Welcome to the St Emlyn’s Podcast, your go-to source for the latest insights, developments, and discussions in emergency medicine and critical care, with in-depth analysis, evidence-based practices, and practical advice to enhance your clinical practice and professional development. Whether you are a seasoned practitioner or just starting your journey in the field, our podcast aims to provide valuable knowledge and foster a community of learning and support.
This post is the second of a series looking at diagnosis in the Emergency Department.
Listening Time – 18:53
In this post, we delve into the crucial skill of making accurate diagnoses in the emergency department (ED). Our latest podcast episode features Iain Beardsell and Simon Carley discussing the nuances of diagnosing diseases and the strategies we use to decide when patients either have or haven’t got a disease.
What is a Diagnosis?
A diagnosis is essentially a label that we put on a patient to indicate what they have, which then guides our treatment decisions. In the ED, our primary focus is on identifying life-threatening conditions. This approach often involves working backwards by first ruling out serious conditions before considering what a patient might actually have.
Initial Diagnostic Approach
As emergency physicians, our initial approach is to use tests with high sensitivity. These tests are designed to pick up anyone who might have the disease. Once we rule out the serious conditions, we look at tests with high specificity to confirm the diagnosis, as treatments often carry risks. For example, therapies such as thrombolysis come with significant risks, so we need to be fairly certain before proceeding, unlike less consequential treatments like wrist splints.
Understanding Probabilities in Diagnoses
When we say a patient has a diagnosis, we’re essentially saying it’s likely enough to treat. Conversely, when we say a patient doesn’t have a diagnosis, we mean it’s unlikely enough to withhold treatment. This probabilistic approach is vital in the ED and can be surprising to many people.
Case Study: Cardiac Chest Pain
Let’s apply this to a patient with cardiac-sounding chest pain. Our goal is to either rule out or confirm the disease and start appropriate treatment. We start with specific tests to rule in a diagnosis, such as an ECG. A positive ECG with significant ST segment changes indicates a high likelihood of disease, warranting immediate treatment. This approach quickly sorts out high-risk patients.
For patients with normal or near-normal ECGs but still concerning symptoms, we need sensitive tests to ensure we don’t miss anyone with myocardial disease. About 10% of these patients might have underlying issues, so we need to ensure our tests are sensitive enough to catch these cases.
Using Prevalence and Pre-test Probability
To decide if a patient has the disease, we must consider the prevalence or pre-test probability in our population. For example, in patients with normal ECGs and no alarming history, the pre-test probability might be around 10%. This isn’t low enough to rule out the disease but also not high enough to justify treatment without further testing.
Diagnostic Processes in the ED
We use a step-by-step diagnostic process. Starting with the most specific tests to rule in a diagnosis, we then use sensitive tests like high-sensitivity troponin to rule out diseases. High-sensitivity troponin tests are great for ruling out diseases due to their sensitivity. If the test is negative, we can be confident the patient doesn’t have myocardial damage. If the test is positive but not dramatically high, we may need additional tests to confirm the diagnosis.
Each diagnostic step adjusts our patient’s probability of having the disease. Our goal is to reach a probability low enough to safely rule out the disease or high enough to justify treatment. This process is continuous, and we apply it to every patient, whether they have chest pain or another symptom like a headache.
Understanding Likelihood Ratios
We often use likelihood ratios to interpret diagnostic tests. A positive likelihood ratio increases the probability of the disease, while a negative likelihood ratio decreases it. For example, a high-sensitivity troponin test is excellent at ruling out myocardial infarction because of its high sensitivity, though it’s not as good at ruling in due to lower specificity.
Optimising Diagnostic Tests
Diagnostic tests like troponin can be optimized by adjusting the threshold levels. For instance, a higher threshold might improve specificity and thus be better at ruling in the disease, while a lower threshold improves sensitivity, making it better at ruling out the disease. This principle applies to various tests, including white cell counts and amylase levels.
Continuous Assessment and Reassessment
In the ED, we continuously assess and reassess patients. Each diagnostic step, whether it’s asking a question about symptoms or ordering a lab test, adjusts our understanding of the patient’s condition. This iterative process helps us make informed decisions about treatment and ensures that we don’t miss critical diagnoses.
Applying the Approach to Different Symptoms
This diagnostic approach isn’t limited to chest pain. Whether a patient presents with a headache, abdominal pain, or any other symptom, we apply the same principles of sensitivity, specificity, and likelihood ratios. Each question we ask and each test we perform helps refine our assessment and move closer to a definitive diagnosis.
