The Subarachnoid Haemorrhage in the Emergency Department (SHED) Study

Welcome to the St Emlyn’s podcast. I’m Iain Beardsell, and I’m absolutely delighted on this special episode of the podcast to be able to welcome back Dan Horner, who will be well known to many of you as a contributor to the blog and podcast in the past, and Tom Roberts to discuss their study, which has just been published in the Emergency Medicine Journal. More of that in a moment, but first Dan and Tom, perhaps you could introduce yourselves to our listeners.

Thanks Iain. I’m Dan, and hopefully people will know me from previous podcasts and blogs as you say. I’m a consultant in Emergency Medicine and Neurocritical Care working in Northwest England.

Hi Iain, thank you very much and hello listeners. My name is Tom Roberts, I’m an Emergency Medicine Registrar down in Bristol and a clinical lecturer, so a clinical academic based at the University of Bristol.

And the main reason we’re here is to discuss your paper, “Subarachnoid Hemorrhage in the Emergency Department: The SHED Study,” which has many take-home points for anybody who’s working day-to-day in the ED. But before we start, I wondered Tom if you could give us a bit of a background into the authorship of this paper, specifically the Trainee Emergency Research Network, called TERN, which I know you were a fellow for and Dan, I know you’ve led on, but may be unfamiliar to some of our listeners.

Yeah, so TURN was set up in 2018 by Dan and myself, with a lot of support and funding from the College of Emergency Medicine. Essentially, its purpose was to increase research opportunities for trainees, so it was essentially established in what we did is went out to lots of trainees and said, “What’s important to you in terms of research?” And then we designed research that would be easily performed by trainees in busy emergency departments. A lot of our designs are very pragmatic, like this study, and a lot of them, we hope that the questions answer important clinical questions.

TERNwas set up, there’s a fellow in post, we’ve now had four fellows, who essentially supports the day-to-day running of TERN as an organisation, and then there’s always a college professor, so we’re all College of Emergency Medicine professors that provide that senior mentorship. Our studies are large, so generally we enroll participants in patients from around 80 to 100 UK hospitals, so we do things on a very large scale and we try to do things quickly. So that’s their kind of premise of TURN and where it’s come from.

And Dan, I know you were the Royal College of Emergency Medicine professor who was instrumental in setting this up when it started. Give us a bit of an idea about how you chose this particular study, so what was it about subarachnoid hemorrhage that made you think this was a topic that was worth pursuing?

We know that research-active institutions tend to have better clinical outcomes from a lot of data over the last 20 years, and in the UK in particular, we’ve had a UK government policy paper, multiple recommendations from the Academy of Medical Royal Colleges that research should be embedded in clinical care and it should be front and centre. But we know that’s really tricky in emergency care, and the academic support is a bit of a geographical lottery, you know, some sites will be very research active, some sites won’t do anything at all, and trainees will get rotated between the two and don’t often have much choice about it.

The concept of TURN was to try and create research opportunities for any trainee in any hospital. So when you’re thinking about it like that, you have to think what kind of study can we do realistically that will allow us to offer opportunities at every hospital site? And that tends to limit your methodology, so running a randomized controlled trial is very complex, testing a drug or intervention is very complex, and requires a lot of research governance, a lot of research oversight and research support.

But what we can do with TURN, and what we’ve already done very successfully with several studies, is we can look at the care that’s delivered routinely in our emergency departments for specific clinical presentations. And we can have trainees perhaps collect a little bit more data and follow up those patients a little bit more robustly than they would in usual practice. By doing that, you can create these tremendous snapshots of what we do in emergency medicine at the moment, how good we are compared to how good we think we are at making certain diagnoses, how good some of our tests are compared to how good other specialties might think they are when it comes to certain diagnoses.

We started talking about these ideas in the early formation of TURN, and subarachnoid hemorrhage or the early exclusion of subarachnoid hemorrhage was an absolute single contender for this type of observational cohort study where we just look at how we evaluate these patients and we collect some robust data and we look at the success of our diagnostic strategies. Subarachnoid hemorrhage is right up there along with aortic dissection and PE as one of the nightmare diagnoses that emergency physicians don’t want to miss. So this is a topic that’s close to all our hearts, I think.

And if you want to read the paper, it is available in the Emergency Medicine Journal. It’s entitled “Subarachnoid Hemorrhage in the Emergency Department (SHED): A Prospective Observational Multi-center Cohort Study,” and it was published in September 2024. There is also a comprehensive blog post on the St. Emlyn’s website, so you may want to refer back to that after our conversation.

