Introduction

In June 2016, the World Health Organization (WHO) classification of tumors of the central nervous system (CNS) was updated, restructuring solitary fibrous tumor (SFT) and hemangiopericytoma (HPC) into one combined entity [1]. This decision was driven by a number of recent pathological studies demonstrating a common genetic etiology between the two tumors [2,3,4,5,6,7,8]. Despite the fact that the WHO soft tissue classification scheme introduced this change in 2002 [9], the decision to combine these two tumor types in the CNS was delayed due to their distinct biological behaviors [10,11,12,13]. SFTs of the CNS are usually benign and well controlled with surgery alone [5, 11, 13]. HPC, on the other hand, is notorious for its propensity to recur [14,15,16,17,18], almost invariably [19,20,21,22], and metastasize extracranially [14,15,16,17, 22, 23].

The first cases of CNS SFT, also known as meningeal SFT, were described in 1996 [24]. Although CNS HPC of the meninges was described over 70 years ago [25], the WHO did not recognize the tumor as a distinct entity, but rather a malignant variant of meningioma, until 1993 [26]. In the most recent update [1], CNS HPC is by default grade II or III. CNS SFT is grade I, unless it exhibits anaplastic features, in which case it is upgraded to III. Thus, the two tumors are now considered part of the same spectrum.

To our knowledge, there is only one small case series to date that has analyzed treatment-related outcomes of this combined entity [12]. These are rare tumors and little is known about optimal management. As of 2010, there were only 220 CNS SFT cases published in the literature [11]. The majority of our knowledge about CNS HPC is derived from small case series, and the role of adjuvant radiation therapy is still unresolved. Thus, the Surveillance, Epidemiology, and End Results (SEER) Program and other cancer registry databases offer unique opportunities to perform large population-based studies on these rare tumors [15, 27,28,29].

Our current study is the largest single-patient series to date for the combined entity of CNS SFT/HPC, and the first population-based study to examine outcomes based on the new WHO guidelines. In this study, we use the SEER database to assess age-adjusted incidence, demographics, and prognostic factors of SFT/HPC of the CNS.

Methods

Age-adjusted incidence was calculated using SEERSTAT and the Incidence—SEER 18 Database, which covers approximately 27.8% of the US population. The SEER database (November, 2015 submission, 1973–2013) [30] was then queried to collect all SFT (International Classification of Diseases for Oncology, 3rd Edition [ICD-O-3] code 8815) and HPC (ICD-O-3 code 9150) tumors within the brain or CNS diagnosed after 1998 (n = 454). This year was selected ensure consistency in coding for surgical procedures. Information on age, sex, race, histological tumor type, primary site, WHO grade, ICD-0-3 behavior, extension, size, distant metastasis at diagnosis, surgery at primary site, and radiation were collected and coded. Cases with incomplete follow-up (n = 2) and cases that did not receive surgery (n = 25) or for which surgery status was unknown (n = 1) were excluded. From these cases of interest (n = 426), 10 were excluded (2.3%) that had incomplete information on receipt of radiation.

ICD-0-3 behavior coding in SEER is based on histological morphology and indicates the likely behavior of the tumor regarding its potential to invade surrounding tissue, based on what most pathologists believe is usual for that tumor. However, ICD-0-3 behavior can be changed at the discretion of the coding pathologist. Histological (ICD-0-3) behavior classifies tumors as benign, borderline malignant, or malignant and was analyzed in this study as an ordinal variable. Tumors may be coded as borderline based on a pathologist’s observations that the tumor has “low, borderline, or uncertain malignant potential.” WHO grade was not included in the analysis due to obvious inconsistencies between historical and current WHO grading guidelines in the tumors that were diagnosed before 2007 [31]. Tumor location was coded as supratentorial, infratentorial, or unknown/other based on the Primary Site and the Collaborative Staging (CS) Extension variables. The unknown/other category included spinal tumors and tumors for which location in relation to the cerebellar falx was unclear. Tumor size was coded as < 7, 7 cm or greater, or unknown. Treatment was coded as subtotal resection (STR), gross-total resection (GTR), subtotal resection plus radiation therapy (STR + RT), GTR plus radiation therapy (GTR + RT), or unknown extent of resection with or without (±) radiation therapy based on the surgery at primary site and radiation variables. The “unknown extent of resection” category included “local excision of tumor/excisional biopsy” and “surgery, not otherwise specified”. Invasiveness [15] and metastasis [16, 22] at diagnosis were not analyzed as this information was not available for tumors diagnosed before 2004. Patients in this expanded series overlap with those in our previous published work [15]. Patients in this series are also presumed to also overlap with those in previously published SEER studies [32,33,34], although all data extraction, coding, and analysis were conducted independently.

