High failure rate in spinal ependymomas with long-term follow-up

Abstract

Data on spinal ependymomas are sparse, and prognostic factors remain controversial. The primary aim of this study is to review a historical cohort, with large patient numbers and long follow-up, and provide estimates of time to progression (TTP) and survival after progression. As a secondary aim, we assess the effects of potential prognostic variables. Thirty-seven patients with spinal cord ependymomas received postoperative radiation therapy from 1955 to 2001. The influences of radiation dose, extent of resection, Karnofsky performance score, tumor location, and multifocality were assessed in univariate analyses by using the Cox proportional hazards model. The median follow-up for patients who did not fail was 121 months (range, 8–312 months). Kaplan-Meier estimates of 5-, 10-, and 15-year percentage progression free are 75% ± 7.4%, 50% ± 9.1%, and 46% ± 9.3%, respectively. Median TTP, for those who recurred, is 68 months (range, 2–324 months), with 12 of 21 failures occurring after five years. Of the prognostic factors examined, only greater extent of resection significantly correlated with longer TTP (P = 0.02). Local relapse rates for spinal ependymomas are higher than previously cited, with a large proportion of failures occurring more than five years after diagnosis. Extensive surgical resection correlates with longer time to recurrence, and we thus recommend maximal excision while avoiding surgical morbidity. The overall high rate of recurrence leads us to recommend radiation to doses of 45 to 54 Gy for all patients who do not have gross total resections, and long, close follow-up.

Spinal ependymomas are relatively rare tumors, representing less than 5% of all gliomas. They are, however, the most common intramedullary spinal cord tumors, comprising more than 60% of all spinal cord malignancies (Karlsson and Brady, 1987; Slooff et al., 1964). Compared with intracranial ependymomas, spinal ependymomas are less prevalent, occur in a younger population, and exhibit a better prognosis (Henson, 2001). They thus constitute a unique clinical entity and require their own algorithms for clinical management.

Although more than 15 retrospective studies have examined spinal ependymomas, most are hampered by small patient numbers, lack of statistical methods for analysis, or short follow-up. Studies have been published that range from 15 to 80 patients, with approximately half of these studies containing fewer than 30 subjects. We thus consider our study, with 37 patients, as relatively large compared with prior studies.

As a primary goal, we sought to assess the overall failure rate and 5-, 10-, and 15-year percentage progression free (PPF)³ estimates with a relatively long follow-up period. As a secondary goal, we attempted to determine the effect of prognostic factors such as extent of resection, Karnofsky performance score (KPS), tumor multifocality, tumor location, and radiation dose on outcome. The effect of these prognostic factors is, for the most part, still being debated in the literature. For example, whereas most studies have demonstrated that more extensive resections correlate with improved outcome (Asazuma et al., 1999; Fischer and Mansuy, 1980; Hanbali et al., 2002; Sonneland et al., 1985), at least one large, well-controlled study with long follow-up failed to reveal any effect of extent of surgery (Waldron et al., 1993). Thus, key questions remain in the management of spinal ependymomas. The primary aim of this study was to review a historical cohort of spinal ependymoma patients treated with radiotherapy and to capitalize on large patient numbers and long follow-up to provide estimates of time to progression (TTP) and survival after progression (SAP). As a secondary aim, we assessed the effects of potential prognostic variables and attempted to address key issues of controversy that persist in the current literature.

Materials and Methods

Patient Characteristics

This study was approved by the Institutional Review Board at the University of California, San Francisco. Between 1955 and 2001, 37 patients with spinal ependymomas received postoperative radiation therapy at the University of California, San Francisco. Patient characteristics and treatment details are presented in Table 1. Median age was 40 years, and 31 patients were adults, older than 18 years of age. Five patients had KPS values of <70, and the remaining had scores of >70. Thirty-three patients had low-grade tumors, and four had high-grade features. Of the low-grade tumors, seven were myxopapillary, 20 were cellular, and three were low-grade ependymomas, not otherwise specified. Prior to radiation therapy, 13 patients had only biopsies, 20 underwent subtotal resections (STRs), and four had gross total resections (GTRs). All patients in this study received radiotherapy at the University of California, San Francisco. The range of radiation doses used was relatively narrow, with all but two patients receiving between 45 and 54 Gy and a median dose of 50.4 Gy. Radiation therapy began one to two months after diagnosis, consisting of 1.5- to 2.0-Gy daily fractions, administered over five to six weeks. Nearly one third of patients received 50.4 Gy in 1.8-Gy, once-daily fractions (12 patients).

