Graphical Abstract
Abstract
BACKGROUND AND PURPOSE: Spontaneous intracranial hypotension (SIH) due to CSF-venous fistula (CVF) is increasingly recognized as a secondary cause of headaches, with symptoms often overlapping with primary headache syndromes such as migraine. While brain MRI studies have focused on features indicative of SIH, findings that support an alternate headache etiology, such as the bifrontal white matter hyperintensities (WMH) often seen in migraines, have not been explored in this context. This study assesses the following: 1) the quantity and distribution of WMH, and 2) the presenting clinical features in patients with and without CVF found on dynamic decubitus CT myelography (dCTM).
MATERIALS AND METHODS: Seventy-two consecutive patients underwent clinical work-up for SIH due to suspected CVF, including preprocedural brain and spine MRI followed by dCTM. Brain imaging features were analyzed, including the Bern score, quantitative WMH burden, and WMH distribution. Demographics and clinical symptoms present at the time of presentation were recorded. Imaging features were compared between groups with and without CVF using parametric or nonparametric comparisons according to variable normality. Multivariate logistic regression explored the relationships among imaging features, clinical symptoms, and the presence of CVF.
RESULTS: The cohort included 40 patients with (CVF+) and 32 patients without (CVF–) CVFs, with no significant age or sex differences. Patients with CVF+ had significantly higher Bern scores and significantly fewer WMH. There were significant differences in the frequencies of WMH patterns between groups, with a migrainous pattern observed most frequently in patients with CVF–. Logistic regression combining the Bern score, WMH burden, and WMH pattern demonstrated a better fit for predicting CVF than using the Bern score or WMH features alone. Fourteen clinical symptoms showed the greatest differences between CVF+ and CVF– groups. Logistic regression demonstrated a positive association between CVF detection and a pressure/throbbing headache quality and negative associations for neck pain, facial pain, phonophobia, and anhedonia/depression.
CONCLUSIONS: These findings suggest a negative association between CVF detection, increased burden of WMH, and a migrainous WMH pattern. Symptom analysis describes distinct clinical phenotypes, challenging orthostatic headache as a defining characteristic. These results support a comprehensive assessment of imaging and clinical presentations in the work-up of suspected SIH.
ABBREVIATIONS:
- CVF
- CSF-venous fistula
- dCTM
- dynamic decubitus CT myelography
- dDSM
- dynamic decubitus digital subtraction myelography
- NPV
- negative predictive value
- PPV
- positive predictive value
- rpb
- point biserial correlation coefficient
- SIH
- spontaneous intracranial hypotension
- WMH
- white matter hyperintensities
SUMMARY
PREVIOUS LITERATURE:
The Bern score reflects the likelihood of detecting a spinal dural defect or CVF on subsequent myelography but is an imperfect measure. Imaging features suggestive of an alternate headache etiology, particularly those of a primary headache syndrome, have not been examined in the context of suspected SIH. Additionally, symptoms of patients with SIH have been reported but not compared with those of patients suspected of having SIH with ultimately negative myelography findings.
KEY FINDINGS:
Patients with CVF on dCTM have higher Bern scores and lower WMH burden, including a paucity of a migrainous pattern; patients without CVF on dCTM have lower Bern scores and greater WMH, particularly in a migraine-associated pattern. Symptom analysis demonstrates distinct clinical phenotypes but also overlapping features between the 2 groups.
KNOWLEDGE ADVANCEMENT:
Significant differences in the burden and patterns of WMH in patients with and without myelographically-localized CVF suggest a possible alternative headache etiology in those suspected of having SIH but in whom no CVF is found. Together, findings support a comprehensive review of imaging and clinical presentation in the work-up of suspected SIH.
