My department has just started using the HEART score to help risk stratify chest pain patients. Now chest pain/ACS rule out is a large portion of the ED’s workload, with that workload spilling out over to the acute medics, as we wait for 12 hour troponins. Now your own department will have it’s own ways of sorting through all this chaff, and of course you should follow your own local protocols.
HEART Score allows us to risk stratify low risk chest pain patients and get them out of our department with a single troponin measurement taken at triage. Here it is in all it’s glory.
Now the original derivation study is available here.
The first thing that strikes me is the size of the initial study. The ED it was performed in was a 265 bed community hospital. The cohort size was small as well (only 122 patients).
The authors basically sat down and came up with a scoring system based on the components of the chest pain assessment that we normally do. They didn’t derive their score by complex regression analysis they literally thought it up.
Most of it is pretty objective, and therefore repeatable and reliable. Apart from one bit. That’s the history.
Heres what the authors wrote about the history.
“For the purpose of this study, patient history was classified by two investigators, based on the narrative in the hospital charts written in the emergency room and not allowing for risk factors, ECGs, laboratory results and later developments. In the absence of specific elements in terms of pattern of the chest pain, onset and duration, relation with exercise, stress or cold, localisation, concomitant symptoms and the reaction to sublingual nitrates, the history was classified as ‘nonspecific’ and granted zero points. If the patient history contained both nonspecific and suspicious elements, the history was classified as ‘moderately suspicious’ and granted one point. If the history contained primarily specific elements, the history was classified highly suspicious and granted two points “
So – if the history had ‘bits’ that made the investigator think it was cardiac then they got 2 points, if they had bits that were ‘a bit cardiac’ and ‘a bit not cardiac’ they got 1 point. If there was no parts of the history that was suggestive you got 0 points.
Now the problem I have here is that what one of us thinks is ‘definitely cardiac’ someone else might think was a bit soft. They’ve not broken this down much.
They’ve talked about pattern, onset, duration, relation with exercise, stress, cold, localisation, concomitant symptoms, and the reaction to sublingual nitrates. They’ve not given more detail as to exactly what feature of the the pain, onset, duration etc.
They’ve also not checked to see if there is any agreement or disagreement between people using the score (kappa). So we don’t know what the inter-rater reliability is.
The people who decided how suspect a history were experienced ED clinicians. So surely they are going to have a slightly different opinion to an F2 whose just started their first rotation. These two senior ED clinicians worked in the same department, that doesn’t see STEMIs. So is there experience going to be skewed? We have no way of knowing. We also have no way of knowing if they’d agree with my (or your) assessment of chest pain as ‘highly’ or ‘moderately’ suspicious.
AND if there was nothing in the history that was suspicious then why the hell are you doing a cardiac workup?
Lets look at their results.
So the authors recruited 122 patients over a 3 month period, and followed them up for a minimum of about 300 days. They looked for ‘end points’ which they defined as AMI, PCI, CABG, Death or all together. They managed 98.3% follow up rate.
24.1% or 29 patients got to one of the end points. This was how their scores broke down.
|HEART SCORE||Number||Number of reached any Endpoint|
|0-3 (low risk)||39||1||2.5%|
|4-6 (med risk)||59||12||20.3%|
|7-10 (high risk)||22||16||72.7%|
The authors conclude that if their TINY study is correct it might be possible to discharge people with a HEART score of 3 or less without further investigation as the risk of MACE is <2.5%.
From ED and AMU perspectives that would be wonderful. Chest pain rule outs take up a massive proportion of ED and Acute Medical Unit time, so by allowing us to discharge people prior to that 12 hour trop we might save ourselves a lot of work, and our patients a lot of time, and unnecessary tests.
This is a tiny study! It was retrospective. It is possible to have a Trop >3x normal and non-specific ECG changes and be classified as low risk. I do find that worrying, and I’d feel uncomfortable sending those people home, as I’m sure you would.
I’m going to look at the first validation study, then I’m going to run through a few others. First validation study is available here.
This was a multicentre prospective validation study.
Patients were enrolled aged 21+ if they were having a cardiac workup in one of 10 hospitals in the Netherlands. They had a proforma filled in, which was entered into a database along with their initial ECG and initial troponin result. The database calculated GRACE, TIMI and HEART score. ECGs were blinded and read by two cardiologists using the Minnesota criteria. Ambiguous ECG’s were arbitrated by a third cardiologist. The primary endpoint was MACE at 6 weeks (ACS, PCI, CABG, or Death), which was collected by trawling patient records, phoning the GP, or contacting the patient.
They recruited 2440 patients, lost 7 due to incomplete form filling in, and lost 45 to follow up. This left 2338. 408 patients had MACE at 6 weeks. 155 AMIs, 251 PCIs, 67 CABGs, 44 Medical Mx, and 16 deaths.
Their results are below:
|HEART SCORE||n(%) of study population||Number who reached end point||6/52 mace|
|0-3 (low risk)||870 (36.4%)||15||1.7%|
|4-6 (med risk)||1101(46.1%)||183||16.6%|
|7-10 (high risk)||417(17.5%)||209||50.1%|
The authors report
“The average HEART score was 3.96+/−2.0 in the non-MACE group and 6.54+/−1.7 in the MACE group.”
Now the standard deviations do OVERLAP. So i’m not entirely sure that they can ascribe significance to their finding. The c-statistic (now this is like area under curve, so the closer to 1 the better) was 0.83, which is damn good really. They didn’t do a sample size calculation. So I’m not sure they’ve got enough participants to robustly test this test.
Good things: they’ve reproduced the results on a larger scale, using a computer to calculate the scores (less chance of fudging). I’d have loved a look at their standard clerking form however. As I’m still a bit concerned about the HISTORY segment, as this is enough to move someone from a low to high risk category.
Bad things: Again theres little information about the HISTORY segment of scoring. If they used a computer then either the form MUST have asked for a gut feeling, or got the database to calculate it based on ticked boxes. Again no assessment of inter-rater reliability.
Well the HEART score has been derived in a small ED based on a undifferentiated sample of patients presenting with chest pain. It’s certainly useful, but it does have limitations especially when it comes to history, and if some of the more objective measures change. I do however like how it allows you to consider a minuscule frustrating trop rise that you know is a false positive and still allow discharge.
As with most scores, they shouldn’t replace the grey squashy stuff between your ears, but maybe they’ll help us all risk stratify a little better.
Next time – So I guess what we should talk about next is features of a chest pain history actually increase the likelihood of ACS? What makes a ‘suspicious’ history?