Cranial dural arteriovenous fistulas (dAVFs) comprise a heterogeneous array of pathologies with unique anatomical features and require specific and individualized treatment, including transarterial and/or transvenous embolization, radiosurgery, and open surgical ligation. Despite successful fistula obliteration at the time of treatment, recurrence of dAVF is estimated to occur in up to 14.3% of cases,^1–4 requiring additional unplanned treatments and sometimes resulting in an unexpected intracranial hemorrhage or nonhemorrhagic neurological deficit (NHND). The precise rate, timing, and clinical implications of recurrence are not well elucidated. Currently, there are few studies with adequate sample sizes available for analyzing treatment trends and risks of recurrence.

The objective of this study was to evaluate a large cohort of patients with dAVF treated with various types and combinations of therapies for an accurate estimate of recurrence. We defined recurrence as angiographic recurrence after an initial angiographic cure, excluding cases of exacerbation or progression of known residual dAVF after treatment.

Methods

The infrastructure for the Consortium for Dural Arteriovenous Fistula Outcomes Research (CONDOR) is described elsewhere in detail.⁵ Briefly, collaborating institutions obtained individual institutional review board approval to conduct a retrospective analysis of all dAVFs treated at each institution. Details were collected from the clinical record and imaging databases, de-identified, and shared via a third-party host institution. Discrepancies in coded patient information were addressed directly with each contributing center. Twelve centers contributed a total of 1077 patients who presented between 1990 and 2017 with dAVF. Patients were included in the study if there was a history of dAVF treatment of any kind, a history of angiographic cure with a precise date of cure, and at least 30 days of angiographic follow-up from the date of cure. Treatment regimens were performed at the discretion of the contributing center, based on the institutional members’ experience (which varied within each institution over time and between institutions). Patients were excluded if the recurrence status or cure status was unknown or missing. Two groups of patients were compared: those with a sustained, durable cure, and those with initial cure followed by recurrence.

This study’s primary outcome was the rate of recurrence after initial cure. Demographics, presenting features, angioarchitecture on diagnosis, Borden type, Cognard grade, clinical status (via the modified Rankin Scale [mRS] score), treatment course(s), and follow-up (both angiographic and clinical) were collected for all patients. Potential risk factors for recurrence were investigated. A time-to-recurrence analysis was performed, with stratification based on potential risk factors. Secondary outcomes included 1) clinical condition at the most recent follow-up; 2) morbidity during follow-up (including subsequent dAVF-related hemorrhage, NHND, or flow-related symptoms, such as pulsatile tinnitus, cranial neuropathies, or other neurological issues presumed to be related to increased venous hypertension); and 3) dAVF-related mortality. Risk factor analysis was performed comparing the group of patients who experienced recurrence with those with durable cure. Time-to-event analysis was performed using all collective recurrence events (multiple per patient in some cases).

Statistical analysis was performed by an independent statistician. Bivariate relationships among patient characteristics and risk of recurrence over time were assessed statistically using log-rank tests for categorical effects and Cox model regression for continuous effects. Continuous variables are presented as the mean ± SD. For a specific predetermined subset of 9 risk factors of particular interest, significance values were further adjusted for multiple comparisons per Benjamini-Hochberg using a 5% false discovery rate.

Results

Of the 1077 patients in CONDOR, 457 were eligible for inclusion in this study (Fig. 1). All patients included in the analysis were treated and achieved radiographic cure between 1991 and 2017. Thirty-two patients (7%) experienced 34 events of recurrence. The mean overall clinical follow-up was 3.35 ± 3.03 years, and the mean angiographic follow-up was 2.2 ± 2.4 years. The demographics of both subgroups are listed in Table 1; patients were similar in age, sex, presenting symptoms, total number of dAVFs, and mRS score at presentation. Among 11 possible dAVF locations, there was a higher representation of tentorial dAVFs in the recurrence group (34.4%) compared with 15.4% in the cure group (p = 0.007), whereas carotid-cavernous fistulas (CCFs) were more represented in the cure group (although not significant, p = 0.077). Tentorial location (p = 0.073) did not remain significant after post hoc examination. Additionally, the years of initial presentation were comparable between the two groups, as were the rates of preexisting medical comorbidities, including hypertension, coronary artery disease, and antiplatelet and anticoagulation use (data not shown).