But Tom, let’s dive into this paper and let’s think about what was it that made you choose this? What is the background to subarachnoid hemorrhage? Where are we now, and what was the study aiming to tell us about where we might be in the future?

So in terms of where we are now, this study, we came up with the idea in 2018 and actually the literature and the evidence base around subarachnoid hemorrhage has moved on obviously since 2018. So there’s been new UK NICE guidance, which is essentially our national body that produces guidance which says that ruling out subarachnoid hemorrhage is safe if that CT brain is done within six hours of headache onset and that’s consistent with a lot of literature that’s come out of Canada; the Perry team have done a lot of work on this.

There’s also been some papers, like one from New Zealand published in 2022, which was a retrospective review of all patients presenting with subarachnoid hemorrhage, and they actually identified that for patients presenting with aneurysmal subarachnoid hemorrhage, the CT brain was 100% sensitive, so it didn’t miss any cases of aneurysmal disease within the first 24 hours. The NICE guidance and the Perry paper have always said that CT is very good up to six hours, but it’s less good after that, and that’s consistent with what we know about how blood behaves in the brain.

Essentially, after you bleed, the blood will start to break down, so it becomes less clear on CT; it’s not bright white, it becomes less and less white, making it more challenging to identify. It also starts to get diluted by the CSF as CSF flows around the brain. The six-hour cutoff is fairly arbitrary, so that was an arbitrary time cutoff created by the team out in Canada. The question is, can we push that cutoff further than six hours? Can CT brain be good enough and reliable enough to rule out subarachnoid hemorrhage beyond six hours?

And not just subarachnoid hemorrhage, but we started to think, especially after that paper from New Zealand, are the test characteristics different for aneurysmal subarachnoid hemorrhage and non-aneurysmal subarachnoid hemorrhage? As we go through the paper, I want you to listen and be aware that we use the term subarachnoid hemorrhage as a catch-all for all types, but there are two different types here that we’re talking about, and they have different mortality and morbidities. As we will discuss, CT is potentially better at picking up aneurysmal disease than non-aneurysmal disease.

So we’ve got this disease that we all have heard of: subarachnoid hemorrhage. One of the things I found interesting reading both the blog post and the paper was that we’re not actually sure of a definition of that, and what is a subarachnoid hemorrhage? How do we define them? We’ve all heard about thunderclap headaches, but it may be worth just thinking about: what is a thunderclap headache, Dan? And how did you define that as part of this study?

I mean, acute severe headache is a common presentation to EDs, I think it accounts for between one and two percent of annual admissions for the UK. Tom and I have done lots of work with international colleagues on acute severe headache and different definitions, and when you start talking to people, it becomes quite apparent that thunderclap is a fairly subjective term. But what we do know is that in that acute severe headache cohort, around 90% will eventually be diagnosed as a secondary benign headache, such as a migraine or tension headache—benign’s perhaps not the right word, but what we mean by that is no structural pathology evident on testing and imaging. Around 10% will have a serious diagnosis among these acute severe headache patients, and some of those will be tumors, some of them will be bleeds, and some of them will be other sinister pathology. These are the one in ten that we definitely don’t want to miss when we see a patient with an acute severe headache in the ED.

When we’re talking about thunderclap, we looked at previous literature to develop our definition for the SHED inclusion cohort, because what we really wanted to look at were these patients where

the headache is not just a gradual headache that’s been building slowly in someone with a long history of migraine. We wanted to capture these acute severe headaches that worry emergency clinicians. Interestingly, Perry et al.’s definition includes a headache that reaches peak onset within an hour, so you’ve got an hour from the start of your headache to it developing to its absolute peak. Now, other people might say their definition of thunderclap is more like reaching peak within five minutes or ten minutes, but the variations in the literature are pretty common. We wanted to pick something that had been used previously, and something that had a bit of reproducibility to it.

What we do know from studies of headache is that people are very variable in writing down timings and subjectively describing the type of headache and getting accuracy around that. We know that you still get subarachnoid hemorrhage in patients that don’t declare they’ve had a thunderclap headache and talk about more nebulous symptoms. So we were desperate to have a broader umbrella to catch all these subarachnoid hemorrhages in the department, rather than restrict our inclusion criteria to something pretty nebulous.

So we’ve got this group—it’s a fairly big group, isn’t it? One to two percent of all emergency department attendances, of which ten percent of those are going to have something serious. So this is a big deal—lots of patients attending every emergency department across the country every day. And you’ve defined there really nicely that actually we’re not doing very well on saying what a subarachnoid hemorrhage is or what the symptomatology is, but we know there’s plenty of patients we need to find, and then we’ve got this difference between aneurysmal and non-aneurysmal subarachnoid hemorrhages. So a really interesting cohort to look at.