In recent years, awareness of spontaneous intracranial hypotension (SIH) as a potential secondary cause of headache syndromes has greatly increased among radiologists, neurologists, and neurosurgeons.1 However, distinguishing SIH from primary headache disorders such as migraine on a clinical basis can be challenging, given the heterogeneous clinical presentations of SIH and the potential overlap with migraine symptomatology.2,3 For instance, an orthostatic headache in which pain is elicited from the patient being in an upright position and relieved when lying flat has been considered the cardinal clinical feature of SIH.4 However, it has been reported that up to 92% of patients with migraine lie down as a main strategy to improve their headaches.5 Additionally, patients with proved CSF leaks have presented with atypical features, including the absence of headache, exacerbated pain from lying flat, or symptoms classically associated with migraines, such as frontal head pain, photophobia, and/or nausea.6⇓-8 Thus, additional specific diagnostic tools are needed to diagnose SIH.
Further complicating the diagnosis, up to 20% of patients with symptoms of intracranial hypotension present without classic findings of MRI of the brain.7,9 This is particularly challenging in the context of CSF-venous fistula (CVF), in which there is no dural defect resulting in epidural fluid but rather an aberrant communication between the subarachnoid space and adjacent venous structures, allowing CSF to directly enter the venous circulation and requiring invasive myelography for its detection.10⇓-12 Additionally, CVF requires definitive treatments beyond standard epidural blood patching, such as percutaneous fibrin occlusion, endovascular embolization, or surgical ligation/resection.13⇓-15 Techniques to detect CVF include dynamic decubitus digital subtraction myelography (dDSM) or dynamic decubitus CT myelography (dCTM).16 While these procedures are minimally invasive and generally considered safe, they are not without risks. For instance, the frequency of iatrogenic postdural puncture headache ranges from approximately 4% to 30%, depending on the spinal needle size and type.17,18 Additionally, to minimize motion and optimize breath-holding during imaging, dDSM is often performed with the patient under general anesthesia,19 which carries inherent risks.20 Both dDSM and dCTM can expose the patient to significant ionizing radiation due to the acquisition of multiple views and/or phases.7,19,21 Considering these risks, it is challenging to identify which patients are most appropriate for invasive myelographic testing, particularly if brain MRI does not demonstrate obvious findings of SIH.
To date, the assessment of imaging findings for suspected SIH has focused on features suggestive of the presence of a CSF leak. These features are frequently reported in terms of the Bern Score, a probabilistic scoring system that reflects the likelihood of detecting a spinal dural defect or CVF on subsequent myelography.22,23 However, in the context of SIH, little attention has been paid to MRI findings that may suggest an alternate headache etiology, particularly those related to a primary headache syndrome. One such potential feature is T2 FLAIR bifrontal white matter hypointensities (WMH), the presence of which has been extensively described in patients with migraine.24⇓-26 While incompletely understood, the pathophysiology underlying these lesions is proposed to include subclinical microvascular ischemia, inflammatory processes, or remodeling of neuronal microstructures.27,28 WMH demonstrate distinct distribution patterns: WMH related to chronic microvascular ischemic disease involving the subcortical, deep, and periventricular white matter without lobar predilection.24 WMH related to migraines tend to have a predilection for the bifrontal white matter, either alone or out of proportion to the overall burden of WMH.25
The purpose of this study was to assess the quantity and distribution pattern of WMH in patients with suspected SIH who underwent subsequent dCTM. We hypothesized that findings of migrainous WMH would be inversely associated with the likelihood of CVF detection. Secondarily, we sought to explore whether there were differences in the presenting clinical symptoms of patients with headache in whom a CVF was found compared with those in whom it was not.
MATERIALS AND METHODS
Subjects, Image and Clinical Analyses
This retrospective, cross-sectional cohort study was approved by the institutional review board, and informed consent was waived due to the retrospective nature of the study design. The study population consisted of consecutively evaluated patients referred to our institution for clinical suspicion of SIH between February 1, 2021, and December 31, 2023. Patients who did not undergo a pre-dCTM brain MRI, had imaging insufficient for inclusion, had evidence of extradural fluid on preprocedural spine MRI, did not undergo dCTM, had a study with negative findings with only 1 laterality studied on dCTM, or had demyelinating disorders were excluded. No included patients demonstrated imaging features concerning for leukodystrophies. All included patients underwent a preprocedural MRI of the brain at 1.5T or 3T. Because patients were referred from diverse geographic locations and clinical practices, brain scans encompassed a wide variety of imaging protocols; minimal inclusion requirements were a 2D or 3D sagittal sequence and postcontrast T1 imaging allowing determination of the Bern score22 and a 2D or 3D FLAIR sequence for assessment of WMH. Bilateral dCTM using a standard energy-integrating detector CT scanner was performed as previously described.29,30 A Bern score was assigned to each patient by 1 of 4 board-certified neuroradiologists with specialization in CSF leaks using the previously described quantitative and qualitative criteria22 from preprocedural brain MRIs.
