Anatomical and surgical characteristics correlate with pachymeningeal failure in patients with brain metastases after neurosurgical resection and adjuvant stereotactic radiosurgery

Neurosurgery (NS) is an essential modality for large brain metastases (BM). Postoperative stereotactic radiosurgery (SRS) is the standard of care adjuvant treatment. Pachymeningeal failure (PMF) is a newly described entity, distinct from classical leptomeningeal failure (LMF), that is uniquely observed in postoperative patients treated with adjuvant SRS. We sought to identify risk factors for PMF in patients treated with NS + SRS. From a prospective registry (2009 to 2021), we identified all patients treated with NS + SRS. Clinical, imaging, pathological, and treatment factors were analyzed. PMF incidence was evaluated using a competing risks model. 144 Patients were identified. The median age was 62 (23–90). PMF occurred in 21.5% (31/144). Female gender [Hazard Ratio (HR) 2.65, p = 0.013], higher Graded Prognostic Assessment (GPA) index (HR 2.4, p < 0.001), absence of prior radiation therapy (HR N/A, p = 0.018), controlled extracranial disease (CED) (HR 3.46, p = 0.0038), and pia/dura contact (PDC) (HR 3.30, p = 0.0053) were associated with increased risk for PMF on univariate analysis. In patients with PDC, wider target volumes correlated with reduced risk of PMF. Multivariate analysis indicated PDC (HR 3.51, p = 0.0053), piecemeal resection (HR 2.38, p = 0.027), and CED (HR 3.97, p = 0.0016) independently correlated with PMF risk. PMF correlated with reduced OS (HR 2.90, p < 0.001) at a lower rate compared to LMF (HR 10.15, p < 0.001). PMF correlates with tumor PDC and piecemeal resection in patients treated with NS + SRS. For unclear reasons, it is also associated with CED. In tumors with PDC, wider dural radiotherapy coverage was associated with a lower risk of PMF.


Introduction
Brain metastases (BM) afflict approximately 20% of cancer patients [1,2], and portend a poorer prognosis [3]. Multimodal therapies for BM include surgery, radiotherapy, chemotherapy, immunotherapy, and targeted therapies. Neurosurgery is an essential modality for patients with large or symptomatic brain metastases (BM) causing mass effect, providing local control that is significantly increased with the use of postoperative radiation [4,5]. Traditionally, the standard of care for patients with BM undergoing surgery was postoperative whole-brain radiation therapy (WBRT), after the landmark randomized trial by Patchell et al. in 1998 [4]. More recently, stereotactic radiosurgery (SRS) has become the standard of care modality to treat BM, as both a 1 3 definitive and adjuvant therapy to surgery because it lessens cognitive toxicity while achieving equivalent rates of overall survival compared to WBRT [6,7].
However, as the treatment paradigm shifted from WBRT, a distinct pattern of failure has emerged, hypothetically resulting from intra-operative microscopic tumor cell seeding [8], called nodular leptomeningeal (nLM) or pachymeningeal failure (PMF). This relatively new entity is contrasted to classical leptomeningeal failure (cLMF) that can occur with or without surgery [9][10][11]. PMF remains incompletely characterized and its etiology is poorly understood. To that end, we hypothesized that clinical, imaging, pathological, and treatment factors are associated with PMF and sought to identify them. Moreover, we studied its incidence and clinical course.

Materials and methods
From a prospectively collected database of BrM patients, we identified 144 consecutively treated patients; all 18 years or older, treated with upfront surgery, followed by single or fractionated Gamma Knife SRS from January 2010 to June 2021. Patients who received concomitant WBRT were excluded. This study was approved by our local institutional Research Ethics Board at University Health Network.

