27 November 2012


I love, it, I really love it, when one of my strongly-held prejudices is borne out by actual, you know, facts and science.

For years, I have been arguing against the practice of performing a routine lumbar puncture (aka LP or spinal tap) in patients with the "worst headache of their life." This is done after a CT scan of the brain, typically, to look for a subarachnoid hemorrhage (SAH). The SAH is feared because in some cases they represent a leaking aneurysm which is at risk of bursting, often with devastating or lethal consequences.

The need to do the LP is one of the sacred cows of Emergency Medicine, written in stone, and has been for longer than I have been practicing. The reason is that SAH is dangerous, the CT scan is imperfectly sensitive for SAH whereas LP is highly sensitive (in fact, the "gold standard") and relatively easy and safe to do. This was perhaps more true long ago when the resolution of a CT scan was lower than it is with modern machines, but the dogma remains. There is, however, a huge variation in actual practice out there. Many docs seem to do very few LPs for headaches, and some seem to LP everybody. I performed a unscientific survey of ER docs on twitter and found that about half "always" still do the LP or are strongly inclined to do it routinely. Some were, in fact, required by their employer to do the LP!

Now my experience over the years was that the LP seemed to be a horrific waste of time. It was traumatic for the patient, consumed a lot of ER resources, and never ever showed anything. Twice -- twice! -- in a decade I spotted the unicorn and had a genuine negative CT followed by a positive LP. In both cases, the patient went on to have negative angiograms, so either the LP was a false positive or they were non-aneurysmal bleeds (which, as it happens, do not require treatment).

So I dug into our data. Pulling a year's worth of cases, I found that we had about 2,800 headaches present annually, slightly under 3% of all of our visits. 18 of those were subsequently diagnosed as SAH, for a prevalence of about 0.6% within all-comers of headaches. But that's not entirely fair, since over half of the headaches were either migraine type headaches or other chronic/recurrent headaches, and these folks are not those for whom we are highly suspicious of SAH. Of the headache patients, about 900 had CT scans ordered. While I might argue that not all of those truly needed a CT, and certainly not all would have gotten one in other countries, for this discussion it's reasonable to use that as an index of how many headache patients we had for whom our doctors were worried about SAH. So we have about a 2% prevalence of disease in our "acute" headache population (18/900). The traditional data was that CT was about 90% sensitive for SAH, so the negative predictive value of a CT is very good -- somewhere well north of 99% likelihood that the patient does not have SAH. Now you can play with the numbers and tighten it up a bit by more rigorously screening out headaches that are not "worst ever" and not sudden onset, but even if you get to a pretest prevalence of 10%, which would be quite high, the NPV is still very good, certainly better than we can rule out other serious diseases like PE or unstable angina.

But this was very rough math from a single practice with small numbers. So it is not exactly something I was able to endorse as a standard of care. Just contextual information I could offer a patient guiding them whether or not to accept the LP I was offering. Most declined, but some preferred the assurance that the gold standard test offers.

I've been quite pleased, though, to see more and more new and more rigorous data emerge on the topic. It seems, ever so slowly, the tide of opinion is turning against the routine LP.  First David Newman over at SMART EM did a great deep dive on the topic, showing that for LP, the Number Needed to Treat is somewhere around 500, which means that you'll do a lot of LPs to find a single SAH in a patient for whom it will make a difference. (Updated podcast on SAH here - worth listening to!) Then there was the Perry article in the BMJ last year which showed that the sensitivity of early CT is very very good, perhaps as high as 100% for SAH. Then there was this August 2012 article in the highly influential journal, Stroke, authored by none other than Dr. Jonathan Edlow:

Diagnosis of Subarachnoid Hemorrhage : Time to Change the Guidelines?[...] Given this analysis, we believe that practice should change. Neurologically intact patients who present with thunderclap headache and undergo CT scan within 6 hours of symptom onset no longer need an LP to exclude SAH if the CT scan is negative.
This is the same Dr Edlow who was lead author on the ACEP clinical policy, only 4 years ago, which did recommend routine LPs! (Link: PDF) The times, they are a-changin'!

So I feel comfortable claiming victory here. I was right all along and shame on you for ever doubting me.

(insert nuanced discussion here about shared decision-making with patients and the need to assess each patient as an individual.)


  1. (not in your industry at all, so please forgive my ignorance). If the LP is the gold standard for detecting SAH, why do the CT scan at all? Isn't the CT significantly more expensive, and a radiation risk to boot?

  2. Better late than never, I guess. Had to settle a lawsuit about this in the early 1990s. Not in an ED setting, in primary care, where doing an LP (even having the equipment to do an LP) is well nigh impossible. Even getting the CT urgently in those days was a stretch. HA gone by the time the studies concluded, normal neuro exam, blah blah. Felt awful about it.

  3. CT is still somewhat useful to rule out other etiologies, such as a tumor causing mass effect, and as a screen for hydrocephalus. That being said, there is immense debate on the utility of such studies, and a CT is in all likelihood less harmful than an LP and with fewer risks, and much less discomfort. Practitioner preference will vary with available equipment and skill.

  4. Yeah, half the reason I enjoy conferences is to feel smart 'cause it's something I've been saying all along. I take one small issue with your post, and that is that the #s analysis gives no weight to spectrum bias. The high 90's percentages that CT generates are all comers. Do we ever consider D/Cing a pt. with neuro deficits or who's in a coma? And those sentinel bleed guys, rare though they be, won't they ALWAYS be the guys you kinda want to D/C? Spectrum bias, the little bleeds make up the few % that we then let leave.
    Having said so much, if you're still reading, I do agree that CT (2nd or 3rd gen at least) within 6 hrs obviates CT - UNLESS they scare me. So young, non-vomiting, no neck pain folks will go home. 60 yo with meningismus and vomiting get an LP even if they feel great after the Reglan and the CT's negative. And I think my practice is way less critical of the Perry data than the ER critique of Hoffman/Bukata.

  5. What I never understood about the "worst headache" standard was that everyone whom has ever had a headache has to have a headache that is a "worst". So someone who doesn't routinely suffer from headaches comes in with a mild tension or sinus headache but its their "worst" then gets an LP? Makes no sense.

    I've had migraines since I was 12 and if someone asked when I was lying there waiting for the subq imitrex to kick in, I'd say hell, yes, worst of my life.

  6. Yes, yes, yes, I'm all for less testing. My blog is rabid on less testing. But - a more recent post of yours gives an malpractice example. You used the example as one of negligence, but that doc probably would have gotten sued even if he hadn't done the LP, probably for not doing the LP. You do what you think is best and, congratulations! you've been sued!

  7. you can code for a procedure, why not?
    Some places routine do LPs all HA.