Tom, on to the methods, we’ve got all of these patients, got all this data that we can collect. Tell us about getting the data that you needed, involving other colleagues, and what were the major inclusion and exclusion criteria you had for the study?

Our inclusion criteria, which Dan alluded to, were acute severe headache with maximum onset within one hour, and a key inclusion criterion was that these patients had to be GCS 15 with no focal neurology. So once you develop focal neurology, the kind of pre-test probability for something going on internally is a lot higher. These are patients who are sat there, they’re alert, GCS 15, and had this maximum intensity within an hour. There’s a long list of exclusion criteria, which are fairly obvious, like previous subarachnoid hemorrhages or transfer from another hospital with subarachnoid hemorrhage—you can have a look at them in the paper.

Dan mentioned that people record headache onset time very poorly, and this is something we identified with some audit work prior to the study. So what we did, and this is how we were able to recruit across 88 emergency departments, was create a simple A4 sheet of paper with an inclusion checklist. This included key pertinent information that we know was poorly collected from routine records. Clinicians who were recruiting patients to SHED would write the date and onset of headache clearly documented, which meant we had really good capture of that, and then they’d answer six key questions around the Ottawa clinical decision rules, like: was there neck pain and neck stiffness, was this a definite thunderclap headache, etc.—key pertinent information useful for our analysis.

After that, the clinicians would see the patient 24/7 in their emergency department, do the inclusion checklist, and that was their job done—the patient was enrolled. We then retrospectively looked through their hospital records to document their CT scan outcomes, lumbar puncture outcomes, and any other pertinent information, including onward investigations or neurovascular MDT discussions.

In terms of how someone was diagnosed with a subarachnoid hemorrhage, we had a clear diagnostic criterion derived again from the Perry criteria, to keep things consistent. To be diagnosed with a subarachnoid hemorrhage, you had to have blood present on CT brain, a lumbar puncture consistent with xanthochromia or positive according to the UK biochemist guideline, or greater than 5 x 10^6 red blood cells in your lumbar puncture plus an aneurysm present on angiography (either CT angiography or DSE). We also followed patients out for 28 days to ensure that there were no missed cases. If you had a positive subarachnoid hemorrhage within the subsequent 28 days, you were also defined as being positive for subarachnoid hemorrhage.

Something unique to this study was that we also had a committee which looked at all the unclear findings. This isn’t an easy group to definitely define as having subarachnoid hemorrhage or not—there are lots of false positives and false negatives. So we had an independent committee of a neurosurgeon consultant, an acute medic, and a consultant emergency physician who reviewed all the cases where the findings were unclear, and they independently adjudicated whether this was a subarachnoid hemorrhage or not and whether the initial CT identified or missed that subarachnoid hemorrhage.

You mentioned the Ottawa subarachnoid hemorrhage clinical decision rules briefly. That’s interesting mainly because it’s incredibly sensitive but poorly specific, and I know it’s being used as a sort of initial filter if you like before patients get into this sort of investigative testing. Tell us a bit more about that clinical decision rule and whether you think it’s actually useful in the emergency department.

Yeah, so Ottawa are well known for their clinical decision rules, aren’t they? They’ve done them for ankles, the cervical spine, and I think there are knee rules on there as well. They’ve got tremendous expertise in this kind of research design and methodology. So it’s not surprising that they’ve taken this difficult population and applied some of that expertise to say, well, what about if we just build a set of clinical criteria that reduces your probability to such a low extent that we could say you don’t have a subarachnoid hemorrhage without any test at all—no CT, no lumbar puncture, nothing.

Similarly to a clinical decision rule like the PERC rule, which many of your listeners might know, it’s a set of clinical criteria designed to reduce your pre-test probability to such a low proportion that you don’t think it’s worth investigating at all; you can just scratch off the possibility of disease. The problem with these rules is that sometimes they can be developed to provide a very high sensitivity, but that’s not really the only question. The question is also about specificity, so there’s no point having a clinical decision rule that identifies everyone as potentially having a subarachnoid hemorrhage, because then you’re essentially using that rule to no effect.

You are using it to try and assure yourself that no one has a subarachnoid hemorrhage, but actually, you’re convincing yourself that most people might have. The Ottawa clinical decision rule has always suffered a bit from that, so sensitivity is high in the validation cohorts, but the specificity has always been poor across a number of settings. That’s because it has some very generic criteria in it—like if you’re over 40 years of age, you fail the rule automatically and need further investigations irrespective of other criteria.