Treatments
All patients were treated with upfront surgical resection followed by adjuvant Gamma Knife Radiosurgery (Elekta AB) to the surgical cavity. Adjuvant SRS was usually delivered 2-4 weeks after surgery to the cavity in 1-3 fractions. SRS dose was chosen as per our institutional policies and at the discretion of the staff radiation oncologist for single-fraction SRS (SF-SRS) ≤ 4 cc/21 Gy, 4-10 cc/18 Gy, > 10 cc/15 Gy. This scheme was used for cavities until the fractionated SRS (F-SRS) technique was implemented in our center in 2017, using the ICON frameless system, utilizing the following dosing regimen: 4-< 10 cc 27 Gy/3, 10-< 20 cc 24 Gy/3, and > 20 cc 21 Gy/3. Our SRS treatment procedure for cavities has been previously published [12]. We generally use the contouring guidelines for cavity SRS by Soliman et al. [13], which recommends including the surgical tract (in addition to the surgical cavity) and a CTV expansion of 1-10 mm based on preoperative tumor contact with venous sinus or dura. All treatments were prescribed to a median isodose line of 50% (range 40-60) with a GTV coverage > 98%. Each SRS treatment plan is reviewed and approved by two physicists, a radiation oncologist, and a neurosurgeon.

Surveillance
Each patient was followed every 8-12 weeks after SRS with an MRI brain examination during the first year, then every 3-4 months for the second year, unless additional imaging outside of this schedule was clinically indicated.

Outcomes
A diagnosis of failure type [PMF, LMF, Distant failure (DF), and Local Failure (LF)] was made by consensus based on MRI. Since PMF is a relatively new pattern of failure, associated with surgical cavities and can be misclassified as LMF, LF, or DF, the patterns of failure for each patient were analyzed by two radiation oncologists and one neurosurgeon attending our weekly multidisciplinary BM conference and have been trained according to the proposed module from Turner et al. [14]. PMF was defined as an enhancing nodule (or nodules) arising from the pachymeninges (dura mater) with no involvement of the skull to indicate secondary extension to the pachymeninges, extending 1 cm beyond the planning target volume of the stereotactic field [9,15] (Fig. 1). The preoperative brain MRI, simulation MRI, including treatment volumes (CTV, PTV, OARs), SRS isodose lines, and the MRI brain of the date of recurrence if occurred were imported and fused into the Raystation planning software. Neurologic death (ND) was defined as per the definition established by Patchell et al. [4]. Regardless of extracranial disease status, if a patient died with evidence of advanced neurologic dysfunction and had treated or untreated brain disease, including BM, LMF, and PMF on their last imaging prior to their death, their death was classified as neurologic.

Statistical methods
Variables were summarized using mean/sd and median/IQR/ range for continuous variables and counts with percentages for categorical ones. The Kruskal-Wallis test was used to assess the correlation between the treatment type and GPA. The time to PMF diagnosis is calculated in months from the date of surgery to PMF or death; death without PMF was treated as a competing event. In cases of no PMF or death, the follow-up time was used as a censoring event. The difference in time to PMF was assessed using the Fine-Gray method and hazard ratios with p-values were reported. In situations of complete separation, the pseudo-likelihood ratio test was used. The PMF rates were evaluated using the cumulative incidence function (CIF). Overall survival (OS) was calculated in months from the date of surgery or PMF/LMF to death. The survival percentages were calculated using the Kaplan-Meier estimator, and the difference between groups was assessed using the Cox Proportional-Hazards model. When comparing the overall survival of PMF vs LMF patients, starting at the date of surgery, a timedependent covariate was used for PMF/LMF. P values ≤ 0.05 were deemed statistically significant.

Results
One hundred forty-four (144) consecutively treated patients harboring at least one brain metastasis treated with surgery and adjuvant SRS for the index lesion were examined.
Baseline patient and tumor characteristics are listed in Table 1. In 73 patients (51%), the index lesion was a single brain metastasis. Most lesions were supratentorial (116, 81%) and the median tumor volume was 12.1 cc (Q1 6, Q3 21). Moreover, the contact of the tumor with the pia/ dura and the dural venous sinuses was recorded. In these cases, a subgroup analysis regarding the correlation of the extent of radiation contouring with PMF was performed. Treatment characteristics are described in Table 2. In 66 patients (47%), en bloc resection was achieved, while 75 tumors (53%) were resected in a piecemeal fashion. Administration of systemic therapy (ST) in the perioperative period (2 months before surgery to 6 months after surgery) was noted; 81 patients (56%) received ST during this period. The type of ST, the extent of resection, the number of SRS fractions, the total SRS dose, and the inclusion of the surgical tract in the radiation plan were also recorded.