Study design with two subgroups including patients with a durable cure during follow-up ("cure") and those with angiographic recurrence ("recur").

Arterial supply and venous drainage were statistically similar between the two groups, but patients in the recurrence group had a trend toward higher rates of deep cerebral venous drainage (30% vs 15%, p = 0.062) and cortical venous drainage (CVD) (84.4% vs 73.6%, p = 0.089). Table 2 lists the rates of various venous drainage patterns in each group. Although this latter parameter did not achieve statistical significance, CVD emerged as statistically more prominent in the recurrence group through the analysis of sinus outflow patterns on angiography and sinus flow direction. Both Borden type III (71.9% vs 56.6%) and Cognard grade III (54.8% vs 26.1%) representation were higher in the recurrence group. Table 3 stratifies patients by Borden type and definitive treatment modality (surgery vs endovascular), with recurrence rates listed. Surgery was more commonly employed for Borden type III dAVF. Type III dAVFs treated with endovascular therapy were statistically significantly more likely to experience recurrence than those treated with surgery (13.3 vs 0%, p = 0.0001).

A summary of primary (i.e., initial) dAVF treatment modality approaches is listed in Table 4. Most patients underwent endovascular embolization (Fig. 2), and the average number of embolizations was higher in the recurrence group. Radiosurgery was also used more commonly in the recurrence group. In all 32 patients who experienced recurrence, primary treatment was endovascular embolization; subsequent treatments performed were multimodal. Of the 32 patients with recurrence, 29 (90.6%) of 32 underwent re-treatment, and 25 (86.2%) of those 29 achieved a second angiographic cure. Of the patients treated with radiosurgery for recurrence (n = 9), most (n = 6) achieved complete cure (Table 4). None of the 3 patients with untreated recurrence experienced any morbidity or mortality during follow-up. Two patients who underwent re-treatment for recurrence experienced a second episode of recurrence, requiring a third treatment. Supplemental Fig. 1 depicts the extent of initial success associated with dAVFs treated with primary endovascular embolization; note that paradoxically, patients in the recurrence group had significantly higher initial rates of complete angiographic obliteration (p = 0.026). Table 4 lists the rates of cumulative success for each treatment modality. The use of ethylene vinyl alcohol copolymer (Onyx, Covidien) was comparable between the two groups. Surgical cure was achieved at a higher rate in the cure group than in the recurrence group.

Summary of treatment modalities. A: Primary treatment for patients with durable cure. B: Primary treatment for patients who experienced primary cure but subsequent recurrence during follow-up. C: Treatment modalities used for secondary treatment of patients with recurrence. embo = endovascular embolization; surg = resection; XRT = radiosurgery. Figure is available in color online only.

A summary of primary (i.e., initial) dAVF treatment modality approaches is listed in Table 4. Most patients underwent endovascular embolization (Fig. 2), and the average number of embolizations was higher in the recurrence group. Radiosurgery was also used more commonly in the recurrence group. In all 32 patients who experienced recurrence, primary treatment was endovascular embolization; subsequent treatments performed were multimodal. Of the 32 patients with recurrence, 29 (90.6%) of 32 underwent re-treatment, and 25 (86.2%) of those 29 achieved a second angiographic cure. Of the patients treated with radiosurgery for recurrence (n = 9), most (n = 6) achieved complete cure (Table 4). None of the 3 patients with untreated recurrence experienced any morbidity or mortality during follow-up. Two patients who underwent re-treatment for recurrence experienced a second episode of recurrence, requiring a third treatment. Supplemental Fig. 1 depicts the extent of initial success associated with dAVFs treated with primary endovascular embolization; note that paradoxically, patients in the recurrence group had significantly higher initial rates of complete angiographic obliteration (p = 0.026). Table 4 lists the rates of cumulative success for each treatment modality. The use of ethylene vinyl alcohol copolymer (Onyx, Covidien) was comparable between the two groups. Surgical cure was achieved at a higher rate in the cure group than in the recurrence group.