When we looked at the Ottawa rule in a UK population, because it’s never really been validated in a UK setting apart from some retrospective work, we found that it didn’t add much incremental value. The post-test probability after applying the rule in this cohort of over 3,000 patients didn’t really shift very much. So you end up with a rule that doesn’t reassure many people but tells you to investigate lots of people. When you apply that rule in practice, does it offer any benefit to you as a clinician or the patient in front of you? That’s pretty questionable, and we didn’t really feel from our results that it could be recommended for use.

Dan, you’ve touched on a few things there about Bayesian analysis, and if you’re interested in sensitivity and specificity, likelihood ratios, and all that sort of thing, there’s plenty more to look at on the St Emlyn’s website, and we’ve done podcasts in the past. It is worth trying to get a handle on those things, but anyone who has used them regularly will know that good sensitivity but poor specificity means that your likelihood ratios are not going to be very helpful. They don’t really change your pre-test probability to a post-test probability that helps with any decision-making at all. So it’s really attractive if we can rule things out, but that’s got to be balanced with a specificity that helps us to try and rule things in. Ottawa was there as a filter, but probably in our setting, not that useful, which makes your findings even more important to us.

So we’ve thought a bit about the people you’ve recruited, the types of patients we’re talking about. Tom, what happens next? You’ve collected all your data. What did the data tell us about these patients who are presenting to the emergency department with acute severe headache that’s reached its maximum intensity within an hour?

We ended up recruiting 3,633 acute severe headache patients from 88 UK emergency departments, which is a great result in terms of the methods that we used. Of that, there was a prevalence of subarachnoid hemorrhage of 6.5%. So that means patients in your emergency department presenting with acute severe headache have a prevalence of disease of 6.5%, which is, I think, probably greater than people might think. So that’s a key point that people should take away.

In terms of the key question: What was the sensitivity of CT brain? I want to take a second to step away from sensitivity because I think your listeners and even myself, who have been doing this for years, find sensitivity and specificity incredibly difficult to understand in terms of how it is useful for your patient. Post-test probability is something you can take away practically to discuss with yourself, your patients, and your colleagues.

So when we talk about pre-test probability, we are talking about

the prevalence of subarachnoid hemorrhage in our cohort, and post-test probability uses the data from sensitivity and specificity to create a likelihood ratio, which then creates the actual value: your post-test probability, which is easily understandable for everyone. If you look at our cohort, this is table two in our paper.

From 0 to 6 hours, our prevalence of disease is 13.1%. So, our pre-test probability of subarachnoid hemorrhage if you present within six hours is 13%. Our sensitivity is 97%, which means that we do miss some cases of subarachnoid hemorrhage, and our specificity was 100%. What does that mean in terms of your post-test probability of disease? If you present with subarachnoid hemorrhage with acute severe headache and have a CT brain, then the likelihood of you having a subarachnoid hemorrhage is 0.5% after your negative CT brain. The confidence intervals around that are 0.2 to 1.3%.

If we look at times greater than six hours, so between 6 to 12 hours, our prevalence drops to 7.8%, the sensitivity drops slightly to 94.5%, but our post-test probability remains consistent at 0.5%, again with confidence intervals between 0.2 to 1.3%. If we stretch out to 12 to 18 hours, the prevalence drops again to 5.3%, sensitivity remains about 94%, but our post-test probability remains at 0.3% with confidence intervals of 0.1 to 1.7%.

We know that tests are good; sensitivity tells you how good the test is, but that value on its own isn’t important. It’s also about what population you are applying your test to. So what we find and what I think is important in this study is to say that yes, your test gets worse with time—we understand why that is—but importantly, the actual likelihood of disease doesn’t significantly change with time, especially within that 0 to 18/24-hour period.

That was the key result that we want people to take away, and you can read into that as much as you want and discuss it in your journal clubs and as a department. There’s some interesting data, as Dan was talking about with the Ottawa clinical decision rule. Specifically, your pre-test probability of having a subarachnoid hemorrhage if you’re under the age of 40 is 2%, so your pre-test probability before applying the Ottawa clinical decision rule is 2%. If that test is positive, it increases the likelihood of disease from 2% to 6.9%, and if it’s negative, it decreases your probability of disease from 2% to 1.4%.

When you talk about sensitivity and specificity, that’s the impact on post-test probability. So what we’re saying is this test only marginally increases your chance of disease if it’s positive and only marginally decreases your risk of disease if it’s negative. That’s why when you think about it practically in post-test probability terms, it might not be that clinically useful for you on the shop floor of the ED.

