Are we Missing Acute MIs with Clinical Risk Scores?

Background: In 2011, we saw 7 million patients in the emergency department (ED) complaining of chest pain. Most of these patients did NOT have an acute coronary syndrome (ACS) or an acute myocardial infarction (AMI). Missing an AMI is one of the biggest fears we have in the ED. By using validated risk scores, we can help decrease the risk of missing AMI and the resultant adverse events. There are multiple scores available for our use. Thrombolysis in Myocardial Infarction (TIMI) predicts risk of adverse outcomes in the next 14 days. Global Registry of Acute Coronary Events (GRACE) predicts outcomes at 6 months. ED specific scores include HEART and Emergency Department Assessment of Chest Pain (EDACS). But, how well do these scores actually perform? Are we missing AMIs by using these clinical risk scores?

What They Did:

  • This study was part of a larger, multicenter, prospective observational study. The original study was looking to validate a novel cardiac troponin for ED use in patients suggestive of ACS.
    • 7 geographically different academic US EDs
    • Inclusion: patients age 21 or older with symptoms suggestive of ACS
    • Exclusion: implanted defibrillator discharge < 24 hours, cardiac procedure or bypass < 30 days, recent chest trauma
    • Recorded: clinical gestalt, TIMI, GRACE, HEART-1, HEART-2, and EDACS to separate into low, intermediate, and high risk. EDACS only uses low and not low risk classifications. (Difference between HEART-1 and HEART-2 was using one or two troponins.)
  • Cutoffs used: HEART 0-3, TIMI = 0, GRACE < 50, EDACS < 16


  • Primary—presence or absence of AMI. Final diagnosis was determined by a committee of 3 (2 cardiologists, 1 ED), blinded to each other’s determination. The ESC/ACCF/AHA/WHR 3rd Universal definition of MI was used.


  • 434 patients included in the final analysis
    • Median Age: 57 years (Range: 49 – 64)
    • Diagnosis of AMI: 18.4%
    • Median Time Between Serial cTn: 4.1 hr (Range 2.7 – 7.8hr)


  • Adequate enrollment: Study needed at least 45 patients with AMI and assumed a 10% prevalence for AMI. They would need 450 patients enrolled in the study (They enrolled 459 of which 434 were included in the final analysis).
  • Use of all major risk assessment scores at recommended cutoffs


  • Only 82 patients had confirmed AMI: This is a small sample size
  • Possible selection bias, excluded many patients due to incomplete data: This could lead to under- or over-estimation of clinical risk scores
  • Only used AMI, likely missed other ACS and longer-term major adverse cardiac events (MACE)
  • Each hospital troponin was different, with variable cutoffs. This may have caused variability in the results of the prediction scores this study
  • Half of the false negative cases involved point of care assays. The high rate of missed AMI in patients classified as low risk may be due to using older, less sensitive cTn assays, which have high coefficients of variation
  • Some facilities used POC cTnI as their initial test and laboratory cTnI as their follow up test, which makes it difficult to evaluate rise and fall as two different assays were used


  • Of all the symptoms described diaphoresis was much more common in patients diagnosed with an AMI vs patients without AMI (43.8% vs 5.4%). Take diaphoresis seriously.
  • All assessments were within statistical variability, but high miss rates were noted in all except TIMI, and gestalt + troponin.
  • Reducing score cutoffs would result in zero misses, but much fewer patients would be classified as low risk. The lower cut-offs were defined as:
    • HEART-1 = 1
    • GRACE </= 48
    • TIMI= 0
    • HEART-2 </= 2
    • EDACS </= 11
  • The percentage of patients that would be considered low risk with these lower cut offs come at the price of less patients identified as low risk:
    • HEART-1: 1%
    • GRACE: 3.3%
    • TIMI: 7%
    • HEART-2: 19.5%
    • EDACS: 33.6%
  • It is important to note that the original HEART score and EDACS score have had prospective evaluation. Most of the HEART score studies are performed in European populations and the EDACS score was derived and validated in New Zealand but used hs cTnI. The findings of this study may simply be due to different populations (European vs US) or the use of hs cTnI (which until recently was not available in the US). Even so, due to the smaller population of this study, 95% CIs were very wide.
  • Just a little discussion point about risk scores. Risk scores are tools to help clinicians have a standardized approach to patient complaints. There will always be exceptions to the tools. Not to pick on anyone one score, but lets talk about the HEART-2 score as an example. It is possible that someone could have a low risk score (0 – 3) and not truly be low risk.
    • Lets start with risk factors. If a patient doesn’t go to see a doctor, then they won’t have any past medical history, when in fact they may. This will grossly underestimate their risk.
    • Although only one ECG is required for the HEART-2 score, there is something to be said for multiple ECGs. Be sure to look at pre-hospital ECGs, old ECGs in your EMR, and for god’s sake, get serial ECGs in patients you are concerned about. Remember, a single ECG is one point in time. We don’t want to miss dynamic changes and the best way to find these subtle changes may be looking at pre-hospital ECGs, old ECGs, and serial ECGs.
    • What about troponins? It is very possible that you could have a patient that doesn’t go to doctors, and therefore has no risk factors, a normal ECG and have a positive troponin. By definition, this would still be low risk if you use the HEART-2 score at face value. I think most of us would agree that positive troponins are not low risk. Even if it is not ACS that is causing the elevation, there maybe another pathology at play. This is not a low risk patient, and should not be sent home without further risk stratification or workup.