Incidence and risk factors of PMF
PMF was identified in 31 (21.5%) cavities with a cumulative incidence (95% CI) at 12 and 24 months of 18% (13%, 26%) and 22% (16%, 29%), respectively (Supplementary Fig. 1). The earliest PMF event was recorded during the 3rd postoperative month and the vast majority of events occurred within the first postoperative year. No events occurred after the 16th postoperative month. The median time from surgery to PMF diagnosis was 5.4 months.
The variables correlated with PMF on UVA are shown in Table 3   Patients harboring tumors in contact with dura/pia (n = 76) and dural venous sinuses (n = 38) were analyzed in a subgroup analysis according to the dural contouring margins (1-5 mm vs 6-10 mm) of the adjuvant SRS treatment. Wider contouring (6-10 mm) was associated with a reduced risk of PMF for tumors in contact with dura/pia [HR: NA (no PMF when contouring 6-10 mm), p = 0.046]. No statistically significant difference was observed between the 2 groups for tumors in contact with venous sinuses (p = 0.87) (Supplementary Tables 2 and 3).
In a subgroup analysis, we attempted to correlate the survival from the time of PMF diagnosis with the number of PMF metastases (solitary vs multiple), PMF subtype

Discussion
Pachymeningeal failure is an emerging pattern of failure, distinct from the classical (sugarcoated or linear) leptomeningeal failure (cLMF), and unique to postoperative patients   [9,12]. Past studies did not differentiate between cLMF and PMF [16][17][18]. Herein we report our institutional experience regarding PMF incidence and risk factors, its prognostic significance, and factors affecting survival following diagnosis. PMF is underrecognized with wide variability in reported incidence; 1-year rate varies from 7 to 31% [17][18][19][20][21][22][23]. A plausible explanation for its variable incidence is the unfamiliarity among clinicians leading to misdiagnosis of PMF as LMF, local or distal failure [6]. In our cohort, PMF determination was conducted rigorously, by three assessors (two radiation oncologists and one neurosurgeon). The cumulative incidence of PMF at 12 and 24 months was 18 and 22%. We demonstrated an association between PMF and piecemeal resection. While an increased risk of LMF [24] and local recurrence [25] with piecemeal resection has been reported in other studies, this is the first to correlate this factor with PMF. Piecemeal resection may increase the risk of intraoperative tumor spillage [24,26]. Spillage in turn might lead to seeding in the well-vascularized pachymeninges. Since this is a potentially modifiable risk factor, en bloc resection is recommended [27,28], though piecemeal resection is often unavoidable for very large BM or for BM in eloquent areas where resection is performed via a narrow corticectomy and manipulation of the healthy brain needs to be minimized [28]. Other factors previously correlated with LMF, such as the cystic or hemorrhagic nature of the lesion, [9], did not correlate with PMF in our cohort.
We found that PMF correlated with tumor proximity to pia or dura mater (presumable dissemination by direct extension during surgery) [29,30], particularly lesions in contact with the falx/tentorium, which are deep, require a piecemeal resection, and are approached through narrow sulcal/subarachnoid corridors where spillage of tumor cells might be facilitated by surgical instruments. The falx and tentorium are supplied by several overlapping vascular Statistically significant values are indicated by bold sources, including branches of the meningeal arteries and cerebral arteries, making them an anastomotic pathway between the dural and parenchymal arteries [31]. The rich vascularization of the falx and tentorium might contribute to PMF, rendering them an ideal location for seeding. We found that PMF tends to occur in patients with controlled extracranial disease, in line with Cagney et al. [9], in contrast to LMF, which correlates with extensive systemic disease [29]. This suggests that iatrogenically introduced PMF may be biologically distinct from LMF caused by natural disease progression [11]. Similarly, and potentially for related reasons, GPA also correlated with PMF risk on UVA. We explored the possibility that this association was related to treatment with ST following resection but, as with Cagney et al. [9], we did not identify a relationship between ST and PMF. Cagney et al. [9] suggest that a higher rate of death among patients in their cohort with active extracranial disease accounted for a lower rate of PMF. We believe that the higher cumulative incidence in our analysis, in which death is considered a competing risk, argues against that explanation.
On univariate analysis, prior radiotherapy was inversely correlated to PMF risk, contrary to prior reports [9]. Previously radiated BMs requiring salvage resection may contain a lower tumor burden [32], as they represent a mixture of a viable tumor with necrosis [33]. Moreover, radiation causes corrosive damage to the cell membrane, which disrupts cells' integrity, and mobility, and can affect cell survival [34]. These factors can translate into a lower risk of intraoperative pachymeningeal seeding. Alternatively, prior radiation may alter the adjacent dura so that it is at a lower risk of being seeded. Female gender was also associated with PMF, in congruency with one study on LMF in melanoma [35], but not with others [36,37].
In our study, PMF was correlated with reduced OS, though to a lower extent than LMF, in line with prior reports [8]. This difference in the natural history of the disease supports the theory that classic LMF and PMF are distinct with the former being more aggressive and untreatable [8,38].
WBRT was the most commonly used treatment for PMF in our study. Interestingly, we found that patients with PMF can also develop spinal deposits, [39], which may denote a generalized spread, associated with reduced OS. Death was the result of a neurological decline in most of the cases [9].
Our analysis suggests that en bloc resection can reduce the risk of pachymeningeal seeding, while a radiotherapeutic field covering a wide margin of the at-risk dura may be beneficial for patients with risk factors including stable systemic disease and tumors in contact with dura/pia [7]. Hypofractionated SRS may make this strategy more feasible and analysis of the recently completed CEC7 trial may provide some insight into that, although for that trial treated radiation volumes were, by design, identical regardless of treatment  arm. Although the contouring consensus guidelines for BM cavities by Soliman et al. recommend including 1-10 mm dural/bone flap based on preoperative tumor contact, it is unknown if this approach decreases the incidence of PMF [13]. Our data suggest that a wider dural margin (6-10 mm) is associated with a lower risk of PMF for tumors in contact with pia/dura, while other studies argue against that and otherwise reports have been mixed. Neoadjuvant stereotactic radiosurgery (NASRS) is an emerging treatment currently under investigation, which aims to effectively control intracranial disease while minimizing the risk of PMF [40]. Hypothetically, after NASRS, the tumor undergoes necrosis and reduction in immunomodulatory cell populations, which may make intraoperatively spilled cells less likely to seed [41,42]. The role of neo-adjuvant radiotherapy to improve rates of local control has been established in other disease sites, such as sarcoma, rectal, esophageal, and pancreatic cancer [43][44][45][46][47]. Currently, although it is still under investigation for brain metastases in many centers including our [40], preliminary data has shown benefits for the use of NASRS, not only to improve local control and PMF but also to allow a better delineation of the target and a potential reduction of radiation toxicity [21,42].