So this is a really important reminder, isn’t it, that actually we’re dealing in probabilities and not definites? We always seem to talk in black and white in the emergency department: they’ve got this, they haven’t got it, we’ve ruled it out, we haven’t ruled it out. And actually, even in this cohort that we’re talking about, when we’re talking about the rule-out of a CT scan within six hours, something that we’ve all hung our hats on and really depended on, you still have a 1 in 200 chance of having a subarachnoid haemorrhage, a 0.5% probability after that negative CT.

We have to understand that we are not perfect, and so there will be patients who we do not diagnose. I don’t like the phrase “missed” because it’s not that we’re trying to miss it. Far from it; we just haven’t managed to diagnose it with the technology and the patient we have in front of us. But it might surprise many listeners to think that actually these things are not as black and white as we thought, even when we’re doing evidence-based investigation.

Perhaps that’s one of the big things to take away from any paper like this: it’s all about probability. And so when you have colleagues who talk in definite yes and no’s, black and white—they have got it, they haven’t got it—whatever the disease is, it’s worth questioning whether that’s actually real, because it’s actually more likely a probability that makes the disease more or less likely, and getting to a threshold where we’re happy to start treatment for the disease, or conversely, to say, “We’re fairly confident you haven’t got it.”

I think this is one of the times where shared decision-making is important because we can involve patients in understanding their chance of having these issues, these diseases. Of course, how accurate we need to be does depend on the nature of the disease and how dangerous that is, but I think most data would suggest that emergency physicians are happy excluding if the chance of having a disease with your post-test probability is less than 1%. But again, that does still mean we are saying we accept one in a hundred patients may have the disease, but we haven’t actually managed to diagnose it with the testing we have.

There’s loads more on the website about this. Simon Carly has done lots of work about being a probabilistician. I do recommend you go back and have a look, but this paper really highlights how we work in that way, and we don’t work in definites.

So we’ve got these answers now, Dan, of where in your cohort we think the post-test probability lies after CT scanning. And it seems to me that that golden six-hour time, if the prevalence of disease in your cohort is, probably can be extended. What do you think this means for people in the emergency department who are doing CT scans beyond six hours, maybe up to 24 hours, and the need for a lumbar puncture in those patients?

That’s the crux of this, I suppose, in that we found that extending the window for a CT exclusion of subarachnoid hemorrhage may be safe. I say “may” because this is a prospective observational cohort study—we’re not randomizing patients to different diagnostic strategies here. We’re essentially following up on NICE research recommendations that already say we need further data on the safety of CT exclusion strategies out to 24 hours to see what the numbers look like.

Our numbers look pretty good, as Tom says. The post-test probability of CT between 6 to 12 hours was 0.5% with similar confidence intervals. At 12 to 18 hours, we’ve got fewer patients, so although we’ve got a lower post-test probability at 0.3%, our confidence interval, a 95% confidence interval, upper range extends up to 1.7%. And then at 18 to 24 hours, it starts to get a lot greater; our post-test probability then was 1.5%, and the upper limit of the confidence interval goes up to 4.1%.

It’s interesting that you mentioned thresholds, and you talked about the kind of magic number of less than 1% in terms of what emergency physicians think. There has actually been work done on different diseases looking at test thresholds and the kind of acceptable proportion of patients that you would fail to diagnose with that strategy, and what people would accept. If you ask neuro-interventional radiologists, emergency physicians, and neurosurgeons how willing they’d be or what post-test probability they would accept without further investigations, you get slightly different answers, but they tend to range between about 1 to 2%.

So we’re in exactly the right territory here of reducing that post-test probability down to a level that people feel would be acceptable, up to almost 18 hours, we would argue. This is a really important finding and certainly something that you can share with patients when you’re having a discussion. What we are saying to people is, we can’t categorically guarantee you haven’t had a subarachnoid hemorrhage without some further tests, but the chances of those tests being positive are probably about 1 in 200, so we need to do 200 lumbar punctures in people like you to pick up the one case that we’ve missed.

Now might be a time to come on to the actual importance of that one case, because this is really dependent then on whether that subarachnoid hemorrhage is an aneurysmal subarachnoid hemorrhage, which will need further investigations and may need a coiling procedure or a clipping procedure to make sure that the risk of re-bleeding is dramatically reduced, or it might actually be a non-aneurysmal subarachnoid hemorrhage, which in our experience of this study, were all managed conservatively with nothing more than outpatient follow-up.