Table 1.

Patient characteristics and treatment details

Variables	Number (%)*
Number of patients	37
Median age (range), years	40 (2–66)
Median KPS (range)	80 (30–90)
Female:male	14:23
Focality of tumor
Unifocal	30 (81)
Multifocal	7 (19)
Tumor location
Lumbosacral/cauda equina	19 (53)
Non-cauda equina	18 (47)
Median radiation dose (range), Gy	50.4 (45–59.4)
Radiation fields
Craniospinal irradiation	4 (11)
Whole spine only	2 (5)
Tumor volume plus margin	31 (84)
Extent of resection at diagnosis
Gross total resection	4 (11)
Subtotal resection	20 (55)
Biopsy only	13 (34)
Tumor grade
Low	33 (89)
High	4 (11)

Abbreviation: KPS, Karnofsky performance status.

^*Data are numbers with percents in parentheses except as noted for median and ratio values.

Fields of irradiation consisted of craniospinal in four patients, whole spine without cranial irradiation in two patients, and tumor plus margin in 31 patients. The two patients who received whole spine irradiation had multifocal disease; one had neurofibromatosis type II and prior history of a meningioma. This patient was the only patient in our cohort diagnosed with neurofibromatosis. Of the four patients who underwent craniospinal irradiation, two had multifocal presentations, one had a high-grade tumor, and one patient had neither high-grade histology nor multifocal disease but received craniospinal irradiation at the discretion of the treating physicians. The imaging modality used to make the original diagnosis was documented for 32 patients and consisted of CT and/or MRI (most commonly both) in 16 patients, and myelogram without CT or MRI in 16 patients.

Disease relapse was defined as recurrent/progressive symptoms or radiographic evidence of disease recurrence. Extent of resection was determined by a combination of operative reports and postoperative imaging studies. Of note, the method for estimating TTP was slightly different from that used to estimate overall survival (OS). In determining TTP, for patients without documented progression at the last clinic visit, if a death certificate was obtainable and did not document disease progression at the time of death, the TTP was estimated to be equal to survival time. If, however, a death certificate was unobtainable or if progression occurred before the time of death, then the time period before evidence of progression was the TTP. Those patients who did not progress were censored at their last follow-up date. In contrast, to calculate OS, any documentation of the date of death was used, regardless of the cause of death. For nine of the 19 patients who died, death certificates were obtained through the National Death Index. For the other 10 patients, details of patient deaths were obtained by contacting family members and the referring physician and through documentation in medical records.

For the majority of patients, follow-up was carried out at the University of California, San Francisco, and progression was documented by imaging studies and clinic visits. For the minority of patients who were followed primarily at outside institutions, conclusive documentation of disease progression was acquired through imaging studies or written communication of clinical progression.

Statistical Analysis

The primary end point of our study was PPF assessed at 5, 10, and 15 years by the Kaplan-Meier method. Secondary end points were 5-, 10-, and 15-year OS rates, as well as SAP. We limited our analysis to potential prognostic factors supported by current literature, including extent of resection, age, tumor location, KPS, and radiation dose. The impact of each variable was assessed by using a Cox regression model, and hazard ratios with 95% confidence intervals were calculated with the statistical software SPSS (SPSS, Inc., Chicago, Ill.). We chose to focus on TTP and SAP because after decades of follow-up many of our patients died of causes entirely unrelated to their tumors or treatments, and thus OS provided limited information about the natural history or appropriate treatment of spinal ependymomas. In addition, our cohort included only four patients with high-grade spinal ependymomas, which precluded an analysis of the relationship between prognosis and tumor grade; however, all univariate analyses were performed both with and without the four patients with high-grade tumors in order to ensure that these four patients were not driving any of our findings.