Author Conclusion: “Using their recommended cutpoints and non-high sensitivity cTn, TIMI and unstructured clinical impression were the only scores with no missed cases of AMI. Using lower cutpoints (GRACE ≤48, TIMI = 0, EDACS ≤11, HEART ≤2) missed no case of AMI, but classified less patients as low-risk.”

Clinical Take Home Point: Clinical gestalt remains the most useful tool for assessment of risk of ACS/AMI. The use of risk assessment tools should be regarded as just that, tools. Keep in mind the limitations of each as you are taking care of patients.


  1. Singer AJ et al. Missed Myocardial Infarctions in ED Patients Prospectively Categorized as Low Risk by Established Risk Scores. Am J Emerg Med 2017. [Epub Ahead of Print] PMID: 28108220

For More on This Topic Checkout:

Post Peer Reviewed By: Salim Rezaie (Twitter: @srrezaie)

Cite this article as: Matt Astin, "Are we Missing Acute MIs with Clinical Risk Scores?", REBEL EM blog, April 13, 2017. Available at:
The following two tabs change content below.

Matt Astin

Clinical Assistant Professor of Emergency Medicine/Internal Medicine at Medical Center of Central Georgia
REBEL EM Associate Editor and Author

Like this article?

Share on facebook
Share on Facebook
Share on twitter
Share on Twitter
Share on linkedin
Share on Linkdin
Share on email
Share via Email

Want to support rebelem?

9 thoughts on “Are we Missing Acute MIs with Clinical Risk Scores?”

  1. Hi Matt,

    Thanks for posting this review. It shouldn’t be a big surprise that gestalt + cTn outperformed ACS risk scores (except TIMI) with regard to not missing an AMI. The fear of missing an AMI has us tending towards being conservative in our assessment leading to biased decision making. Our false negative rates will fall, but our false positives rise. And this is what gave rise to the development of ACS risk scores in the first place; to try and reduce the quantity of unnecessary coronary catheterisations and associated complications, while minimising missed AMIs. This is similar to the development of NEXUS and the Canadian C-spine rules to reduce the number of unnecessary c-spine CTs. There is a trade-off.

    The TIMI risk score for unstable angina/non-ST elevation MI: A method for prognostication and therapeutic decision making. JAMA 2000. Antman et al.

    So, leaving aside technical criticisms such as small sample size and inclusion bias for a moment, while it might be tempting to superficially invoke evidence based medicine and state that ACS risk scores miss AMIs and therefore should be abandoned in favour of Gestalt + cTn or the TIMI score, it would be more accurate to adopt an evidence influenced stance and screen patients with a standardised, validated tool while recognising the limits of that tool and when to make an exception. Which is why well trained and experienced human clinicians still have the edge over medical robots.

    • Thanks for the comment, Matthew. I totally agree, part of the reason for writing the post. The medical robots haven’t won yet. Until then, nothing can beat the well trained and experience clinician. Using validated tools in our practice should be viewed as just that, a tool to help verify or discard hypotheses.

  2. Hi,

    British EM Doctor here.