So now you’re talking about, do we lumbar puncture 200 patients to pick up the one person who’s had a non-aneurysmal subarachnoid hemorrhage that we can reassure and say, “We’ll see you in clinic in six months,” or do we do none of those 200 lumbar punctures?

Tom, you highlighted earlier the difference between aneurysmal and non-aneurysmal, and Dan’s touched on it there. Now we’ve got to the stage of knowing what your post-test probability is, but what happens next? There are some things to consider about what happens for those patients, and there is a population issue here about the knock-on effects to other patients in your department, and their chance of being admitted. There are also the side effects of the procedure we’re talking about. I’ve had a lumbar puncture. In fact, I ended up

having two lumbar punctures because the first one didn’t actually happen in the way it needed to, and they are not comfortable. So there are side effects to all these things.

Why do we need to track down these aneurysmal subarachnoid hemorrhages? What makes them so important? And does that change how we view the importance of doing lumbar punctures in these patients with negative CTs who present later?

The key problem here is this is all very grey. Everything we’re talking about here is stuff we are hypothesizing, yes, with an evidence base, but we’re hypothesizing. The way I’ve come to understand this and think about it is that aneurysmal subarachnoid hemorrhage is an arterial bleed; therefore, when it ruptures, it’s likely to be a large volume bleed. A large volume bleed is more likely to be detected on CT brain. A non-aneurysmal subarachnoid hemorrhage, which we think is probably venous, is therefore more likely to be a small volume bleed. You can get large volume non-aneurysmal subarachnoid hemorrhage, but the cohort of patients who are LP’d are CT negative. So you’re more likely when you go through your lumbar puncture to pick up a small volume bleed that is more likely to be of non-aneurysmal nature.

If you look at our zero to 24-hour cohort, there were 2,008 patients in that, and there were two patients who were missed with aneurysmal subarachnoid hemorrhage. Both of those have got really interesting stories around them, and this is where you get into when you look at any of these papers, you go back to the Perry papers and you look at missed aneurysmal subarachnoid hemorrhage, there’s always a human element to why that diagnosis wasn’t picked up initially.

For example, the aneurysmal subarachnoid hemorrhage in our paper that wasn’t picked up on CT between 0 and 6 hours, that patient presented with a strong family history of aneurysmal subarachnoid hemorrhage. The treating emergency clinician did a CT angio on arrival because their pre-test probability and their clinical gestalt were so high that it identified two large aneurysms. They then had a subsequent lumbar puncture, which was positive for xanthochromia, and they’d presented with a thunderclap headache, so they went ahead and coiled those aneurysms. Why those aneurysms weren’t picked up on plain CT brain, I don’t truly understand, but if you think about it, the ED clinician picked that up from the outset because there was such a strong family history.

The second case was a patient who presented at 18 hours, had a fairly common trajectory for these patients, was admitted, had a negative CT, an inconclusive lumbar puncture, an MR angiogram, which didn’t show any blood but did show an aneurysm, and then because they had greater than 5 x 10^6 in their lumbar puncture plus an aneurysm subarachnoid hemorrhage according to our diagnostic criteria, they did go ahead and have coiling.

So those are the only two missed cases. There’s a lot more non-aneurysmal cases that were missed, and I think you always have to just take any evidence like this and then put it in front of the patients that you’re seeing. Is there something about this patient or something about the diagnostic test that you are questioning? If you think your patient looks really sick and you’re suspicious of an aneurysmal subarachnoid hemorrhage, definitely go ahead and further investigate them. We’re not saying that all these patients should be discharged saying that everything’s fine because there is always that risk.

We’re saying we are getting to the realms where we don’t think you’re going to miss by sending someone home an aneurysmal subarachnoid hemorrhage if you use an individual clinician’s judgment. We’re talking in the realms of one in a thousand for aneurysmal disease. You’re not going to see a thousand acute severe headache patients in your career—you might feel like it, but you’re not—so the chances are, for you personally, you can be relatively safe. If that patient looks well and you don’t have any other risk factors that you’re concerned about, you probably can discharge them knowing there’s almost certainly not an aneurysmal diagnosis there.

You might miss non-aneurysmal disease, and that’s where you’ll miss one in two hundred, but we don’t understand, and neurosurgeons don’t understand, what this clinically means for patients. That’s where the kind of research needs to go, but how you get clear clinical answers for that to a level of probability where we have complete certainty, we’ll never get there.