FLAIR sequences were analyzed by a board-certified, subspecialty-trained neuroradiologist to manually quantify the total number of hyperintensities in the supratentorial white matter, and the percentage of lesions occurring in the frontal lobes was calculated. Distribution patterns of WMH were independently scored by 2 board-certified, subspecialty-trained neuroradiologists, with disagreements in pattern classification resolved by consensus. The distributions of WMH were classified into the following patterns: 1) none to minimal, 2) migrainous, or 3) microangiopathic (Fig 1). A minimal pattern was characterized by no discrete WMH or the presence of <10 punctate lesions (<3 mm) without lobar predominance and that could include thin WMH caps at the frontal and occipital horns.31 The migrainous pattern was defined here by small (<3 mm) WMH with a clear frontal lobe predominance and without significant confluence.25 The diffuse microangiopathic pattern was defined by the presence of larger lesions (>3 mm), overlapping lesions or ill-defined borders, broad involvement of the ventricular margins, and diffuse distribution without lobar predilection.24 Patients were stratified according to the presence or absence of CVFs on dCTM. Patient demographics were extracted from the electronic medical record. Additionally, patient records were assessed for the presence or absence of 34 symptoms and clinical features occurring at the time of presentation as documented in clinical progress notes at our institution. The clinical symptoms included have been described in patients with SIH in multiple prior reports.3,6⇓-8,32
Illustrative examples of the classifications of FLAIR WMH. A, Patient with minimal punctate FLAIR hyperintensities without frontal predilection (arrows). B1, Patient with frontal-predominant migrainous-type WHM involving the subcortical and periventricular white matter (arrows; detailed inset B2), characteristically seen with migraines. C, Microangiopathic-type WMH with diffuse supratentorial lesions.
Statistics
Continuous variables were assessed for normality using the Kolmogorov-Smirnov test. Normally-distributed variables were compared via standardized Welch 2-sample t tests; categoric and non-normal variables were compared using 2-sample Mann-Whitney-Wilcoxon tests. Differences in pattern frequencies of WMH between the CVF+ and CVF– groups were assessed by the Pearson χ2 test. The Kruskal-Wallis test was used to examine differences in CVF presence or absence by the pattern of WMH, followed by post hoc paired Mann-Whitney-Wilcoxon comparisons. The Bonferroni adjustment was applied to account for multiple simultaneous comparisons. The point biserial correlation coefficient (rpb) was calculated to explore the association between the Bern score and CVF positivity. Sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were calculated in the standard fashion for features predictive of CVF absence on dCTM. Multivariate logistic regression analyses were performed to explore the relationships among the Bern score, WMH imaging features, clinical symptoms, and the presence or absence of CVF, followed by ANOVA analysis of the model deviance from null and goodness-of-fit assessment via the McFadden likelihood ratio index. All statistical tests were conducted in R Version 4.4.0 (http://www.r-project.org) and RStudio Version 2024.04.2 + 764 (http://rstudio.org/download/desktop).
RESULTS
The cohort consisted of 40 patients with myelographically-proved CVF (CVF+) and 32 patients with no CVF (CVF–) on dCTM. Patient demographics and imaging findings are summarized in Table 1. Groups did not differ by age (mean, range; CVF+: 61.6, 34–83; CVF–: 58.2, 22–71) or sex. Although there was a higher percentage of female patients in the CVF– cohort (72%) versus the CVF+ group (58%), this difference did not meet statistical significance (P = .21).
