Transcript
It’s great to be here and talk to you about this. It’s always difficult to follow both Wouter and Peter because they’re so good, but I think we all have individual differences in how we do things, and we’ll talk a little bit about why that might be.
Disclosures – I showed this QR code yesterday. All the papers I’m going to talk about are in the Google Drive folder that I share here. I’ll show it at the end also.
So, why might the three of us do things differently in a patient with orthostatic headache and a normal brain MRI? I think there are two main reasons. When I was thinking it through and putting together the talk, I came up with eight reasons, but these are the two main ones.
So, one is that we view the likelihood that a brain MRI is a false negative differently. And two is – in this circumstance of a negative brain MRI in a patient with an orthostatic headache – we’re inherently balancing the risk of treating unnecessarily a patient who is truly not leaking versus the risk of failing to treat a patient who is truly leaking and just has a false negative brain MRI.
So, the false negative rate – how often a study will say that something is negative or normal – is one minus the sensitivity of the study. Sensitivity of the study is, kind of as it sounds, the rate of being test positive when in fact you are true positive. And the sensitivity of a brain MRI is the ratio of test positive divided by all true cases of CSF leak that were tested.
And the test positive rate, i.e., what is a positive brain MRI, is the rate at which the MRI is interpreted by real-world radiologists with some kind of positive or affirming statement of a CSF leak. That’s what I think a normal brain MRI is – that’s the definition. It’s an MRI that’s interpreted by a real-world radiologist with a positive or affirming statement of a CSF leak. And we could define it further – maybe 50% of real-world radiologists.
And in many conditions, the sensitivity of any given test is known to be a function of disease severity. So, for instance, in COVID tests, whether you’re looking at the antigen test or you’re looking at the PCR test, they don’t have defined sensitivities for all cases of COVID. It’s known that the sensitivity is different at the beginning of the disease, when you’re producing less virus, and at other stages of the disease, and different in people who are vaccinated.
In fact, it’s very common in diagnostic tests for the test to, in fact, be more sensitive in cases of greater disease severity where the pathophysiology is more robust, and the disease can be picked up more easily. But what we’re interested in for brain MRI is not the sensitivity as a ratio of test positive cases over only the most severe cases, right? We know that when people have a large single-channel CSF-venous fistula, the likelihood of them having a positive brain MRI is really quite high. But we’re – where you’re really more interested in is, is it possible, and to what extent does it happen, that somebody has sufficient loss of CSF spinally that they get symptoms, but not sufficient CSF loss spinally to create pachymeningeal enhancement or venous distension or brain sagging, right? Like, to what degree is loss of CSF in the spine coupled to creating anatomic and physiologic changes in the brain that are of the kind that can be measured and picked up by MRI?
That is fundamentally an experimental question that someone could do – an enterprising neurosurgeon could do when they’re putting lumbar drains in people who don’t have leaks. They could actually measure, like, how much CSF do I have to drain per hour to get somebody to first start feeling a symptom of headache, or neck pain, or ringing in the ears, and if I image them then, what percentage of those people would be brain-positive? That would be a really interesting study, and it would show, kind of as an experiment, what we need to do to detect if everybody, when they first start having some symptoms from losing CSF, have positive brain. But we don’t have direct experimental evidence.
What we have is clinical studies that make me think that the sensitivity of brain MRI for a symptomatic CSF leak is low, and I’ll go through what I think are the four reasons that make me think that. One has been alluded to earlier today, which is that after people have had a known, unintended large-bore dural puncture, the rate of having new chronic headaches is high. The first paper I saw about this was by Pam Flood, who looked at 40 cases of unintended dural puncture and 40 cases of people who had epidurals for their labor and delivery at the same time. You can see here that among those who had a known dural puncture, the rate of having new chronic headaches a year to two years later was 28%. If they didn’t have a known dural puncture, it was only 5%.
One of the things that would help support the idea that this was not central sensitization but actually was related to a leak – is that if they were treated with a blood patch, the likelihood that they in fact went on to have chronic headaches is substantially less. It’s hard to explain that association of a patch with long-term headache if the long-term headache isn’t related to a leak.
That’s not the only study that has found that with these unintended dural punctures, we have long-term rates of headache, which conceptually could not – could possibly not – be due to a leak. But again, patching seems to prevent it. But nonetheless, if you look at the four studies, when they imaged people by brain MRI afterwards, the rate of actually finding anything positive on brain MRI was less than 10% for all the studies. The 100% is the fantasy.