Results

Time to Progression and Survival After Progression

Median TTP and OS rates were 82 months (range of time to events, 2–324 months) and 180 months (range, 8–492 months), respectively (Figs. 1A and 1B), with a median follow-up of 121 months for patients who did not fail (range, 8–312 months) and 192 months for surviving patients (range, 25–364 months).

Fig. 1A.

Time to progression. Assessed at 5, 10, and 15 years, PPF estimates were 75 ± 7.4%, 50 ± 9.1%, and 46 ± 9.3%, respectively. Twelve of 21 failures occurred after five years.

Fig. 1B.

Overall survival. Assessed at 5, 10, and 15 years, OS rates were 83 ± 6.2%, 74 ± 7.5%, and 61 ± 8.5%, respectively.

Assessed at 5, 10, and 15 years, PPF estimates were 75% ± 7.4, 50% ± 9.1, and 46% ± 9.3, respectively, and OS rates were 83% ± 6.2, 74% ± 7.5, and 61% ± 8.5, respectively. In determining TTP, data was collected on 14 of 37 patients at 10 years, and 10 of 37 patients at 15 years. Twenty-one of 37 patients failed radiation therapy, either clinically or radiographically. The median time to failure was 68 months, with 12 of 21 failures occurring after five years. The median SAP was 88 months (range, 0–256 months), as depicted in Fig. 2.

Fig. 2.

Survival after progression in the 21 patients who recurred. The five-year OS rate for those patients who recurred was 65 ± 10.7%.

Notably, of the 17 patients who were diagnosed with spinal ependymomas before 1980, 11 patients (64.7%) failed treatment. Among patients who were diagnosed after 1980, 10 (50%) failed treatment. Analyzing three time periods of treatment (1955–1970–1971–1985, and 1986–2001) revealed no relationship between treatment period and progression-free survival (P = 0.272). Salvage therapy consisted most commonly of a combination of chemotherapy and re-resection.

Three additional findings are of importance about specific small subsets of patients within the cohort. First, three of the four patients with high-grade tumors failed therapy, at 6, 68, and 98 months. Second, in the subset of four patients in which GTR was achieved, only one patient failed therapy and did so at six months. It should be noted, however, that this patient also had an anaplastic histology. Finally, of the 21 patients who failed radiation therapy, 17 had unifocal tumors, and 15 of these patients failed within the original treatment area. Of the six patients with extended field irradiation, four failed treatment.

Variables Predicting Clinical Outcome

Current literature points to several variables as predictors of clinical outcome for patients with spinal ependymomas. We performed univariate analyses to evaluate whether age at time of diagnosis, KPS, tumor location, multifocal presentation, radiation dose, or extent of surgery was associated with TTP. Table 2 depicts the hazard ratios and P values for the five variables tested. Note that extent of resection was the only variable that correlated with TTP (P = 0.006), with more aggressive resections associated with improved outcome. To ensure that the four patients with high-grade histology were not driving this association, the analysis was repeated without these four patients, again revealing a significant correlation between greater extent of resection and improved TTP (P = 0.001). Neither imaging modality nor histologic type affected TTP (P = 0.61, and P = 0.33, respectively).

Table 2.

Correlation of tested variables with time to progression as determined by univariate analysis

Variable	Hazard ratio (95% CI)	Pvalue
Karnofsky performance status	1.0 (0.97–1.03)	0.99
Multifocal disease	1.25 (0.36–4.4)	0.72
Tumor location*	1.7 (0.61–4.8)	0.35
Total radiation dose (Gy)	1.0 (0.99–1.002)	0.75
Extent of resection**	0.34 (0.15–0.75)	0.006

^*Tumor location was classified with the following values: cervical = 0, thoracic = 1, and lumbosacral = 2.

^**Extent of resection was classified as follows: biopsy = 0, subtotal resection = 1, and gross total resection = 2.