All statistical analysis was carried out using the IBM SPSS Statistics software. Median survival time was determined using the Kaplan–Meier method. Univariable (UVA) and multivariable analyses (MVA) were conducted using Cox-Regressions to determine hazard ratios (HR). Variables with a statistically significant univariate association with treatment or survival were included in the MVA. Association between tumor type or treatment versus other variables were determined using the Pearson’s Chi square test. Median values were compared based on the column proportions above versus below the median. Tests with two-tailed p values < 0.05 were considered statistically significant.

Results

Incidence and demographics

The age-adjusted incidence of SFT/HPC of the central nervous system from 2000 to 2013 was calculated to be 3.77 per 10,000,000 people (Supplemental Table S1). The age-adjusted incidence of SFT and HPC of the CNS were 0.58 and 3.18 per 10,000,000, respectively. The incidence of SFT/HPC was similar in men (3.71) and women (3.78), but higher in Asian/Pacific Islanders (4.55) than other races. Age-adjusted incidence had increased from 2000 to 2013, reaching almost 6 per 10,000,000 in 2013 (Supplemental Figure S2). Racial demographics did not change significantly over this period of time, and thus did not account for the increasing incidence. Given the increasing incidence, we approximate 230 cases to be diagnosed in the United States in 2017.

416 tumors were identified that matched search criteria, 350 of which were HPC and 66 of which were SFT. Demographical and clinical characteristics are displayed in Table 1. There was a slight female predominance (52.9%), and most patients were white (78.8%). Median age of diagnosis was 54. 84.1% of tumors were HPC, 76.9% were considered borderline malignant or malignant by ICD-0-3 behavior, 7.2% occurred infratentorially, and 10.3% were 7 cm or greater. The percentage of patients that received STR, GTR, STR + RT, GTR + RT, and unknown extent of resection ±RT was 3.4, 31.3, 5.5, 24.3, and 35.6%, respectively. STR (8.9% total) may have been underestimated due to strict inclusion criteria. Radiation (45.4% total) may have been underestimated, as a recent study revealed the SEER database to be 80% sensitive in capturing this variable [35]. HPC tumors were more likely to be borderline malignant or malignant (p < 0.0005), supratentorial (p < 0.0005), and treated with RT (p < 0.0005). Differential treatment was significantly associated with tumor type, histological behavior, location, and size (Supplemental Table S3).

Table 1 Demographics
Full size table

Survival

Median and mean follow-up times for all tumors were 46 and 56 months, respectively. Over this time period, 95 deaths occurred and 321 cases were censored. Median survival was 155 months (Supplemental Figure S4), with 5- and 10-year survival rates of 78 and 61%, respectively. Median survival time was 138 months for HPC and was not reached in SFT. In UVA, older age (HR = 1.040, p < 0.0001), increased malignant histological behavior (HR = 1.384, p = 0.034), and location (p = 0.002) were significantly associated with decreased overall survival (Table 2, Supplemental Figure S5). Asian/Pacific Islander race was significantly associated with increased survival when compared with White race (HR = 0.341, p = 0.020), but overall, race was not a significant prognostic factor. GTR (HR = 0.332, p = 0.042), GTR + RT (HR = 0.301, p = 0.029), and unknown extent of resection ±RT (HR = 0.243, p = 0.010) were significantly associated with increased survival over STR alone. Univariable subgroup analysis by tumor histology is displayed in Supplemental Table S6.

Table 2 Univariable and multivariable analysis of overall survival
Full size table

Age, tumor type, histological behavior, location, size, and treatment were included in the multivariable analysis. Tumor type and size were included due to their association with the treatment variable (Supplemental Table S3). Older age (HR = 1.038, p < 0.0001) and infratentorial versus supratentorial location (HR = 2.019, p = 0.038) were significantly associated with decreased survival (Table 2). Size > 7 cm was associated with decreased survival but did not reach statistical significance (HR = 1.710, p = 0.115). GTR (HR = 0.313, p = 0.041) and GTR + RT (HR = 0.215, p = 0.008) were associated with an incremental increase in survival (Table 2, Supplemental Figure S7). STR + RT was associated with increased survival over STR alone but did not reach significance (HR = 0.303, p = 0.134). Unknown extent of resection ±RT was significantly associated with increased survival over STR (HR = 0.251, p = 0.016).