Conclusion
Mastering diagnostic skills in the ED involves understanding and applying probabilities, using specific and sensitive tests effectively, and continuously reassessing the patient’s condition. By focusing on these principles, we can make more accurate diagnoses, provide appropriate treatments, and ultimately improve patient outcomes.
Podcast Transcription
Welcome back to the St. Emlyns podcast, I’m Ian Beardsell, and I’m Simon Carley. This is the second of a series of podcasts we’re going to be doing talking about a key skill in the emergency department, that of making diagnoses and how we decide when patients either have or haven’t got a disease.
So Simon, could you just take a couple of minutes to remind us where we got to after the first podcast and some of the things we talked about?
Sure, I think there are a few things that we talked about last time. The first was what is a diagnosis, and we decided that a diagnosis is the label. It’s just a tag that we put on somebody and we say if you’ve got this diagnosis, it’s a tag that says you’ve got that label, this is what you have, we’re going to do something about it. If we don’t give somebody the label, we don’t give them the diagnosis, we’re saying, well I don’t think you’ve got this, so we’re not going to treat you or do anything about it, but it’s really just a label that we attach to something.
We also talked about how, as emergency physicians, the diagnosis we’re really interested in is the one that’s going to kill you. So we almost work backwards from a perspective of deciding what somebody doesn’t have rather than what they do have. That’s our initial approach to many conditions. Diagnostically speaking, we’re most interested in tests which have a high sensitivity. These are tests which will definitely pick up anybody who might have the disease. That’s our usual first approach. Once we’ve been through that, we start thinking about other tests which are specific and will tell us whether somebody definitely does have a disease because there are consequences to treatment for many of the things that we do. Deciding when that tipping point is about when we’re going to go forward and treat somebody depends on how consequential the therapy is.
Many of our therapies have risks, such as thrombolysis, and we want to be fairly sure about that, whereas other things like putting somebody in a wrist splint, the consequences are pretty minimal so we don’t have to be as sure. If you put all of those things together, it really does put us into a situation where we have to accept that when we say you have a diagnosis or you don’t, we’re really talking about a probability. When we say you have it, we think it’s pretty likely that you do and it’s worth treating. When we say you don’t, we mean it’s pretty unlikely, and therefore by not giving you that label, you’ll probably be okay. But it does mean we’re probabilisticians, not diagnosticians, and that’s a bit of a surprise to many people.
So let’s take that onto the ED shop floor and talk a little bit about how we might be able to use some of that in a patient who’s presented to the ED with cardiac sounding chest pain. We need to get to the stage where we can either say with a degree of certainty that they haven’t got the disease or that we want to rule in the disease and start treating them for whatever it may be, like ACS type treatment. How do we get from a patient who pitches up with cardiac sounding chest pain to that end stage that we’ve just been talking about? We need to go through some sort of diagnostic process. Chest pains are great ones to think about how diagnostic tests really work in practice. Let’s take that group of patients.
You’ve got this group of patients who’ve got potentially myocardial disease or symptoms which are suggestive. Let’s go for the big wins early. Let’s go for something which is really specific which will rule in the diagnosis, a spin test, and that would be an ECG. If your ECG is positive with big ST segment changes, you’re going to go off to the lab. They’ve almost certainly got the disease, well certainly 95% certain, and it’s worth treating that group of patients. We’ll go for a big win early specific test ECG. It’s positive. Get rid of them. Move them onto other things. They’re an easy group of patients to deal with because it’s all protocolised medicine.
Similarly, those who’ve got pretty dodgy looking ECGs can go the same way really. You’re not going to be sending those patients home. So we’re going for those big wins early. We’re then left with a group of patients who, in my experience, have got normal ECGs or very nearly normal ECGs and the symptoms aren’t that dramatic. They’ve not recently had an MI. They’ve not got chest pain at rest but it’s potentially possible and those are patients who we don’t want to send home without doing anything. But we need to make sure that we pick up those who have got myocardial disease. About 10% of that group have actually had a myocardial damage event.
We need something which is sensitive and that’s when we start thinking about sensitive tests which will pick up anybody who might have the condition. To reiterate, the ECG is a relatively specific test because there are few false positives. If we see an abnormality on an ECG, it’s likely that the patient may well have disease. It’s still important to relate it to what the patient looks like. We see some young patients with what might be described as high takeoff and having an idea of their age and a few other bits and pieces helps us. But it’s a quick way of ruling patients in because it’s a specific test. We’ve got those ones sorted. Then we want to now work on saying who hasn’t got the disease. We must have an idea of prevalence in our population and in these terms, we can really equate prevalence to pre-test probability. Can we use those terms relatively interchangeably at the start?