The durations of angiographic and clinical follow-up are listed in Table 5. The mean time to angiographic recurrence was 368.7 ± 607.3 days. Patients with a durable cure had a mean angiographic follow-up of 623 ± 716 days. Supplemental Fig. 2 depicts the time to recurrence; although most recurrence events (23/34, 67.6%) occurred within 9 months, there was a bimodal distribution for timing of recurrence, with some patients experiencing a delayed recurrence (or at least detection of recurrence) beyond 2 years from primary cure. The rate of recurrence was calculated as follows: 34 recurrences/(264,967 days of recurrence-free follow-up for the cure group + 12,712 days prior to recurrence for the recurrence group) × (365 days/1 year) × 100% = 4.5% recurrence/year.

The clinical course was similar between the cure and recurrence groups; despite a durable angiographic cure, 17.8% of patients in the cure group had either worsened or new dAVF-related symptoms during follow-up, compared with 33.3% in the recurrence group. There were 2 incidents of hemorrhage (6.5%) and 3 incidents of NHND (10%) in the recurrence group at the time of recurrence; both of these were statistically significantly higher compared with those of the cure group. Despite this, functional outcomes were comparable between the two subgroups at the most recent follow-up, with roughly 80% of patients in both groups achieving an mRS score of 0 or 1. Rates of flow-related symptoms were comparable between the two groups. There was no dAVF-related mortality in the recurrence group.

The cumulative Kaplan-Meier curves are plotted in Fig. 3, with the number of patients remaining at various follow-up intervals on the x-axis. Figure 3A is limited to 3 years of follow-up, at which point 77 (16.8%) of 457 patients had follow-up; this estimates a recurrence rate of 11%. Figure 3B depicts all of the data, which estimated a long-term recurrence rate of 26% at 10 years. Patients with recurrence did have longer angiographic follow-up than the cure group (Table 5). Supplemental Fig. 3 displays Kaplan-Meier modeling of recurrence stratified by various risk factors. Since CVD was assumed to be the responsible feature for Cognard grade III dAVFs demonstrating higher rates of recurrence, subsequent multivariate modeling was conducted including 1) CVD, 2) deep cerebral venous drainage, and 3) tentorial location as potential predictors for recurrence. Tentorial location remained statistically significant as a predictor of recurrence risk after adjusting for CVD (HR 2.38, p = 0.022 in multivariate modeling).

Kaplan-Meier survival analysis curves for recurrence. Number of patients remaining at time points are listed on the x-axis. A: Data plotted for patients up to 3 years of follow-up. B: All data plotted up to longest follow-up (i.e., 10 years).

Discussion

We present the largest follow-up series to date to determine the rate of recurrence of dAVFs after initial cure. From a cohort of 457 patients, the overall rate of recurrence was 7%, within the range of previously computed rates of 0%–14.3%.^1–4 Projected long-term rates approached 11%, although our calculations were in the setting of stringent exclusion criteria with regard to follow-up. A large number of patients who were treated and in whom the dAVF was cured were lost to follow-up, suggesting that the cohort of patients with follow-up probably overrepresents the patients with recurrence. The risk of recurrence was 4.5% per year after cure, which is not insignificant. This is based on collective data from 457 patients with adequate angiographic follow-up after cure. Recurrence occurred in a bimodal fashion, with a much more prominent peak from 0 to 9 months (mean time to recurrence 1.01 years, median time 6.3 months), with some patients experiencing recurrence (or at least formal angiographic detection) beyond 2 years from cure. Most cases of recurrence are clinically silent, with very rare yet statistically significantly higher rates of hemorrhage or NHND. Interestingly, a significant proportion of patients with initial cure still experienced either persistent symptoms (8.1%) or worsening/new symptoms (17.8%). Thus, clinical follow-up cannot be used as a surrogate for angiographic evaluation. Importantly, the clinical implications for angiographic recurrence are marginal, as most patients still have good functional outcomes.