Patient demographics and imaging characteristics
Patients with CVF+ had significantly higher Bern scores compared with patients with CVF– (median, 6 versus 1; P < .001), and there was a strong correlation between the Bern score and CVF positivity (rpb = −0.82). Patients with CVF had significantly fewer total WMH (mean, 7.9 versus 20.7; P = .008) and fewer frontal WMH (mean, 5.4 versus 14.6; P = .002). Differences in the percentage of WMH occurring in the frontal lobes did not reach statistical significance between the CVF+ and CVF– groups using the adjusted α (median, 53.3% versus 70.6%; P = .017). There was a significant difference in pattern distribution of WMH between the CVF+ and CVF– groups (Table 1 and Fig 2A; χ2 = 10.54, P = .005). There was an overall significant difference in the frequency of myelographically-localized CVF among different patterns of WMH (Fig 2B; P = .006), with post hoc analyses demonstrating a significant difference in CVF positivity between the minimal and migrainous WMH classes (68.8% versus 25.0%; P = .002) but not between the minimal and microangiopathic classes (68.8% versus 65.0%; P = .79). A trend toward differences in CVF positivity between migrainous and microangiopathic classes was present that did not meet the adjusted α level for significance (25.0% versus 65.0%; P = .01).
A, Bar graph illustrating the percentage of patients with patterns of WMH, which differed significantly between those with and without a myelographically-proved CVF (P = .005). B, Bar graph illustrating the frequency of myelographically-localized CVF in patients according to the pattern of WMH. There is an overall significant difference in CVF positivity across patterns of WMH (P = .006), with post hoc analyses demonstrating a significant difference between the minimal and migrainous WMH classes (*P = .002).
Sensitivity, specificity, PPV, and NPV were calculated for imaging features potentially predictive of CVF absence compared with the “true” presence or absence on dCTM, including the intermediate-probability Bern score (<5), low-probability Bern score (<3), the presence of a migrainous white matter pattern, or a combination of Bern score threshold and migrainous WMH as summarized in Table 2. The features of a low-probability Bern score with the presence of migrainous WMH resulted in the highest combined values for sensitivity, specificity, PPV, and NPV.
The combined features of a low-probability Bern score and the presence of migrainous WMH resulted in the highest sensitivity, specificity, PPV, and NPV
Figure 3 summarizes the occurrence of clinical features in patients with CVF+ and CVF–. Symptoms with the greatest differences in frequency between patients with myelographically-detected CVF and those without were the following: headache quality of pressure/throbbing (70% versus 47%), symptoms worse with the Valsalva maneuver (48% versus 17%), neck pain (35% versus 63%), fatigue (23% versus 43%), photophobia (20% versus 43.%), worse when lying flat (15% versus 3%), difficulty sleeping (15% versus 43%), radicular symptoms (15% versus 27%), anhedonia/depression (13% versus 37%), facial pain (8% versus 60%), lightheadedness (8% versus 17%), ear pain or fullness (5% versus 23%), phonophobia (5% versus 33%), and scalp or facial hyperesthesia (5% versus 17%). Frontal head pain (30% versus 40%) and nausea (30% versus 43%) were reported slightly more frequently in the CVF– group, while periorbital pain (30% versus 20%) occurred slightly more frequently in the CVF+ group. Other clinical features including dizziness, relief when flat, tinnitus, occipital pain, temporal pain, headache quality sharp/stabbing, exertional headache, exacerbation with head movement or position changes, memory issues, holocephalic pain, hearing loss, blurry vision, syncope, and preference to sleep at an angle were all reported at similar frequencies (within 10%) for both groups.
Bar graph illustrating the distribution of clinical symptoms and characteristics between patients with CVF+ and CVF– at presentation. Features identified via multivariate logistic regression to have a significant effect on the presence of CVF on dynamic CT myelography are denoted by positive (+) or negative (–) associations.
