Subarachnoid Haemorrhage or: How I Learned to Stop Worrying and Love Lumbar Puncture.


Like a lot of people, when I read this article in the BMJ that claimed a 100% sensitivity of CT for SAH within 6 hours of onset of headache, I was pretty excited. Finally we don’t have to do the LP anymore! I posted the paper on this blog here, without much comment, hoping smarter people than me would start to talk about it. Well they did, and a lot of people are excited about it. But good old Jerry Hoffman has taken the paper apart on the January 2012 Emergency Medicine Abstracts tape. His expert analysis is well worth a listen if you are a subscriber (if you are not you should be or you should make friends with someone who is). Below is a summary of all the reasons this paper doesn’t tell us that we don’t need to do LPs anymore (with thanks to Rick Bukata and Jerry Hoffman).

But first, why don’t we want to do the LP?

"The patient won’t like it. It will hurt, they’ll be frightened of it, I’ll need to take time convincing them to have it and then if they hate it and I’ve told them that it won’t be so bad they won’t trust me anymore."

"What if I get a traumatic tap? Or can’t get it at all? Do I just tell the patient that I’ve changed my mind and we don’t need the test after all? Do I get an alternative test? What alternative is there? If there’s an alternative why don’t we just do that instead of the LP?"

"It will take forever. I’ll need to get all sterile and position the patient. I’ll need to find a nurse to help or the patient is going to roll around and move. Just when I get sorted the AO phone will ring or there’ll be a trauma or the patient will decide they need to go to the toilet. Worse still, the resident will want to be supervised doing it and it will take three times as long, be even more likely to be traumatic and the patient will look at me with those eyes that say “You want to stick a needle where?! And you want to let that guy do it!? And he has never done one before? How come you don’t do it if you’ve done heaps of them?!” The patient might need some sedation while we do it because they are totally freaked by the whole thing so there goes another doctor for the duration of the procedure."

"The CT scanner is really good. It’s brand new and we had to put up with a week of jackhammers and workmen just through the wall from resus when they installed it. It can CT a whole body in 10 seconds and it has 2000 detectors and it has really nice pictures and I can look at them on my beautiful giant screen in the staff base magnified so big it could be the head of a wooly mammoth. It must work. It should work. Why doesn’t our technology work!???"

So, why doesn’t CT work for SAH?

No matter how good the scanners get, why can’t we get beyond about 95% sensitive?

Well it is not about the scanner. It is about the blood and the CSF and the difference between them. Take 1mL of blood and dilute it in 100mL of CSF and you have something that looks on CT pretty much like CSF. Take 5mL of blood and dilute it in 10mL of CSF and you have something that looks quite different. The smaller the bleed and the longer it has had to diffuse away from the site of haemorrhage the less the “haematocrit” of the CSF and the harder it is to detect on CT. Don’t forget it is small bleeds that we want to find. That is, the GCS 15 patient with a warning leak who we can save from the big one; the guy who comes in with stertorous breathing, GCS 5 and a dense hemiplegia has a big bleed we are rarely going to miss on CT and his prognosis is not great anyway.

So, to the paper.

The gist is this. 3000 odd people with essentially a thunderclap headache were looked at prospectively in a bunch of EDs in Canada. The patients and the settings were very similar to ours in Australia and the inclusion criteria were realistic so it was certainly going to be generalisable to our setting. Overall there was about a 7.7% incidence of SAH which is about what we would expect with this presentation. And overall the sensitivity of CT was 93% which is again what we would expect. It is a bit lower than what is usually quoted but it included people with some delayed presentations and we know that sensitivity of CT after 24 hours is hopeless (~80% by day 3 and 50% at a week). The CTs done within 6 hours however (953 with 121 cases of SAH) had 100% sensitivity. With such big numbers the confidence intervals were good. Done deal.

Wait a second. Here are the problems.

Workup bias.

Only 40% of the people with a negative CT actually got the gold standard test of an LP. In the rest they trusted to phone follow-up to catch any missed bleeds. So the first problem is the assumption that phone follow up (and a search of the coroner’s records for the province) is equal to LP as a diagnostic strategy for SAH.

Failed follow-up.

Only 78% of people who needed follow up (negative CT, no LP) got the 6 month telephone follow up. 50 patients (about 2.5%) were lost to all follow-up, 157 got a call at 1 month but not later, 200 (10%) were still having entries in their medical records at 6 months so were therefore alive, and 8 were “dead of other causes” (see below).


Anyone who has ever filled in a death certificate knows that what killed the patient and what is listed as COD are linked only by the tenuous thread of an intern’s brain. Also, we don’t know if they died of something else while waiting for the big SAH that was just around the corner. One death was “presumed cardiac” on the basis that it was a sudden death 32 days after his headache and 4 days after a presentation with chest pain; no autopsy was done.

Final read of the CT

The CT was called positive if the final radiology report said it was positive. 3 patients got a false negative read at the time and were called back the next day after the radiology report, fortunately before a re-bleed. What we don’t know is in how many cases the ED doc read the scan as negative, did the LP, found blood or xanthachromia, and then went back to the CT and said “oh! There it is. I wondered about that bit.”

100% specificity

CT had 100% specificity in all comers in this study. Fantastic! But a true positive was defined as blood on CT or blood /xanthachromia on LP. So it had to be 100% specific because it was its own gold standard for a positive result.

So, what we have is a big study that comfirms some things we knew already about SAH.

  • About 8% of thunderclap headaches have a SAH.
  • CT can pick up more than 90% of them in the first 24 hours.
  • CT probably gets better the closer to the onset of the symptoms you do it.
  • But, “CT is 100% sensitive for SAH if done within the first 6 hours of the symptoms”? No, I’m afraid we are not there yet. 

So what to do about LP avoidance.

Do more LPs. The more you do the easier it gets.  It is one of the best tests we have in medicine. Does my patient have meningitis? The LP tells me yes or no. Does my patient have SAH? The LP tells me yes or know.

When you find a bit of blood in the CSF wait a few minutes and let it wash out of the needle before you start collecting. Look closely and you’ll see the blood has settled to the bottom of the hub and clean CSF is flowing over the top of it. Now take a syringe and blunt 18G needle and aspirate that bloody CSF from the bottom of the hub (never suck on the spinal needle with a syringe unless you want to suck out a piece of nerve root!). Clean CSF can now flow out without picking up cells from your traumatic blood.

Have a plan for what to do with a failed or traumatic tap. If you work in a place where the radiology department is guarded by registrars who feel it is their job not to do tests, get your plan agreed on at an interdepartmental level. I would suggest CT, followed by LP if negative. If the LP fails or is indetirminate follwo up with a CT angiogram (or in a pregnant patient an MRI/MRA). Remember that theset ests don't look for a bleed, they look for an aneurysm and abuot 1/50 ro 1/200 people have one, the great majority of which will never bleed. But if you save these tests for where CT/LP fails you, at least you won't be creating too much disease where there is none.

Stop doing CT’s for headaches that never needed one in the first place. One of the reasons people feel like the LP after CT is pointless is that they have such a low yield but this is because they are sending a whole lot of people down that pathway who never had a thunderclap headache, “just in case”.  Take a good history. You need a CT for a headache in ED if it is thunderclap in onset, if it is associated with trauma, if the patient has a coagulation problem or immunosuppresion (especially HIV) or cancer, or if it is a chronic headache that might be a tumour.

Remember that although it feels like CT is the nice test and LP is the nasty one, no one ever got cancer from a lumbar puncture. If the headache was bad enough to justify irradiating a young person’s brain it was bad enough to justify LP.

What's with the title of the post?

You need to watch more old movies.