I think that this, to me, suggests that these patients are the true positives. They had a known dural puncture. They have persistent headaches that started after that known dural puncture. Epidural patching appears to reduce the rate of that happening. 30% of those patients, when looked at five years later, still report that the headache is, in fact, postural, and many of those patients get better when, in fact, they’re operated on or they have patched.
So, I think the most likely, most reasonable explanation of those studies – and they’ve really influenced my thinking – the most reasonable explanation of those studies is that it is totally possible to have a small puncture of the dura that is absolutely sufficient to cause chronic symptoms that just is imaging negative with our current imaging techniques.
And if it’s true that when we cause such an injury, that it’s totally possible to have long-term CSF loss causing chronic symptoms, why should that not be possible from natural or spontaneous tears too? I don’t see any compelling physiologic reason for believing that is not so. And it seems a parsimonious explanation for the observation we have – that, in fact, many people with orthostatic headaches are imaging negative.
This was a hole that was found at surgery in the dura after a lumbar drain had been placed in a patient to help manage her when this physician had had a venous sinus thrombosis and a venous infarct. And she survived her venous infarct but had a year’s worth of orthostatic headaches that was worked up at an outside institution. Her brain MRI looked like this with her orthostatic headaches.
I’m going to show you some additional images. You can see here that I think most neuroradiologists would look at that and say, “Well, that pituitary is a little big,” which they did, in fact, comment on in her MRI reports at the outside institution. But there’s no pachymeningeal enhancement, and her venous sinuses are just fine. And in fact, she was seen at a major institution, and while they said, you know, her pituitary is kind of generous, this was not read with any affirming or positive statement about CSF leak at all.
If you can have a brain MRI that’s just normal when there’s a big hole like that, and just so you know, her spine imaging was also read as normal, and I think would be read as normal at Stanford too. And I think when you know there’s a hole, you would look at this little area here and say maybe you’re losing some dural margin there, and her epidural fat is in fact a little brighter than it should be. I pay a ton of attention to this kind of thing now in part because of experiences of finding patients like this, and I think Dr. Callen pays a lot of attention to these small things now, and you can see there’s no hint of a dural bleb there.
So these are what I think are the true positives that are the most appropriate way to assess what MRI sensitivity is in the real world. As I said, I just don’t think I can think of a compelling reason that would cause these leaks to be fundamentally different than a leak caused by a bone spur.
And I’ll just say that if I hadn’t patched this patient who had, by the way, had bilateral DSMs and empiric thoracic patches at the outside institution, if I hadn’t been aware of that literature about chronic post-puncture leaks, I wouldn’t have thought, well, I should really patch the lumbar drain site. This was the only thing that ever led to the discovery of the preventable and fixable cause of this woman’s disability – this teeny little bit of intravasation seen during the patch, which was enough to convince our neurosurgeons to go in and take a look. The brain MRI – normal; the spine MRI – normal; nothing was going to convince our surgeons to do anything for this woman other than that one little finding. And I’ll tell you, she got better when I fibrin glue patched her there; she got better for 24 hours, and that’s it. There was no compelling six-month “I’m better,” you know – it was like a nothing. Her response to patching was zero.
So, reason number one for why I think MRI brain sensitivity is low is because it’s so low in the patients with persistent post-puncture headaches.
Reason number two is because most patients with an orthostatic headache have a normal brain MRI, but at least 10 to 15% of those patients will be found to have a CSF-venous fistula. So, Hosoya and colleagues in Japan prospectively gathered 100 orthostatic headache patients. They evaluated all of them for signs of a leak – 89 brain MRIs, 69 totally normal, only 20 brain MRIs read as positive. This tells you nothing about the sensitivity, but what it does tell you is that only a minority of clinically suspected cases have a brain MRI with SEEPS positivity.
This is exactly what we found when we went back and looked at our data – of the people presenting with orthostatic headaches, only about 20% have a positive brain MRI. Most people who are afflicted with orthostatic headaches have normal brain MRIs, but as Wouter already talked about, as Dr. Schievink already talked about, he now has two separate independent cohorts that shows when you look in these patients with a high clinical suspicion that, in fact, you’re going to find in either his first cohort 10% or his second cohort 15% of patients have a defined, findable, fixable cause, but their brain MRI is normal.
Reason number three that makes me think brain MRI’s sensitivity may be low is evidence that MRI-negative patients improve following epidural patching and that improvement is linked specifically to orthostatic features before patching. Peter mentioned Dr. Choi’s study from Korea, and there’s also this one by Davies suggesting that many people who are imaging negative benefit from patching.
Our data, that Peter talked about – and I guess I’ll spend just one minute on it – the first is among patients with what we have here on the Y-axis, is improvement in the PROMIS physical global health score. So, zero would represent no change. 10 points would represent an improvement of a full standard deviation of global physical health according to the US population, going from the 50th percentile up and improving a full standard deviation – that’s 10 points.
