Estimated reading time: 7 minutes
There is no doubt that sub-arachnoid haemorrhage is an important diagnosis in the emergency department. It is arguably especially important in the cohort of patients who present with lesser signs/symptoms as they are the group in whom preventative measures may be effective. Sadly, the potential for positive outcomes from the GCS 3 patient with a massive bleed is not great, but perhaps we can prevent that if we are able to identify patients with small bleeds, before they have a big one.
Our current strategy of investigation for ambulant patients with concerning presentations is to perform a CT scan of the brain. If that is negative, or if the CT scan was taken over 6 hours from symptom onset our default position is to then perform a lumbar puncture. In reality that often means a shared decision making process with the patient as we know that the yield of positive outcomes from the LP is very low in this group of patients and many patients elect not to proceed.
The data that underpins this approach has changed over the years, largely as a result of the improved performance of CT scanners, and probably of the radiologists who interpret them. Back in 2015 @cgray reviewed the evidence and concluded that a CT scan effectively excluded an SAH using some of the more recent scanner technologies and appropriate reporting, but have things improved since then?
This week we review a paper that might help us answer this question based on more recent data. The abstract is below but as we always say, please read the full paper for yourself and come to your own conclusions.
What kind of study is this.
This is another retrospective review of data. In general we are not fans of this design at St Emlyn’s. Retrospective designs struggle for many reasons, but one of the major concerns is that the data is not specifically tied to the question and may not be of the highest quality. There are other concerns too which you can check out on the podcast I did with Rick some years ago.
However, retrospective cohorts can be helpful to generate hypotheses and in some cases where we just can’t get the data from other means. In this case it may be useful, but we must be a bit cautious.
Who was studied?
The authors sought and documented the results from patients diagnosed with a sub-arachnoid haemorrhage at a single hospital in New Zealand. The key point here is that it is a review of patients who HAD an SAH diagnosed, and that’s not the group of patients I am interested in. As an emergency physician we deal with the group of patients who MIGHT have an SAH. These are very different groups and in terms of diagnostic test evaluation they are chalk and cheese. The basic principle is that diagnostic tests will always appear to perform better if you’ve already selected out the patients who have the condition that you are looking for. Put another way, diagnostic tests should always be evaluated in the population in which you would use them.
What did they find?
The authors looked over a 10 year period (2008-2017) using hospital records. They identified 728 patients with SAH during that time but after exclusions just 346 remained for analysis.
224 patients did not have an accurate time of symptom onset and so that was estimated by the research team as 30min before admission (I have a real problem with this for obvious reasons). With this in place the average time from symptom onset to CT was 3 hours, which logically would suggest that the estimated times are probably shorter than reality. Whilst that may potentially be a ‘safe’ way of shifting the data, it clearly illustrates the estimated group probably has inaccurate data.
In terms of the headline results MSCT identified 253 (97.3%) of all aneurysmal SAH and 332 (95.7%) of all SAH. The 15 patients not identified with MSCT, of whom 7 (47%) were aneurysmal SAH, were diagnosed with SAH using a combination of Lumbar Puncture and/or MRI. These are impressive figures, but I’m not sure that it really answers the question that I want answering.
At 24 hours the authors report a sensitivity of CT scan of 99.3%.
So should we change to a 24 hour time point with CT investigations?
In many ways we are interested more in the time analysis and especially for those patients who had their CT >6 hours from symptom onset. This is the point when my reading/presentation of the data start to differ from the authors.
It is reassuring that if a CT was taken within 6 hours, the sensitivity was 100%. However, this is almost certainly the expectation, because if our current investigation within 6 hours is CT and (routinely) nothing else, then how else would you diagnose it? This accounts for 224 patients in the cohort.
Beyond 6 hours it looks as though we are left with 108 patients of whom 15 patients had a negative CT. That means that the sensitivity for this group is closer to 87% which is (in my opinion) too low to assure me of a diagnostic exclusion. Admittedly just two of the missed patients occurred in the <24 hour window, but these are small numbers and therefore wide confidence intervals.
The reason I look at this differently is because the authors have described cumulative diagnostic performance as opposed to within specific time boundaries. So when they report a high sensitivity at 24 hours it includes the high diagnostic performance at less than 6 hours. Is that right or wrong? It depends on your perspective, but I think my approach makes more sense for the following reasons.
- Patients choose when to present.
- We have good evidence on practice within 6 hours, so our pragmatic question is in how we change practice beyond that.
In addition the methodology here means that we should be cautious about changing practice on the findings. The main concerns are as follows.
- Retrospective designs almost always contain known and/or hidden bias.
- We don’t have accurate times for symptom onset.
- The evaluation is in patients with known SAH and that’s not the population we see in the ED.
- The confidence intervals around the performance of CT beyond 6 hours are wide. There just aren’t enough patients here in those groups of interest to give us a precise estimate.
- There is a pathophysiological basis to treat delayed presentation patients differently than those that present early (since the body clears blood from the subarachnoid space).
- There is the possibility for missed diagnoses as although the authors sought for missed deaths through coroners records, it is possible that there were missed events that did not lead to mortality. Any such cases would reduce sensitivity estimates. The authors could have searched for SAH diagnoses rather than deaths.
- The CT report is both the index test and in many cases the gold standard. This produces some circularity in the estimate. An independent assessment of the CT scans and/or panel confirmation of the diagnosis based on a wider assessment of the patient journey and wider investigations would be an alternative approach.
I had high hopes for this paper from the title. However, the concerns above mean that it’s not definitive and not yet enough for me to change practice. Having said that, this paper does ask a really important question and may lead to further research that may allow us to get greater precision around the estimates. It’s always great to see more information on this topic, and also useful to apply a critical appraisal lens on studies looking at diagnostic. I believe that the UK TERN (Trainees in Emergency Medicine Research Network) will have a prospective study up (the SHED study) and running soon that may give us more data in this area.
The bottom line is that CT scanners and the radiologists who read them may well be getting more accurate, but this paper cannot conclusively tell us whether that is the case.
Gray C, Foex B. BET 2: does a normal CT scan within 6 h rule out subarachnoid haemorrhage? https://emj-bmj-com.manchester.idm.oclc.org/content/32/11/898
Roberts T, Hirst R, Hulme W, Horner D. External validation of a clinical decision rule and neuroimaging rule-out strategy for exclusion of subarachnoid haemorrhage in the emergency department: A prospective observational cohort study. https://www.medrxiv.org/content/10.1101/2021.05.26.21257212v1
Vincent A et al. Sensitivity of modern multislice CT for subarachnoid haemorrhage at incremental timepoints after headache onset: a 10-year analysis. https://pubmed.ncbi.nlm.nih.gov/34819306/
Sub-arachnoid archives at St Emlyn’s https://www.stemlynsblog.org/tag/subarachnoid-haemorrhage/