Recurrence exclusively occurred in cases with primary endovascular treatment, suggesting that this is an endovascular-specific phenomenon. It is well known that dAVF cure requires complete penetration of the fistulous point and draining vein (or common venous outlet),⁶ although there are cases of recurrence despite seemingly adequate venous penetration.¹ Interestingly, the presence of Onyx in the embolysate made no difference to recurrence rates. Although the mechanism of recurrence is not addressed in this investigation, it is generally accepted to be related to suboptimal casting of the draining vein. Nogueira et al.⁴ first reported using Onyx in 12 patients with dAVF, 11 of whom achieved cure, and 9 had angiographic FU at an average of 4.4 months. One patient (11.1%) had recurrence. This same phenomenon was observed in the early Onyx-dAVF experience from the Barrow Neurological Institute.⁷ Although neither the rate nor the timing was reported, the authors speculated that small arterial feeders that initially appeared silent may have persisted after primary “cure” and subsequently enlarged and parasitized nearby sinus structures especially in the setting of treatment-related sinus thrombosis/venous hypertension, and with the release of angiogenic factors from a hypoxic microenvironment. This is distinctly different from a separate, de novo dAVF (which is also reported in that series, but not the topic of this discussion). A subsequent series³ found recurrence in 1 (7.1%) of 14 patients treated with Onyx at 6 months of follow-up. Adamczyk et al. reported 2 cases of cranial dAVF with recurrence,² occurring at 8 and 14 months; both were asymptomatic, and the authors summarized the literature at that time as well as possible explanations for recurrence. Both cases in that series had a well-formed Onyx cast that remained in place on subsequent angiography. Again, the pathogenesis was hypothesized to be related to local vascular injury, angiogenic growth factor release, and venous hypertension (due to venous occlusion with embolization).

Ambekar et al. presented the first large series aimed specifically at identifying features predictive of recurrence in endovascular-treated lesions only.¹ Of the 58 patients treated at their institution (CCFs excluded), they limited the analysis to the 26 patients who underwent endovascular embolization with Onyx, 21 of whom had angiographic follow-up (ranging from 2 to 39 months, average 14 months). Three patients (14.3%) experienced recurrence. In 1 case, the authors found on rereviewing the initial treatment angiogram that there was not in fact complete penetration of the common venous outlet. For the other 2 patients, the authors suggested that there may have been a patent channel within the Onyx cast—given that the agent precipitates in a radial fashion from outward to inward—that permitted recurrence in a delayed fashion. In other words, this highlights the notion that an initial angiographic cure may merely be temporary angiographic nonopacification of the dAVF (due to stagnant flow and a partial cast in the fistula/venous outlet). Subsequent meningeal recruitment can then expand a small residual dAVF.⁶

In the current series, tentorial location emerged as an independent risk factor for recurrence, and CVD (and thus higher Borden type and Cognard grade) and deep cerebral venous drainage also had an association in univariate analysis. A separate analysis showed that as Borden type increased, so too did the proportion of lesions treated with surgery compared with endovascular therapies. This seems reasonable because the surgical treatment of type I and II lesions often requires sinus packing or skeletonization of the sinus, which is uncommon. Borden type III lesions treated endovascularly (as opposed to surgically) were significantly more likely to experience recurrence. Borden type III lesions are effectively treated with clip ligation with occlusion of the immediate draining vein. It is possible that our series reflects a high “initial (false) cure” rate for Borden type III lesions treated with endovascular therapy, whereby a true venous cast of the draining vein was not obtained in all cases, hence the recurrence in this group. Thus, lesions with these features should be considered for closer monitoring and perhaps delayed long-term follow-up compared with more benign, lower-grade dAVFs. Conversely, CCFs were less prone to recurrence. Lawton et al. presented a classification system, algorithm for selecting surgical strategy, and overall microsurgical experience with treating 33 tentorial dAVFs.⁸ Cure was achieved in 31 (93.9%) of the 33 patients, with an average of 4.2 years of follow-up, and no cases of recurrence, although the details for angiographic follow-up are not clear. Early primary surgical intervention may be warranted in select tentorial dAVFs, as our study indicates that surgical success is much less likely (71.4% vs 92.7%, p = 0.041) in cases that have already had a recurrence.