Limitations
The retrospective analysis, albeit from a prospective database, comes with inherent biases. The data are derived from a single institution, thus, limiting generalizability. Although these data are from a large, specialized cancer center, most clinical investigations benefit from a multicenter approach. Due to the limited number of patients some of the statistically significant factors in the UVA could not be included in the MVA. However, we chose a model based on a clinical rationale that we considered the most inclusive. The small cohort of patients suffering PMF renders it difficult to identify factors related to survival and perform confirmatory analysis. Another limitation of our study is the heterogeneity of brain metastases histology, which could potentially influence the risk of PMF/LF and the response to treatment modalities. The identification of En-bloc vs piecemeal resection was based on the retrospective review of operative reports, which may be subject to variability in reporting and interpretation. Finally, classifying deaths as neurologic versus non-neurologic is inherently challenging, particularly in a retrospective study.

Conclusions
PMF is an understudied phenomenon that correlates with pre-operative pial/dural contact, piecemeal resection, and controlled extra-cranial disease in patients treated with NS + SRS for BM. The latter indicates that iatrogenically introduced PMF may in fact be biologically distinct from LMF that occurs as a result of natural disease progression. In tumors, in contact with pia/dura, wider dural margins in radiation delivery are associated with less risk of PMF. While less morbid than LMF, PMF is a critical event that deserves increased vigilance and study of alternative methods such as neoadjuvant SRS or alternative postoperative radiotherapy strategies covering a larger area.