We are the owners of patients presenting with acute severe headache. It’s up to us to understand how we want to manage them and where our risk thresholds are. The NICE guidance suggests discussing with a specialist if you’re unclear on what the next step should be after a negative CT. I would argue the specialist you need to discuss with in terms of acute severe headache patients is probably an emergency physician, because we are the specialists in undifferentiated acute severe headache.

This SHED study data provides you with the power to understand patients’ risk for specific conditions. It’s one in a thousand for aneurysmal disease up to 24 hours, and it’s one in 200 for non-aneurysmal disease, and that’s what we need to have in our brains. We always need to go back to that patient and think negative CT, what do I think is actually going on here? And go on from there.

I think one of the really interesting things you said there is in the patient you mentioned who had a strong family history. Obviously, their chance of disease, their pre-test probability, is already greater than the prevalence in the cohort, which generally are quite low risk. In front of you already had a patient with a higher pre-test probability, and actually, if they had a negative CT scan, if you’re doing Bayesian thinking, even with a decent negative likelihood ratio, it still might not take you below that 1% that we’ve talked about earlier to satisfy you that you’re going to say to that patient, “Okay, I don’t think you’ve had an aneurysmal subarachnoid hemorrhage.” So there is that difference between the patient in front of you versus the population that you’re talking about.

The other thing I really want to emphasize, which I couldn’t agree with you more, is that emergency physicians have to take ownership of all of these different presenting complaints which have a serious diagnosis related to them, but we are the experts in ruling out those serious diagnoses. So it’s not just subarachnoid hemorrhage and headache. It’s acute coronary syndrome and chest pain. It’s aortic dissection in chest pain. It’s PE and shortness of breath. Yet, for some reason, we’ve decided we want to give that decision-making away to people who are used to seeing patients with disease, rather than the patients who are trying to say you don’t have the disease. One thing from this discussion, I really take away is that, yeah, we need to be in charge of that. All too often, at the end of algorithms, it says, “Talk to a specialist.” Somehow emergency doctors think that’s never them. But actually, it really is quite a lot of the time.

Dan, what would you say to people who are going to take this paper to a journal club? What should they do with it next?

I think they should take it into their journal clubs and critically appraise it. Tom and I will give you a rose-tinted view of it because it’s been Tom’s life work for the last eight years, and we’re both invested. But we’re very interested to hear what people think. The trainees in the UK delivered this study, and we’re now interested to hear what they have to say about the results.

In terms of whether you can take these findings forward to influence practice, I think the first thing this paper does is tell you that the NICE recommendations, which have been contentious and have been debated online, are probably safe. That using a CT within six hours of headache onset to exclude the diagnosis of subarachnoid hemorrhage on its own does actually confer a very, very low post-test probability, and there will be patients that present atypically or unusually or just give you a really bad belly feeling, and of course, you have to do whatever you think is best with those patients.

But for the vast majority where patients have a GCS of 15 and no focal neurology and you have a negative CT within six hours of headache onset, their post-test probability is rock bottom, and you have national guidance to support discharging those patients without recourse to lumbar puncture. Between six and 12 hours of headache onset, between 12 and 18 hours of headache onset, this data suggests that you have similar post-test probabilities. So for me, I’m going to sit down with patients and be a bit more informed about their discussions.

Interestingly, in this study itself, we saw lots of patients presenting after six hours of headache onset who didn’t get lumbar puncture in routine practice. I feel like we’re already doing this a little bit. We are already seeing people at 8, 10, 12 hours and saying, “Your CT is negative. I could refer you for a lumbar puncture, but is that what you want? Are you willing to accept the side effects and consequences of that?” A lot of patients are saying no because ultimately this is the patient’s decision. We can only present the probabilities and the potential disease and then we have to ask them what they think.

So I feel like we’re already doing that, but this paper adds a bit of weight to those conversations, and it perhaps might push us more towards having frank conversations about post-test probability and making sensible admission decisions.

I think I’ve learned a lot from our conversation and there’s a lot to think about and a

lot for people to take away and discuss. As you say, if you’d like to get comments into Tom and Dan on the blog post, please feel free to write in the comments section there. We’d love to hear what you think.

I want to make three quick points that I think are important to consider in this SHED study before we finish off. So number one, this discussion around missed diagnoses is interesting. Actually, if you look at our cohort, 90% of acute severe headache patients were scanned or had a CT brain. When you look at all the retrospective literature around why patients have a missed diagnosis, it’s because they weren’t imaged. So the key intervention you can do is image these patients, which we’re now doing at a really high rate and a higher rate than we’ve ever done before, which I think is capturing a lot of these patients that previously might have been missed.