Discussion

We present here a relatively large cohort of spinal ependymomas with long follow-up, two characteristics that allow conclusions about a rare and indolent disease. Although our study is not the largest reported in the literature, it has unique aspects that bring new insight into the true failure rate of spinal ependymomas and potential prognostic factors. We report 10- and 15-year PPF estimates of 50% and 46%, respectively, as well as an overall failure rate of 57% (21 of 37 patients), with 12 of 21 patients failing at least five years after diagnosis (median time to recurrence of 68 months). These failure rates are higher than the 20% to 50% rates seen in previous large studies (Cervoni et al., 1994; Sonneland et al., 1985; Waldron et al., 1993; Whitaker et al., 1991).

A higher failure rate than previously reported is somewhat surprising because our results are conservative and may in fact underestimate the true number of failures that occurred. The results may underestimate recurrences because patients whose death certificates indicated causes of death unrelated to tumor without prior documented disease failures were estimated to be progression free until death. It is possible that some patients experienced tumor recurrences but were successfully salvaged between their last disease evaluation and time of death.

Several explanations exist for higher recurrence rates reported herein compared with some previous reports. Whitaker et al. (1991) reported a 10-year progression-free survival rate of 68%, but with a follow-up period of only 70 months compared with our median follow-up of 121 months. Sonneland et al. (1985) reported failure rates of 10% to 20% but included only myxopapillary ependymomas. However, the reasons underlying lower failure rates reported by Cervoni et al. (1994) and Waldron et al. (1993) are less clear. Both studies have long median follow-up times of 204 and 131 months but observed failure rates of only 32% and 18%, respectively. Furthermore, the median time to recurrence in the study by Waldron et al. was two years, with nine of 11 recurrences occurring within three years, in contradistinction to the 12 of 21 failures in our study that occurred at least five years after diagnosis.

Several factors may underlie the higher recurrence rates observed in our study. Although we could not demonstrate a correlation between historical treatment period and outcome, our study included patients who were diagnosed in years up to and including 2001. Patients in such recent periods likely had more frequent follow-up examinations with more sophisticated imaging technologies that may have revealed failures potentially missed in older studies. In addition, myxopapillary histology represents a larger proportion of histologies in a study by Cervoni et al. (1994) than in our current cohort (46% vs. 27%, respectively). That said, differences in myxopapillary histology cannot explain higher recurrence rates reported herein compared with those reported by Waldron et al. (1993), as both studies have nearly identical percentages of myxopapillary tumors, presumably because both studies included only patients who received postoperative radiation therapy. Finally, because we were unable to assess the proportion of patients for whom the determination of recurrence or progression was based solely on clinical rather than radiographic grounds, sequelae such as radiation necrosis might have been mistaken for late recurrences, which could have led to overestimating of recurrence rates. Of note, differences in age of diagnosis cannot explain higher recurrence rates reported herein, as virtually all published studies report similar age ranges, reflecting the incidence of spinal ependymomas in the population.

The study reported herein is unique in that the cohort size is relatively large, follow-up time is substantial, and both Kaplan-Meier and univariate Cox regression analyses are used to evaluate the data. This is in contrast to other retrospective studies of spinal ependymomas. Cervoni et al. (1994) reported what is possibly the largest cohort of spinal ependymoma patients in the literature, 78 subjects, but examined risk factors for recurrence using only a chi-squared statistical test. Mork and Loken (1977) assessed OS in 53 patients but did not utilize additional statistical analyses.

To further evaluate the prognosis of patients who failed treatment, we examined survival after failure in the 21 patients with progression or recurrence. Interestingly, three years after failure, survival was approximately 60%, but beyond three years mortality increased more slowly, with some patients surviving up to 20 years. These findings may reflect the efficacy of salvage therapy and/or the indolent nature of spinal ependymomas.