Absolutely. I think it’s sometimes confusing when people talk about them as if they’re very different things. Prevalence in that population is good and that’s going to be our pre-test probability if we’re going to do some testing. For that group of patients, normal ECGs, no particular things in the history, not in heart failure, no major features on examination, about 10% of them have actually got underlying myocardial damage. So it’s still quite high.
So 10% is our pre-test probability. It’s obviously not low enough to say they haven’t got the disease because we’d be missing quite a lot of patients, but it’s also not high enough to start giving them the treatment because the treatment could have harm. We need to take those patients now and move them further down that diagnostic pathway. We should really start with the history, shouldn’t we? Let’s think how we can use some of those things we’ve just talked about but equate them to a patient in the ED. We see lots of patients with chest pain and lots of patients with cardiac sounding chest pain. So why don’t we try and relate a little bit of what we’ve learned so far into a real patient scenario?
A patient turns up with cardiac sounding chest pain. How do we get to the stage where we can either say you haven’t got the disease, you’re safe to go home, or you have got the disease, we think it’s more probable that you have, and we’re going to start treatment on you. What’s the best way to go about that?
Well, great example. Chest pain is a beautiful one if we want to think about how diagnostic tests work in clinical practice. So you’ve got that group of patients you said and they’re cardiac sounding chest pain patients. Yeah, so these are the patients that we describe as having a cardiac sounding chest pain. We want to get them through to be sure they haven’t got an ACS type problem. We’re thinking about things like myocardial infarction and aortic dissection, that kind of thing, and that’s great. I think everybody will understand what we’re talking about. Let’s think about how we do our diagnostic test. Let’s go for some easy wins early. Let’s use something which is really specific and which is pretty good and can rule in. So a spin test. Let’s use an ECG. An ECG is quite a good test because it’s specific. If you’ve got big ST segment changes, that’s good. If you’ve got lots of ST segment depression, that’s also good. It identifies a high risk group of patients who are either going to be definitely admitted and given anti-platelet medication or are going to go off to the lab get PCI or thrombolysis. Use of an early specific test is fantastic. Move them out, dead easy to deal with. By specific, we’re meaning there’s few false positives. So it’s more likely if we see these abnormalities that they’re true positives and the patient has the problem.
So we can use that test. It’s still important, I guess, to have a bit of an idea about the patient because there will be other things that we know about that will cause some of those changes. Having an eyeball of the patient from the end of the bed and having an idea of their age perhaps might give us a bit of a hint. But generally, that’s going to sort out one set of patients. Off they go, decision made, you’re staying in, you’re having your treatment.
How do we go about the rest of the patients who we’ve got with cardiac chest pain who aren’t so obvious?
Well, I think you’re going to be left then with about still about 50% of the patients are going to have a pretty normal looking ECG, symptoms compatible with myocardial disease but no major examination findings and nothing particularly alarming in the history like the fact that they had a myocardial infarction last week. In that group of patients, you’ve still got about 10% of them are going to have underlying myocardial disease if you look hard enough for it, which is still pretty high. So 10%, this is now what we’re describing as our prevalence or we could maybe even call that pre-test probability, we can use those terms interchangeably, that’s okay.
I think so. I think when we’re talking about pre-test probability, it’s quite good that we’re talking about the specific group of patients who we’ve now filtered down to, whereas prevalence people might say, well, it’s the prevalence of people who turn up in your ED. But pre-test probability, I think that’s really essential for an emergency physician to understand that’s our group because the test performance is going to vary depending on your pre-test probability. This is the probability that that patient has disease in the group that we’re talking about. Now we’ve got to a group who have a 10% probability of having a bad outcome or a disease that we want to diagnose.
We’ve said already that’s not enough to say we’re not going to treat you anymore. We want to be less than 2% really for that. But because of the nature of the disease and the nature of the treatment, which can have harm, it’s also not high enough to start doing treatment for those patients. We need to start moving that 10% either up towards a level where we’re happy to treat or down to a level where we’re happy to say it isn’t an ACS type picture. How do we go about that? We’ve got to start with history, haven’t we?
To some extent, we’ve done some of that already. So we’ve got ourselves to the point where on history and examination, we’re about 10%. We’ve done as much as we probably can. We’ve now got to move on and think about laboratory tests. In our practice, we use troponins, I don’t know what you use in Salford. We’re the same with troponin. We’re about to move to a high sensitivity troponin.