Overall, GTR + RT approached but did not attain a statistically significant increase in survival compared with GTR alone (HR = 0.574, p = 0.057, Table 3; Fig. 1). However, in the HPC and borderline malignant/malignant subgroups, GTR + RT was associated with significantly increased survival compared with GTR alone (HR = 0.537, p = 0.039 and HR = 0.525, p = 0.038, Table 3; Figs. 1, 2). MVA could not be conducted in the STR, benign, or SFT subgroups due to a lack of events (deaths). In the subgroup of patients with unknown extent of resection, RT was associated with increased survival but not reach significance (HR = 0.499, p = 0.121, Supplemental Figure S8).

Table 3 Multivariable analysis of overall survival in GTR subgroups
Full size table
Fig. 1
figure 1

Survival in GTR subgroups. GTR gross-total resection, RT radiation therapy, HPC hemangiopericytoma

Full size image
Fig. 2
figure 2

a, b Multivariable Cox-Regression analysis. GTR gross-total resection, RT radiation therapy, HPC hemangiopericytoma

Full size image

To rule out the effects of immortal time bias in regards to radiation therapy, we conducted a separate analysis excluding cases that died within 3 months of diagnosis, resulting in 381 cases available for analysis. GTR + RT (HR = 0.237, p = 0.031) and unknown extent of resection ±RT (HR = 0.238, p = 0.031) were significantly associated increased survival over STR, while GTR alone did not attain statistical significance (HR = 0.280, p = 0.054, Supplemental Table S9).

Discussion

In this study, GTR and GTR + RT were associated with an incremental increase in survival over STR alone. In the HPC and borderline malignant/malignant subgroups, GTR + RT was associated with increased survival over GTR alone. While the importance of GTR in CNS HPC is not disputed [16, 17, 22, 23, 36,37,38,39], the role of adjuvant radiation is more obscure. The majority of published studies support the use of post-operative radiotherapy (PORT) in HPC of the CNS [16, 21,22,23, 32,33,34, 40,41,42,43], while few studies have shown a survival disadvantage [15, 17, 36]. Controversy regarding the use of PORT in CNS HPC has increased recently due to a frequently cited meta-analysis [36]. The study, published by Rutkowski et al., found no survival advantage associated with PORT, and among patients who received radiation, decreased survival in those that received > 50 Gy.

This finding conflicts with several other studies [16, 44, 45], including another meta-analysis [21], which found a dose–response relationship correlating with decreased recurrence or mortality. The meta-analysis by Rutkowski et al. was criticized for including older studies [34] dating back to 1928. Two other meta-analyses, which reviewed more contemporary studies, came to the opposite conclusion when examining the effect of radiation in CNS HPC [21, 23].

There are three population-based studies on HPC of the CNS using SEER [32,33,34], all of which support the use of PORT. Another large multi-institutional retrospective study from Korea (140 patients) found a significant survival advantage associated with PORT [42]. Of the nine large studies published with a cohort > 75 patients [14, 21, 32,33,34, 36, 42, 43, 46], seven show a statistically significant survival advantage associated with PORT. Stessin et al. showed a statistically significant survival advantage while accounting for the immortal time bias by eliminating cases that died within 6 months of diagnosis [34].

The benefits of PORT are not disputed in the setting of STR of CNS HPC. However, to date, there are no studies that show significant survival advantage of adjuvant radiotherapy in patients who have received GTR compared with GTR alone. Recently, Sonabend et al. found that GTR plus radiation was the only treatment modality associated with a significant survival increase over biopsy only [33]. Similarly, Schiariti et al. found GTR plus radiation to confer a statistically significant survival advantage over STR alone [22]. In the study, GTR and radiation significantly improved disease-free interval over GTR alone but did not achieve statistical significance for overall survival. Based on these findings, many authors have recommended routine adjuvant radiotherapy, even when GTR is achieved [16, 22, 39, 47]. CNS HPC is known to be infiltrative/invasive in nature [14, 15, 20]. The tumor’s propensity to recur, almost invariably [20], implies microscopic residual disease exists after GTR. The ability of radiation to target microscopic residual disease provides a plausible mechanism for its efficacy as an adjuvant therapy, even after GTR.

There is less literature published on the optimal treatment of CNS SFT. While several authors state the disease can be controlled with GTR alone [5, 11, 13], a recent study examining treatment-related outcomes in SFT/HPC of the CNS showed a significant progression-free survival advantage associated with adjuvant radiation, which approached significance in the SFT subgroup [12]. Our study did not have enough cases to analyze survival after radiation in the benign or SFT subgroups. Therefore, we cannot comment on optimal treatment in either of these groups based on our results. Our results do, however, imply that GTR + RT is the optimal treatment for borderline malignant/malignant tumors, regardless of histological tumor type.