The second thing to say is in terms of that acute severe headache cohort, 86% of our patients presented within 10 minutes. I’ve already had discussions with acute medics locally who said, “Well, this isn’t the cohort that I’d be interested in. I’m only interested in thunderclap headaches.” But when you look at the onset of headache, 86% presented within 10 minutes, which probably is the cohort that we’re interested in.

The final thing to say is, how does this study actually implement the flow within your department? So certainly, it’s really interesting to note that you’ve got pathologies; patients presenting with acute severe headache have a prevalence of disease of 6.5%, and actually within six hours, if they present within six hours of their headache, that’s up to one in 10, so 13%. So my question to you and your department is: are you expediting these patients to have a scan? Because number one, they’re likely to have something on it, as in one in 10 is quite a good pickup rate, and number two, if you actually look at rebleed timings for aneurysmal subarachnoid hemorrhage, if they do have an aneurysmal subarachnoid hemorrhage that you don’t yet know about, the median time to rebleed is probably within three days. But actually, if you look at some really good data, it says that might be as early as 180 minutes.

So you’ve got a reason in terms of exclusion of disease and a reason because these patients might deteriorate to prioritize these patients for a CT. When we’ve got waits of 12, 24 hours in the emergency departments and long waits to be seen, can you triage that patient for an early CT?

The last thing I just want to ask you, Tom, really is from a career perspective. You were the TURN fellow. We’ve talked a bit about what TURN meant. You mentioned earlier what you’re doing now. Where does this lead you with a career as far as academic emergency medicine goes? Because it’s a growing field. What’s your next steps and what’s your next project?

In terms of academic emergency medicine, I fell into this accidentally. I was really interested in improving, essentially, how we work as trainees. I got frustrated with audits and QIPs. I didn’t think they were making sustained practical change to departments. I thought, actually, sat there and calculated the number of hours that trainees spent on audits and QIPs and then extrapolated out to the population, and thought, this is too many hours not making meaningful change. So I thought, there must be a better use of time, like creating a research network. So that’s where I got into this, and that’s what really interests me.

Moving forward in terms of an academic career, I’m super interested in supporting others who want to do it. My portfolio at the moment is massive. I’m working with Ed Carlton in Bristol; we’ve just submitted a grant to look at large versus small-bore chest drains in the hemo-pneumothorax. I’m also currently working with the National Child Mortality Database looking at the mortality of children who’ve died of stab wounds, which is pretty horrific stuff, but we’re looking at both the injury factors that might contribute to death and also the pre-injury factors—these kids are well known to social services and police as well—anything we can do to prevent them getting into these situations in the first place.

I’m also working with an evidence synthesis group out in York to take the SHED data and reanalyze it with a whole bunch of other data on subarachnoid hemorrhage to work out the most cost-effective way to manage acute severe headache patients. And then some other stuff on the side as well. The great thing about academic emergency medicine is you get time to go away and actually really focus on things that you care about in the emergency department. I’m really lucky I get Mondays, Tuesdays, and Wednesdays just for research. It’s really important that I have my clinical days—I still work full nights and full weekends—to understand what the problems are in emergency medicine.

And finally, TERN’s next study coming up in January or February this year is looking at escalation area care in emergency departments. We’re going to recruit everyone that’s being cared for in an escalation area and understand who these patients are. The hypothesis is that they’re probably older patients, vulnerable patients, and probably patients from high areas of deprivation. So we want to understand what the impact of escalation care is and what proportion of patients are being cared for in our emergency departments in these areas. That’s coming in January or February, which will be a hugely interesting piece of work and very easy to recruit to as well. We’ll hopefully create a live dashboard so you can see what’s happening around the country during our snapshot periods.

I think that’s it, but if anyone’s interested in academic emergency medicine, definitely get in touch with myself. Dan’s a great mentor.

It’s been great having you both on the podcast, and thank you so much for giving us an insight into not just the study but how we think as researchers, how we think as clinicians. The real other take-home for this is we’re obsessed at the moment with process in emergency medicine, and somehow we sort of forget that if we get clinical excellence, processes will get better. One of the big take-homes is that there’s probably a cohort of patients who don’t need to stay in the hospital, who don’t need lumbar punctures. Your research is helping inform that, which will help inform speeding up our processes and improve emergency department care for others as well.

Research isn’t some esoteric thing sitting in a lab; it’s got real-world effects. I hardly encourage anybody listening who’s interested to contact Tom and to think about the TURN group and get involved.

Tom, Dan, thank you again for your time, and I really appreciate you joining us today.