Multivariate logistic regressions were performed to predict the presence of CVF on myelography, given the Bern score, WMH imaging features, and clinical symptoms. The clinical analysis included the 14 symptoms with the greatest difference in reported frequencies between CVF+ and CVF– groups. The resultant model had a McFadden likelihood ratio index (pseudo-R2 for goodness of fit) of 0.60. Headache quality of pressure/throbbing (P < .05) had a significant positive association with the probability of CVF detected on myelography, while the variables with a significant negative association included neck pain (P < .01), facial pain (P < .01), phonophobia (P < .05), and anhedonia/depression (P < .05; Fig 3). For imaging factors, the model assessing the combined Bern score, total WMH, frontal lobe WMH, and WMH pattern on subsequent CVF detection demonstrated a stronger goodness of fit (McFadden pseudo-R2 = 0.766) than the Bern score or features of WMH alone (McFadden pseudo-R2 = 0.689 and pseudo-R2 = 0.263, respectively). In the combined model, the Bern score had the most significant association with the probability of a positive CVF (P < .001), while there was a non-statistically significant negative trend for patterns of WMH (P = .067).
DISCUSSION
In this retrospective cohort study of 72 patients, we found significantly different WMH burdens and patterns of distribution between patients with and without myelographically-proved CVF. On the assessment of patients’ presenting symptoms, distinct clinical phenotypes emerged between the 2 groups, with 14 features reported at disparate frequencies. The Bern score remained a robust predictor of probability to localize a CVF on myelography, which is further strengthened by consideration of the pattern of WMH.
WMH have been reported present on MRI in up to 73% of patients with migraine, with approximately 90% of lesions occurring in the frontal lobes.24 Several studies have examined possible associations between burdens of WMH and migraine clinical severity, progression, subtypes, and symptomatology.24⇓-26 Conversely, a paucity of WMH is observed in patients without migraine in the absence of vascular risk factors.24,26 In our cohort, a migraine-type pattern of WMH was present at a nearly 4-fold greater frequency in the CVF– group versus the CVF+ group. Given the results of the current study, we propose that burden and distribution of FLAIR WMH be considered during patient evaluation for SIH, particularly in the setting of a low-probability Bern score, because it may implicate migraine as a plausible cause of the patient’s headaches. Most important, we recognize that a migrainous pattern of WMH does not exclude a concomitant CVF and that migraine can co-occur with SIH; thus, further diagnostic procedures should not be withheld on the basis of this feature alone.
Microangiopathic WMH are largely related to age and vascular risk factors.24 While they are associated with an increased risk for vascular events such as microhemorrhage or ischemic infarct, they do not appear correlated with headaches when diffuse in distribution.24,26 Our cohort demonstrated similar frequencies of microangiopathic WMH between the CVF+ and CVF– groups, likely related to a mean patient age of approximately 60 years at the time of presentation. Notably, there was no significant difference in the rate of CVF detection on myelography between the minimal and microangiopathic WMH groups.
An examination of symptom data demonstrated distinct clinical phenotypes between patients with CVF found on myelography and those without CVF. Specifically, patients with CVF+ reported a higher occurrence of a pressure-type or throbbing headache, symptoms worsening with the Valsalva maneuver, and symptoms worse when lying flat, while a greater number of patients without CVF reported neck pain, fatigue, photophobia, phonophobia, difficulty sleeping, fatigue, depression, radicular symptoms, lightheadedness, ear pain or fullness, facial pain, and facial and scalp hyperesthesia. Symptom exacerbation with the Valsalva maneuver appears to be a feature relatively specific to SIH due to CVF, previously reported to occur in up to 88% of patients.33 Worsening headache with the Valsalva maneuver is also a characteristic feature of Chiari I morphology,34 thought to be due to reduced intracranial CSF volume and altered flow dynamics.35 Many symptoms reported more frequently in patients without CVF include characteristic migraine features such as photophobia, phonophobia, and facial/scalp hyperesthesia,36 aligning with the higher occurrence of a migrainous pattern of WMH in this group.