Alright, OK. So 21st century stuff for you.
And I would say that because in Manchester, we’ve been doing high-sensitive troponin for a number of years. But that’s because of the wonderful Rick Body, of course.
Yeah, we’re going to use some diagnostic tests. What we’re trying to do with the diagnosis is exactly as you describe. What we’re trying to do is to say that the probability after we’ve done the test is either so low we can let you go home or it’s so high that we’re going to admit you. So we’re trying to move the probability and we do that by a function of diagnostic tests, which you would call a likelihood ratio. A likelihood ratio which is positive, so you get a positive result, it makes it more likely that you’ve then got the diagnosis.
And it moves that probability higher to a certain degree?
Yeah. And if it’s a negative result, it moves the probability lower. And it’s a function of both the sensitivity and the specificity. The likelihood ratio, the more positive it is, the more it will move you up in that direction. In effect, the better the test is for diagnosing. If it’s a negative likelihood ratio, it moves you more down towards that rule out criteria.
So this is again, all about how good that test is at the diagnosis you’re looking for.
Yeah, and absolutely, you can have the same test that’s got a really good positive likelihood ratio and a really rubbish negative one. Or you can have one which has got a really rubbish negative one and a really great positive one. So it does really depend on exactly how you’re using a test. Troponin is a good one, high sensitivity troponin is a good one, because it’s very good at ruling out, because it’s super sensitive, it’s not particularly good at ruling in, because it’s not very specific. And so that affects the function of the likelihood ratio.
So if we have our patient with the 10% pre-test probability, and we’ve got to that via a combination of history and other things, although I think it’s quite hard for physicians to really know what a pre-test probability is. But let’s say it’s about that 10% level, and we do the high sensitivity troponin, in whichever guise you’re going to do that, how many tests you’re going to do. That could well be enough to take us low enough to take us below that 2% threshold and say, we’re happy to stop now.
Yeah, that’s if you’re using high sensitivity troponins just as a yes no type thing. So if it’s either above or below this level, and that is still a function, but that’s still a little bit similar to just using a sensitivity specificity, isn’t it? It’s still a bit like a yes no type question.
I guess the sensitivity and specificity all go to make up the likelihood ratio. So I guess they’re interlinked, aren’t they?
Yeah, they are, but take that one step further. Take that one step further and say, well, okay, say our level of troponin to rule out would be 14 nanograms, okay, that’s fine. Why is it 14? It’s because we’ve chosen that level because at that level it’s particularly good for ruling out. What if we change that level? What if we made it 30? Well, it wouldn’t be as good at ruling out, but would that be better at ruling in? And the answer is it would. So depending on where you put your level on a test like troponin, a continuous variable will affect how good that test is at performing. So you could even have the same test, but with a different value it would be great at ruling in, at a lower value, it’d be great at ruling out. Now we don’t do that very often as diagnosticians, but that is the function of many, many tests that we use. It actually works for things like amylase. It actually works for things like white cell counting appendicitis. Many things which a lot of people would say that test doesn’t work. Well, it does, at very high levels or at very low levels in the middle ground.
And I guess tests like the troponin have been extensively investigated. So we get somebody who’s looked at that for a long time to set that level for us. But as I’m going to call myself without hesitation a box standard emergency physician, I guess I have to develop my own level with the white cell count or those other tests you talked about to say where I’m happy about, oh, do you know what, a white cell count of 18 in this context? I don’t think that’s important. So it’s more difficult with those other less binary tests if you like about how we’re going to use them at the end, those continuous variables.
Yeah, but I think you do it all the time already. I put it to you that if somebody comes in and they’ve got a troponin of 6000, although most of the time we’re told to use, or most people are using high-sensitive troponins as ruleouts, somebody comes through with a troponin of 6000, I think that’s a bit of a rule in, isn’t it?
I guess it is. I guess it is. So we’ve got our patient with the pre-test probability of 10%, and we’re going to do a troponin, a high-sensitive troponin on them. Now, it’s a bit like the D-dimer, isn’t it? It’s a sensitive test. We’re just looking for a negative test because it’s going to take us, we’re just looking for a sensitive test because it’s going to take us to that level where we can rule it out, same as with the D-dimer. What happens with that if we actually don’t get a negative high-sensitive troponin? So it comes back positive. We’re no better off, are we?