    I’m a little confused at the application of the various risk scores here. I thought (for example) GRACE was to be used _after_ you had the diagnosis of MI to predict mortality. Is it really a valid use then to look at it to rule out MI?


    • Hi Andrew,
      Appreciate you reading and commenting. You are hitting the nail on the head. Many scores for ACS prediction were and still are being used in a manner that they were not intended. TIMI for example is the risk of all cause mortality, MI, and severe recurrent ischemia requiring urgent revascularization within 14 days AFTER admission. GRACE is risk of hospital death and post-discharge death at 6 months. Other issues I have with GRACE is that it is very complex to use and a large portion of the score is dependent on patient’s age. Neither of the two scores I just mentioned were evaluated in all comers of patients with chest pain, but instead only patients already known to have ACS. Therefore the scores were not derived for an unstratified chest pain population presenting to the ED. So now to your question of why is GRACE even being studied…in the US, chest pain is one of the highest payouts for missed ACS. Many physicians have been trying to find anything to minimize risk, including studying scores that were not derived for that intention. Due to this many of these scores are included in studies for comparison to newer scores. We have certainly come a long way in chest pain evaluation and although we are not at a 100% no miss rate, I would argue that if we ever hit that, then we are overcasting our patients. Hope this helps.


  3. Just a comment on the prospective (and retrospective, for that matter) evaluation of the HEART score. When you dig into the details of these studies, it is worth noting that the two most subjective elements of HEART – history and ECG interpretation – are NOT determined by the treating clinician.

    In my view, this is a major limitation of these diagnostic studies. One must question the face validity and generalizability of the conclusions if you are an emergency physician in my setting as it is the treating emergency physician who takes the history and interprets the ECG.

    What are your thoughts on this?

    • While it is true that the original papers on the HEART score were chart reviews and the data were NOT obtained by the treating clinician, this is how clinical decision tools are created. Other studies have truly prospectively calculated the score. The overall point should be that no matter who decides what the score is, we should not be relying only on that score to make decisions. The HEART score is a tool to help with clinical decisions, not make the decision for us.

  4. Hi Matt,

    Could you direct to the studies that have truly calculated HEART and applied it as I might in the ED? The ones I have come across have been “prospective” with the caveat that the history is obtained on a data entry form – and presumably determined to be of high, moderate or low suspicion by a researcher at some later time – and the ECGs interpreted by Cardiologists.

    Also, how one defines the history as highly, moderately or slightly suspicious is quite variable from study to study. I think a lot of clinicians use overall gestalt to score the history, but this isn’t how HEART has been studied. Do you think this is problematic?

    I just think it is worth emphasizing that the two most subjective elements of the score seem to not have been applied in studies the way we tend to in real life.

    Having said this, I do like the score. It provides a nice framework. I’m just not sure that it performs any better than clinician gestalt, which naturally encompasses risk factors, history, (serial) ECG and (serial) troponin.


    • Hey Chris,
      Agree with Matt on this. Most Clinical Decision Tools are derived from chart review type studies. Several of the HEART Score studies did actually use clinicians to decide…I will attach below:

      1. Mahler SA et al. The HEART Pathway Randomized Trial: Identifying Emergency Department Patients With Acute Chest Pain for Early Discharge. Circ Cardiovasc Qual Outcomes 2015; 8(2): 195 – 203. PMID: 25737484
      2. Poldervaart JM et al. Effect of Using the HEART Score in Patients With Chest Pain in the Emergency Department: A Stepped-Wedge, Cluster Randomized Trial. Ann Intern Med 2017; [Epub Ahead of Print] PMID: 28437795
      3. Poldervaart JM et al. Comparison of the GRACE, HEART, and TIMI Score to Predict Major Adverse Cardiac Events in Chest Pain Patients at the Emergency Department. Int J Cardiol 2017; 227: 656 – 661. PMID: 27810290
      4. Backus BE et al. Prognostic Factors in Chest Pain Patients: A Quantitative Analysis of the HEART Score. Crit Pathw Cardiol 2016; 15(2): 50 – 5. PMID: 27183254

      These were just a few I was able to find….I think there may be a handful more…hope this helps.



Leave a Comment

Time limit is exhausted. Please reload CAPTCHA.