Interestingly, we did not find any reported cases of recurrence after microsurgical resection or radiosurgical treatment of dAVF. This is consistent with the hypothesis reported previously^1,2 that recurrence is actually a temporarily angiographically silent but still anatomically intact dAVF. This study was not designed to determine the exact timing of the recurrence phenomenon, if it does occur due to an etiology other than incomplete cure. It is not clear from the current series whether or not this “delayed recurrence” is a real phenomenon; we investigated the 4 patients with reported recurrence at 24 months or beyond and none had formal angiography confirming cure status after the initial cure (all had noninvasive imaging without definite cure or recurrence appreciated). Thus, future studies might aim to standardize imaging intervals with formal angiography to precisely define when recurrence can occur. If a delayed phenomenon is observed, it may be due a separate pathogenesis, such as delayed venous hypertension in an adjacent segment of the venous sinus, and parasitization of previously inactive arterial feeders. Similarly, although not observed in this current series, a recurrence after radiosurgically or surgically cured lesions would likely occur at a later time interval and could potentially be due to either 1) an adjacent, separate, but communicating venous structure that becomes active in a delayed fashion; or 2) the initially occluded venous structure reconstituting due to local flow dynamics. There does not seem to be a perfect correlation between clinical symptom resolution and angiographic cure, as patients with angiographically cured lesions can still report new or progressive symptoms in approximately 18% of cases, and persistently stable symptoms in 8% of cases, so there may be additional late recurrences that go undetected due to lack of angiographic follow-up. Close attention to dAVF-specific symptoms and vigilance for pervasive, nonspecific neurological complaints can help with initial counseling and managing patient expectations for improvement after treatment, although overall, good functional outcome was seen in most patients, even with hemorrhage- or NHND-associated recurrence.

Conclusions

Recurrence after initial cure in cranial dAVF occurs at a rate of 4.5% per year. Longer-term projected rates approach 11% at 3 years and 26% at 10 years, although these may be overestimates. Recurrence was only seen after endovascular embolization as the first treatment modality, particularly in Borden type III lesions, but this probably reflects patient selection and frequency of treatment, as microsurgical ligation was seldom used in Borden type I and II dAVFs. Tentorial location, CVD, and deep venous drainage are risk factors for recurrence. Long-term follow-up is important in patients with “cured” dAVFs, even beyond a year from cure, to ensure a durable result and prevent subsequent hemorrhage.

Acknowledgments

We thank Jason Barber, MS, for his assistance with statistical analysis, and Sharon Durfy, PhD, for assistance with manuscript preparation.

Appendix

Disclosures

Dr. Kim reports stock ownership in SPI Surgical. Dr. Levitt reports non–study-related funding support from the NINDS, NIH, AHA, Stryker, Medtronic, and Philips Volcano; consultant fees from Medtronic; equity interest in Synchron, Cerebrotech, and eLoupes; and is an advisor for Metis Innovative. Dr. Alaraj reports funding support from the NIH, and consultant fees from Cerenovus and Siemens. Dr. Starke reports funding support from NREF, Joe Niekro Foundation, Brain Aneurysm Foundation, Bee Foundation, and the NIH; and consultant fees from Penumbra, Abbott, Medtronic, and Cerenovus. Dr. Gross reports consultant fees from MicroVention and Medtronic. Dr. Samaniego reports serving as a proctor with MicroVention.

Author Contributions

Conception and design: Kim, Abecassis, Levitt. Acquisition of data: Abecassis, Meyer. Analysis and interpretation of data: Abecassis, Meyer. Drafting the article: Abecassis. Critically revising the article: Abecassis, Meyer. Reviewed submitted version of manuscript: all authors. Approved the final version of the manuscript on behalf of all authors: Kim. Study supervision: Kim, Levitt.

Supplemental Information