The clinical features examined in the current study are also notable for their points of convergence, with strikingly similar reports of relief when lying flat and occipital head pain, once thought to be the defining symptoms of SIH.3,4 Other symptoms occurring at approximately similar rates include dizziness, tinnitus, location of head pain, and cognitive difficulties, among others. Therefore, this study suggests that the clinical features previously considered indicative of SIH could be nonpathognomonic in the context of CVF. Our prior work has shown that patients with myelographically-proved CVF cluster into distinct typical and atypical symptomatologies, including those without headache and exacerbation of symptoms when lying flat.6
The Bern score is a robust predictor of detection of CSF leak on myelography, whether due to a dural defect22 or CVF,11,23 and this prediction was replicated in the current study with a strong correlation between the Bern score and CVF positivity. The Bern score is far from a perfect metric: it has been reported that approximately 10%–20% of patients with reportedly negative findings on brain imaging were found to have leaks on dynamic myelography, in the setting of both dural defects12 and CVF.9 Additionally, the MRI findings comprising the Bern score can resolve in the setting of chronic SIH37 and do not correlate with clinical symptom severity.38 Nonetheless, the Bern score remains a critical component in the assessment of possible SIH; however, in recognition of the above limitations, other features of a patient’s brain MRI, particularly the burden and distribution of WMH along with the clinical presentation, should be considered in an integrated fashion.
Our study has several limitations, principally its retrospective design, modest sample size, and single-institution experience. The frequencies of clinical features and symptoms we describe differ from those previously reported in SIH,3,7,32 likely due to differences in patient populations, in particular the inclusion of patients with leaks due to dural defects that were not studied here. Collection of symptom data also relied on nonstandardized provider documentation versus validated metrics or formal patient-reported outcome measures.2,8 Additionally, many of the clinical features examined are also frequently present in elevated intracranial pressures, particularly when long-standing.39 Analyses of the imaging findings of intracranial hypertension or elevated opening pressures at the time of myelography could further refine overlapping clinical presentations but are beyond the scope of the current study and were not included. However, intracranial hypertension should remain high on the clinician’s differential when approaching the diagnostic work-up of patients with complicated headache phenotypes, including position- and pressure-dependent features.
Limitations also exist regarding image analysis. Variability in preprocedural MRI brain protocols, specifically the inclusion of 2D FLAIR sequences, may have limited the detection of very small WMH. Additionally, although the number and pattern of white matter lesions were assessed via dedicated review of only FLAIR sequences, complete rater blinding to the likelihood of a CVF was not possible, given the presence of associated brain findings, particularly with higher Bern scores. Due to the time-intensive nature of manual quantification, total analysis of WMH could be completed only by a single rater.
While patients with known demyelinating disorders were excluded, WMH can arise from a variety of vascular, inflammatory, infectious, traumatic, and toxic/metabolic entities.40 These expanded etiologies were not assessed or controlled for in-depth; thus, patients with possible alternative causes of WMH may have been erroneously classified into migrainous or microangiopathic groups. Last, there are limitations surrounding the myelographic technique. CT myelography for the detection of CVF was performed on a standard energy-integrating detector CT scanner. Photon-counting CT is a technology that may increase the sensitivity for CVF detection, particularly with low-probability Bern scores.41,42 Furthermore, multiple different techniques exist for the detection of CVF, specifically dCTM, dDSM, or a combination thereof, including conebeam CT during dDSM.11,16,43,44 Various adjunctive techniques, such as CSF pressure augmentation or resisted inspiration, are also reported to influence the sensitivity of CVF localization.45⇓-47 The myelography methods, equipment, and adjunct techniques which provide the optimal diagnostic test remain debated areas of investigation. Thus, the myelography approach used in the current study represents a subset of possible conditions, and CVF occult on our dCTM technique remains a possibility in the CVF– group.
CONCLUSIONS
This retrospective cohort study of 72 patients demonstrates significant differences in the overall burden of WMH and frequencies of patterns of WMH in patients with and without myelographically-localized CVF, favoring a possible alternative headache etiology for patients suspected of having SIH in whom no CVF is found. Analysis of symptoms occurring among patients with and without CVF revealed diverging clinical phenotypes but also features occurring in similar frequencies between the 2 populations, most notably relief when lying flat and occipital head pain. Last, the Bern score is reinforced as a strong predictor of localizing CVF in this cohort. Taken together, these findings provide further support for a comprehensive review of a patient’s imaging and clinical presentations in the work-up of suspected SIH.
Footnotes
Disclosure forms provided by the authors are available with the full text and PDF of this article at www.ajnr.org.
References
- Received August 20, 2024.
- Accepted after revision October 17, 2024.
- © 2025 by American Journal of Neuroradiology