What you see here is a couple of things. Number one, when people laid down for an hour, when their headache severity went down to a zero or a one, almost all of those patients had significant improvement, and a bunch of them improved more than a full standard deviation in terms of global physical function and health.
In fact, if you look at all the green dots up here, these are all the people who improved a full – a full standard deviation in physical health according to the standard deviation of physical health in the US population as a whole. Not a small change – a big change. That represents roughly a fifth of the people who we treated.
In contrast, the number of people who decreased a full standard deviation is one. We improved by more than a full standard deviation compared to reducing by a full standard deviation, large effects by a ratio of 13 to 1.
Now, there’s a bunch of people who had really small improvements in PROMIS physical health, and maybe they really shouldn’t be patched – maybe that is in the eye of the beholder. The problem is, before I patched them, I couldn’t tell who was going to be down here and who was going to be way up here. Right? So, since they’re improving at a much greater rate than they’re not improving or getting worse, and especially since we’re starting to learn we can predict some of that by how orthostatic their headache resolution is when they stay flat, I think this is worth doing. I also think the fact that orthostatic symptoms in patients who are imaging negative predicts long-term improvement suggests we’re missing something important in our imaging.
If we’re not missing something important in our imaging, why are these patients getting better in a way predicted by orthostatic features? So, that’s reason number three for making me think maybe the real-world sensitivity of brain MRI is low.
Reason number four is kind of exploratory and preliminary, which is that a population of MRI-negative patients with symptoms of CSF leak have Bern scores that skew higher than normal healthy controls and patients with post-traumatic headache. Henrik Winther Schytz – who is the co-director of the Masters of Headache Disorders at the University of Copenhagen and at the Danish Headache Center – came to the Bern meeting a couple of years ago. Being interested in CSF leaks, he decided to go back and look at research he had already done. He had already published research on a cohort of about 100 patients with post-traumatic headache, characterizing them extensively in terms of phenotype and symptomatology, and getting brain MRIs on all of them, and at the same time doing the same for a set of age-matched healthy controls.
He went back and looked at them and is currently submitting this data to a journal to compare whether, in fact, the Bern scores in patients with post-traumatic headache were higher than patients in the normal healthy controls. I had been thinking maybe post-traumatic headache patients who have chronic headache and sometimes have dizziness – maybe they would be enriched for patients who had leaks that were missed, and he thought that was a reasonable hypothesis. He went back, did the work to actually look at that, and his team found the following:
So, in the healthy controls, what they find is that, first of all, in healthy controls, nobody has venous sinus distension. As Peter mentioned, when he sees it, he thinks absolutely that that’s a positive finding. Even though the brain MRI was read as negative, this would support him. It is not a normal finding. In fact, Henrik went back and reached out to Tomas Dobrocky and got information from the original Bern controls, and none of them had venous distension either. Venous distension is never seen in healthy controls in that set of 150 patients. It’s always abnormal.
On the other hand, in the healthy controls as well as the post-traumatic headache patients, the rate of having things like suprasellar cistern effacement, prepontine cistern effacement, or mamillopontine cistern effacement was totally the same. There was no evidence that, in fact, post-traumatic headache patients had any evidence of skewing higher in terms of their Bern scores – no evidence that any of them are leaking. I think this actually – we have the Bern scores here on the left: post-traumatic headache participants, healthy controls. Again, this is Henrik’s data. I think this provides good data about what is a normal MRI. In healthy controls, only 6% of people have a Bern score of four or higher. That’s kind of like what we talk about in medical science as a P value. Basically, it is the likelihood, under the null hypothesis of no leak, that you’re going to find this result or something more extreme in the data. Under the null hypothesis, a Bern score of four has a P value of 0.06, and we should understand it in that context so that when we see somebody who has orthostatic headaches and a Bern score of four, there’s only a 6% chance that you were going to see that finding if they didn’t have a leak, which may help us kind of interpret that data better.
In the original Bern cohort, only 7% of normal healthy controls had a Bern score of three or greater. So now when Henrik did this, it got me starting to think just this last week about what if I went back and looked at the Bern scores from the patients in our imaging-negative cohort – of people who had orthostatic headache and other symptoms of CSF leak – and this is what that looks like. So again, you’ve got the post-traumatic headache patients here, the healthy controls here, and then these are the patients from my study – your imaging-negative patients with symptoms of a CSF leak.