Well, as I say, it depends on the level. If it comes back very, very high, you can use these tests because they continue as variables to rule in a diagnosis, but a lot of the time it’ll come back sort of slightly raised, in which case it’s not particularly helpful. D-dimer is perhaps a better example. A D-dimer, which is negative, so below the investigation level, you can use that to rule out in low probability patients. If it comes back as positive, it just means you’re going to have to go on and do further testing. That might be further serial biochemical testing to look for rises and falls, as we do with troponins. Quick interlude here, we don’t do serial testing for D-dimer, of course. We do VQ scans or further imaging to define whether or not the patient has a disease. Or it might be an alternative test entirely, such as a VQ scan or a CTPA for PE. All the time, we’re taking a pre-test probability, applying a test and getting a post-test probability. Until that post-test probability is satisfactory to rule in or rule out, we keep going.
We keep going, and that may be as part of the history. So we said at the very beginning, we start off with that group of patients where we apply the test. The test is the ECG. That has a good positive likelihood ratio. If it shows changes, it’s likely the patient has disease. That takes our post-test probability high enough to rule in. But in this case, we’ve got to the stage where troponin, that pre-test probability is 10%, we’ve done a test, it’s not changed enough. We haven’t got to the threshold of ruling in or ruling out. We need another test. That’s how we think all the time, I guess. We do that with everything we see, whether that’s the patient who pitches up with a headache. We operate a diagnostic test that might be asking some questions about the headache. Each question is in itself a diagnostic test. Each time striving to get to the stage where we can say, “Do you know what? I’m satisfied that in all likelihood you don’t have that illness. I can move on and do something else with you.”
Yeah, you’re quite right. Every little note, every little point where you’re making a decision or asking a question is a diagnostic test in itself. That will change your pre-test probability to a post-test probability.
So Simon, I think we’ve taken everyone a bit further down that line of working out whether a patient has disease or doesn’t have a disease. Using diagnostic testing to take our pre-test probabilities through to post-test probabilities and just adding a little bit of sensitivity, specificity, and likelihood ratios into the mix to try and give that mathematical bent to think about these probabilities. That’s probably enough for certainly my brain today, maybe not for yours, but it might be for listeners as well. Why don’t we come back again in our next episode to take this further and work out what it means to decide to give a patient treatment and how we decide if that treatment is going to be effective and whether even that treatment might harm as well as do good. Please think about what we’ve been talking about. We’d love to hear from you. Get in touch via all of those methods. You can Twitter via the website. We’d love to hear anything you’ve got to say. We’ll look forward to speaking to you very soon. Take care.
Hey, and just for you, yeah, last time I did ask you what your favourite test was. You were very fond of your old SAM machine, if I remember rightly. I like technology. What’s your least favourite test?
You always just do this. I think we’re about to finish and I’m all just chilled out. I’m about to pop it to the fridge, get myself a beer, pack myself on the back and you just land me with a question. So I’m going to answer for you today. Professor Simon Carley, I’m going to say the White Cell count.
Oh, what’s that?
I don’t know. I just think everyone puts a load of emphasis on the White Count as being important and almost never can I think of a time where it changes my pre-test probability significantly enough about whether I’m going to treat a patient or send them home. Because I’ve already made the decision based on everything else. And apart from the occasional patient I happened to pick up via a screening type test, do I have some awful leukaemia, which has happened about once in my career? I can never think of a time it’s been helpful, apart from convincing a surgeon that a patient might be poorly.
Yeah, I think I said, do you know what my favourite is?
I most certainly do.
Yeah, it’s CRP. Yeah, CRP is always that one that emergency physicians like to beat up. And we manage to stop people doing CRPs, I think, which is great.
So, no one needs to do now, I’ve stopped them doing White Cell counts and both of us can be happy.
Yeah, my misses a vowel though. Anyway, see you another time. Take care.
You needed to say the word vowel a bit clearer. Yeah.
Where to listen
You can listen to our podcast in numerous ways, ensuring you never miss an episode no matter where you are or what device you’re using. For the traditionalists, Apple Podcasts and Google Podcasts offer easy access with seamless integration across all your Apple or Android devices. Spotify and Amazon Music are perfect for those who like to mix their tunes with their talks, providing a rich listening experience. If you prefer a more curated approach, platforms like Podchaser and TuneIn specialize in personalising content to your tastes. For those on the go, Overcast and Pocket Casts offer mobile-friendly features that enhance audio quality and manage playlists effortlessly. Lastly, don’t overlook YouTube for those who appreciate a visual element with their audio content. Choose any of these platforms and enjoy our podcast in a way that suits you best!