and the July 17, 2013 post by Salim Rezaie in ALiEM).

  • NOTE: This is not to say that tall, upright T waves in lead V1 might not sometimes be the result of a repolarization variant or a mirror-image reflection of LV “stain” that can sometimes be seen in anterior leads. Instead — it is simply to say that on occasion — I have found recognition of a tall, upright T wave in lead V1 that is clearly taller than the T wave in lead V6 to be a tip-off to an acute coronary syndrome that I might not otherwise have recognized (For more examples of this finding — See My Comments at the bottom of the page in the October 23, 2020 post — in the June 1, 2022 post — and in the March 26, 2022 post of Dr. Smith's Blog).

Additional abnormal ST-T wave findings in ECG #1:
  • Neighboring lead V3 is also abnormal. To Emphasize: By itself — I would not necessarily interpret lead V3 as abnormal. But in the context of T wave "imbalance" with strong suggestion of hyperacute ST-T wave elevation in leads V1 and V2 — I interpreted the ST segment straightening (GREEN line) in lead V3 as extension of the acute process.
  • P.S.: The fact that a healthy R wave returns in lead V3 — supports our suspicion that loss of r wave (from lead V1-to-V2), and the resultant initial Q wave in lead V2 is "real" and a marker of infarction.

  • There is ST elevation in lead aVL. As I discussed in My Comment in the November 21, 2020 post in Dr. Smith's ECG Blog — Birnbaum et al (Am Heart J 131:38, 1996) emphasized the diagnostic utility of identifying ST elevation in this lead — with this finding pointing to proximal LAD and/or 1st Diagonal involvement rather than LCx occlusion when there is ST elevation in one or more of the anterior leads. 

  • Leads III and aVF manifest the mirror-image opposite ST-T wave picture of lead aVL (Comparison of the ST-T waves within the solid and dotted GREEN rectangles in these leads). These are reciprocal changes.
  • In the context of mirror-image opposite reciprocal ST depression in leads III and aVF — I interpreted the ST segment flattening in lead II (GREEN line in this lead) as a reciprocal change in this 3rd inferior lead.
  • Finally — lead I (the other high-lateral lead) manifests subtle ST elevation with ST segment straightening (GREEN line) — that in the context of other limb lead findings, has supportive implications of the ST-T wave changes in lead aVL.

Putting It All Together:
The initial ECG in Figure-1 is consistent with the history and cardiac catheterization findings of acute proximal LAD (as well as 1st Diagonal) occlusion.
  • The proximal LAD location of acute occlusion is suggested by ST elevation beginning as early as in lead V1, with maximal ST elevation in lead V2. This is further supported by ST elevation in lead aVL with reciprocal ST depression in the inferior leads (Inferior lead reciprocal ST depression is much less common when LAD occlusion occurs more distally).
  • That said — the overall amount of ST elevation is relatively modest in this initial ECG, which could reflect spontaneous reopening of the "culprit" vessel over the 24 hours when the patient was deciding whether or not to come to the ED.
  • Loss of r wave between leads V1-to-V2, with development of an abnormal Q wave in lead V2 on this initial ECG suggests infarction occurred at some point over those previous 24 hours when the patient was still at home. 
  • Serial tracings following stent placement confirmed the large extent of myocardial injury. 

No comments:

Post a 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.

Recommended Resources