And this is really preliminary, because when I first reached out to my statistician, I asked her to see if this was statistically significant, and she did a test of a difference in proportions. But the right way to really look at this is to do a whole Chi-square of this distribution versus this distribution. So we’re going back and doing that, but what you see is the difference in the number of patients who have really low Bern scores will probably turn out to be significant. And in fact, the likelihood of having a Bern score of four or greater, if you were presenting with symptoms of orthostatic headache and symptoms of a CSF leak, you were really very statistically significantly more likely to have a higher Bern score in that group.
In other words, the distribution of Bern scores is skewing higher in these patients. So, I can’t say – maybe I’m wrong, maybe I’m patching some people who, in fact, really aren’t leaking. But in this population who are imaging-negative and ICHD-3 negative, this is evidence that, in fact, they have brain-related changes that, even though called normal, are skewing in a distribution way towards positive findings that we associate with CSF leak. And this is some evidence that the sensitivity of brain MRI as read by normal interpreting radiologists is lower than what we would really want.
So, we talked about that. Has our community misunderstood the sensitivity of imaging testing before? I think it’s pretty obvious that we have. This was this kind of well-known paper out of the Mayo Clinic from 2013 where they looked at the sensitivity of MRI of the spine compared to CT myelography in patients with orthostatic headache with CSF leak. They had concluded that, using CT myelography as the gold standard, the sensitivity of spine MRI for detecting a CSF leak – not a SLEC, for detecting a CSF leak – was 91.7%. And I remember ordering CT myelograms at Stanford early on and being told, “Why are you doing that? We did an MRI. We know there’s no leak. The MRI was negative.” I don’t think there’s anyone in the room that thinks a conventional CT myelogram should be the gold standard. And I don’t think there’s anyone in the room who thinks that an MRI is 91% sensitive for CSF leak.
We now know that 50% of leaks are CSF-venous fistulas. We know that all of them were missed by all of the radiologists and investigators in this study, and they were calling 91% sensitivity when what they really should have been saying is, “Look – MRI and conventional CT myelogram are highly correlated. We don’t really know what it says about the underlying sensitivity.” And I think that we may, in fact, continue to be subject to misinterpreting the sensitivity of our imaging studies.
So, what’s the Stanford approach? We believe in doing a detailed – and not just detailed, but a structured assessment of clinical symptoms. We have a leak review of systems we go through with everyone. It’s in the Google Drive folder that I share. We try to measure a multidimensional assessment of symptom responsiveness to remaining flat, which is the 48-hour flat test, which we hope to publish on this year. We try to get the best spine MRI imaging possible using the sequences that Dr. Schievink’s group has put forward, and we’re finding very helpful. Our yield at Stanford, and the yield reported by Dr. Schievink of lateral decubitus imaging, is about 1 in 10 if brain MRI is normal and spine MRI has no perineural cyst.
So, we don’t advocate myelography for those patients. We tell them, “Look – I don’t know if you’re not leaking, but I know that the likelihood we’re going to find something here is less than 1 in 10.” And most patients, when you say to them not, “I don’t think you’re leaking, so I don’t think we should do a myelogram.” You say “look, I don’t know if you’re leaking or not, but I know that our imaging technique today at Stanford only gets positive findings 1 in 10. Maybe it’s not good enough of an imaging technique. We don’t really understand it, but we know that the yield of subjecting you to two new punctures is going to be less than 1 in 10.” And with that, I think we’ve gotten away from puncturing people who aren’t likely to benefit from it.
But that doesn’t mean we become convinced that they’re not leaking at all. If they have compelling symptoms, we discuss them at an interdisciplinary conference, and if multiple expert physicians think that it’s reasonable, we do offer them a trial of epidural patching, because we do think that for the people who benefit, it can be clinically meaningful and make a big difference in their life. And this isn’t 30 days out or 60 days out. This is over a year out. And a lot of the patients who are down here are people coming back for treatment who were much better for three or four or six months, and then their symptoms recurred.
So, I guess what I would say is, and the last part is, that we’re working hard to try and get the same kind of coronal images to look at optic nerve sheath diameter that Dr. Schievink has proven are correlated, not necessarily with whether somebody’s leaking or not, but when it’s futile to do further invasive imaging. And I think that’s the path forward and why that paper is so nice – not because it tells us who’s leaking, but it tells us when to stop doing something invasive.
And so, what I hope the future will reveal for us is circumstances where we can look at our own data that will tell us when we should stop patching or when we should stop doing other invasive things, so that we know, here’s the group where there’s really a chance of benefit. Here, we don’t have to subject people to risk because we know that the yield is, again, like people with very large optic nerve sheath diameters, the yield is low.
So, thank you very much. The papers that I talked about are in the QR code – as is the structured review of symptoms. Thank you.