This study provides an updated estimate for age-adjusted incidence of CNS HPC [33] and the first estimate of incidence of CNS SFT. We found age-adjusted incidence to be increasing in both tumor types. Racial demographics remained consistent throughout the study, so immigration patterns did not account for the increase in incidence. In general, there has been an increase in the diagnosis of all brain tumors [48], which may contribute to our finding. Another cause could be the reclassification schemes introduced by WHO, leading to the appropriate diagnosis being made more frequently. With an estimated 230 cases per year in the US, it is unlikely that a prospective clinical trial will be conducted. Alternatively, larger studies have been published from Asia, and our data shows a greater incidence in Asian/Pacific Islander populations. Therefore, carrying out a prospective study in these countries may be more plausible.

In contrast to classical literature, which states that HPC is slightly more predominant in men [14, 16, 36, 40, 42], our study found roughly an equal incidence between men and women. Two other population-based studies report similar findings [32, 33]. Median age of diagnosis was 54, in contrast to classical studies, which state median age of diagnosis to be in the 5th decade of life [14, 16, 23, 40, 42]. Our study confirms prior findings of age [32, 33], posterior fossa location [16, 17, 33, 36], and malignant behavior [12, 14, 22] being significant prognostic factors.

It is important to recognize the limitations of this study. This study is retrospective, and can therefore, of course, only prove association and not causality. Our study and previous retrospective studies examining RT in HPC have been susceptible to the immortal time bias, in which one treatment group, by definition, cannot be exposed to an outcome for a span of time [49]. For example, in our cohort, any patient that died after surgery and before RT would automatically be classified as not receiving surgery. We attempted to overcome this bias by eliminating patients that died within 3 months of diagnosis in a secondary analysis. Additionally, population-based studies have their own limitations. For example, a large proportion of patients in our study had an unknown extent of resection. It is unlikely that these patients had data that was missing at random. To account for this bias, we also reported outcomes in this subgroup of patients. The “unknown extent of resection” group included surgeries with “local excision of tumor/excisional biopsy”, which could not be coded as STR or GTR. Thus, many of these tumors may have been smaller or more surgically accessibly, accounting for the associated increased survival.

Recently, the appropriateness of using population-based studies to examine treatment-related outcomes has come into question. A study comparing SEER with SEER-Medicare data found that SEER was 80% sensitive in capturing radiation treatment [35]. However, agreement over radiation was overall still very high, at 91%. Therefore, we find it unlikely that this small discrepancy has caused significant misinterpretation of our data. Furthermore, concurrence between SEER and SEER-Medicare radiation differed significantly depending on the organ-site in question. The study provided no estimate of agreement between the two datasets in the context of brain tumors, which are more likely to be treated at a specialty center and therefore may have more accurate follow-up. Treatment-related outcomes can be studied in cancer registry data as long as researchers carefully consider strengths and potential limitations [29].

Population-based studies offer several advantages. The study populations are more homogenous and less likely to be subjected to systematic bias [29]. For example, Rutkoski et al. postulated that patients with more aggressive or difficult to treat tumors may be more likely to receive PORT [36]. In our own previously published institutional data, radiation was associated with decreased survival but also associated with invasiveness, WHO grade III, and STR tumors [15]. This current study showed that the treatment variable was confounded by tumor type, histological behavior, location, and size. Smaller case series may lack the power to account for these variables.

All current literature on SFT/HPC of the CNS is derived from retrospective data. Despite a lack of definitive evidence, radiation was administered to almost half of the patients in our series and has been recommended in the routine management of CNS HPC. In the absence of prospective trials, it is important to consider all available data to shed light on optimal treatment in this rare tumor type. Due to the inherent biases in retrospective studies and the recent controversy surrounding the use of population-based registries to study outcomes, we cannot recommend routine administration of radiation to patients with GTR based on the data from this study alone. This study does, however, when combined with existing literature, provide compelling evidence towards the use of adjuvant radiation in CNS SFT/HPC after GTR and warrants further investigation through a prospective clinical trial.

Conclusion

This study is the largest single-patient series on CNS SFT/HPC to date and the first population-based study to examine treatment-related outcomes under the new WHO classification scheme. Population-based registries are a powerful resource that can provide information on incidence and prognostic factors. They are especially useful in rare tumors that are difficult to otherwise study. PORT may improve survival, even in the setting of GTR. This study compliments the existing literature on the role of radiation in SFT/HPC of the central nervous system, and together they warrant further evaluation through a prospective clinical trial.