Prognostic factors for spinal ependymomas remain controversial. Although some controversy persists regarding the prognostic importance of extent of resection (Asazuma et al., 1999; Fischer and Mansuy, 1980; Guidetti et al., 1981; Hanbali et al., 2002; Sonneland et al., 1985; Wen et al., 1991), most studies with large patient numbers (more than 30) confirm the importance of extent of resection. Whitaker et al. (1991) found that extent of resection influenced 5- and 10-year progression rates as well as OS. Similarly, Cervoni et al. (1994) reported that extent of surgery affected the likelihood of local failure. Guidetti et al. (1981) recommended postoperative radiation only after biopsies or STRs but not after GTRs because their study of 48 patients with intramedullary tumors revealed improved survival for patients with GTRs, compared with those with biopsies or STRs. Hanbali et al. (2002) restated these recommendations in a recent study that documented only one failure among 26 patients with radical GTRs.

One notable exception is the above-mentioned study by Waldron et al. (1993), which detected no influence of extent of resection. The authors attributed lack of effect of surgery to selection bias of patients undergoing GTRs yet still referred for irradiation. Specifically, patients with GTRs that are referred for radiation therapy likely represent a group in which there is uncertainty regarding the completeness of excision or poor prognostic features. Although this argument certainly has merit, a very similar cohort of patients reported herein disclosed a significant relationship between extent of resection and TTP. However, because all of the patients in our study received postoperative radiation, we cannot arrive at firm conclusions regarding the utility of radiation after GTRs.

Our study could not shed light on the prognostic significance of tumor grade or location. The small number of patients with high-grade ependymomas limited the power to detect any statistically significant effect of grade on prognosis. Similarly, neither myxopapillary nor cellular histological subtypes predicted clinical outcome. Published studies debate the prognostic significance of tumor grade in spinal ependymomas. Mork and Loken (1977) found that myxopapillary subtype was associated with better OS compared with other histologic subtypes, and Waldron et al. (1993) concluded that high-grade tumors are associated with decreased relapse-free survival. In contrast, Cervoni et al. (1994) and Schiffer et al. (1991) found no association between tumor grade and risk of recurrence, and these authors concluded that histologic criteria useful for defining anaplasia in other tumors are not useful in ependymomas.

In assessing the effect of tumor location, our study failed to reveal an association between site within the spine and prognosis, likely because of the small number of completely resected tumors (4 of 37), regardless of location. Previous studies have reported that tumor location affects prognosis, potentially because some spine regions are more amenable to complete surgical resection. For example, myxopapillary tumors demonstrate favorable PPF rates and appear more amenable to complete resections, perhaps because of their propensity to occur in the filum terminale (Mork and Loken, 1977; Whitaker et al., 1991).

Radiation doses and fields constitute key factors relevant to radiation recommendations for spinal ependymomas. Analyses of radiation doses for spinal ependymomas are unlikely to document a dose response, as prescribed doses in virtually all published studies fall within a narrow range. The vast majority of patients receive radiation doses approaching spinal cord tolerance, and given our high local failure rate, we continue the practice of administering doses guided by normal tissue tolerance.

Few studies have examined the effect of extent of radiation fields on clinical outcome. However, as most tumors tend to fail locally, extended radiation fields are unlikely to offer benefit. Indeed, Whitaker et al. (1991) found that 75% of failures were local; two of three patients who relapsed intracranially did so after receiving craniospinal irradiation, and thus no benefit was observed for extended fields of radiation. Our study supports these findings. Although we could not formally assess the influence of radiation field size on prognosis because of the limited number of patients with extended fields (n = 6), almost all patients recurred locally, and four of six patients with extended field radiation still failed treatment. We therefore recommend localized postoperative irradiation in all patients with unifocal spinal ependymomas.

Conclusions

This historical cohort of a relatively large number of spinal ependymoma patients with long follow-up demonstrates that local relapses occur more often and later than previously cited and that aggressive surgical resection correlates with improved outcome. Thus we recommend excision aimed at maximizing extent of resection while avoiding surgical morbidity. In all patients with STR or biopsy, we recommend postoperative radiation to doses of 50 to 54 Gy, and close, long follow-up.