No previous international study has systematically addressed the histopathological evaluation of ependymomas in the manner reported here, providing data on the review of 229 intracranial ependymomas from children entered into 4 European trials. The review was conducted by 5 neuropathologists, all with specialist experience in the field of pediatric neuro-oncology. The final phase of the study (phase 3) employed a novel histopathological grading scheme. This represents a prescribed application of the WHO classification and was derived by consensus from both the pathologists' experience of ependymomas and a joint review of tumors from one of the four trials (CNS9204). Key aims of the study were: (i) to assess whether discussion surrounding the conception of the new grading scheme and its principles could be used to improve concordance on grading among pathologists, and (ii) whether any pathological variable, either grade itself or the status of one of four histopathological features, was associated with outcome in the setting of formal ependymoma clinical trials with different therapeutic approaches.
Our study was prompted by lack of consensus on how to grade childhood intracranial ependymomas; a huge discrepancy exists between the ratio of grade II:III tumors across the literature, and there is considerable scepticism as to whether grading intracranial ependymomas has clinical utility . The 2007 WHO classification distinguishes the anaplastic (grade III) from classic (grade II) ependymoma on the basis of "high mitotic activity, often accompanied by microvascular proliferation and pseudopalisading necrosis" . This reflects a general principle of the pathological assessment of gliomas - that the identification of 'anaplastic' features, such as increased cell density, mitotic activity, microvascular proliferation, and necrosis, can be used to derive a clinically useful grade. In diffuse astrocytic tumors, these features are progressively acquired with increasing grade (fibrillary astrocytoma, grade II - anaplastic astrocytoma, grade III - glioblastoma, grade IV) and are recognized prognostic indicators [17, 1]. Reinforcing the biological relevance of histopathological grading, astrocytoma progression is associated with the acquisition of specific genetic abnormalities . In contrast, anaplastic ependymomas tend to present de novo; it is uncommon for recurrent ependymomas progressively to acquire an anaplastic phenotype, and any genetic basis for this phenomenon has not yet been convincingly demonstrated . In addition, the presence of anaplastic features across an ependymoma is notoriously variable in magnitude and extent, potentially making evaluation of these features difficult and subsequent grading subjective. For example, a pathologist may be faced with a small focus of microvascular proliferation or pseudopalisading necrosis in a tumor devoid of mitotic activity and with a low cell density. Should this discovery prompt a diagnosis of anaplastic ependymoma (grade III), or should the dominant grade II phenotype prevail?
In phase 1 of this study and before discussing the grading of ependymomas, the five study neuropathologists showed only fair concordance for grade among the group, while showing individual consistency across trials. If ependymomas are particularly difficult tumors to grade, it is surprising that the levels of inter-observer concordance recorded in this study are not far removed from those reported for other gliomas. Assessing astrocytomas and oligodendrogliomas, Coons and colleagues reported an initial 4-reviewer concordance on grade of 52%, which compares with 51% for 5/5 consensus on grading in CNS9204 in this study . Mirroring our experience, concordance improved over successive reviews, as their pathologists discussed possible explanations for discrepancies and developed criteria to aid grading. Grade was assigned according to the status of the same histopathological variables used in the present study, among which microvascular proliferation proved hardest to evaluate in both studies, with lower levels of agreement on its status than for other histopathological features.
After discussing the problems of grading ependymomas and devising a novel grading scheme (phase 2), our study pathologists assessed ependymomas from two trial cohorts for a second time at an interval of just over one year (phase 3). Concordance on grading was notably improved at this time, though it was apparent that some pathologists altered their practice to accommodate the new scheme more than others. This outcome does not necessarily suggest that the new grading scheme presented here is better than the WHO classification, to which the neuropathologists were working in phase 1 of the study, just that agreement to work to a scheme in a prescribed manner results in increased concordance. However, one corollary of the improvement could be that it is easier to grade ependymomas consistently using a more detailed and prescribed scheme than the current WHO classification.
Various clinical variables have been associated with outcome in trial cohorts of children with intracranial ependymoma. These include age at presentation, tumor location, and extent of surgical resection [9, 4, 15, 20–24]. In addition, there is undoubted evidence to indicate the benefits of radiotherapy [5, 3, 14]. A trend towards shorter PFS and OS in the infant cohorts (SFOP/CNS9204) was observed, but in the present study a significant positive association between age and OS (but not PFS) was observed only among children from the CNS9204 cohort. Infratentorial tumor location was significantly associated with poorer OS (but not PFS) only among children from the SFOP cohort. Extent of surgical resection, which has been a proven prognostic indicator in most studies of pediatric ependymoma [22, 23], was associated with outcome in only two (SFOP/AIEOP) out of four of the present trial cohorts, and it may be relevant that the proportion of completely excised tumors in these trial cohorts (SFOP = 65%; AIEOP = 69%) is greater than in either the CNS9204 (51%) or CNS9904 (39%) trial.
The study design enabled us to examine potential associations between outcome and multiple histopathological features. With assessments from five pathologists, it was also possible to analyze the relationship between outcome and strength of consensus among pathologists on grade or the status of individual histopathological features. Adjusted for extent of surgical resection, strength of consensus on grade and on each of the histopathological features was significantly associated with PFS in the AIEOP cohort, but in none of the other cohorts. When individual pathologists' calls on these variables were reviewed; grade III, high cell density, high mitotic activity, presence of microvascular proliferation, and presence of necrosis were all significant adverse prognostic indicators for the AIEOP cohort, but not for SFOP or CNS9904. In the setting of satisfactory concordance on histopathological interpretation, our results suggest that grading ependymomas might have clinical utility either in older children (versus infants) or in children that have received radiotherapy immediately post-surgery (usually non-infants). The latter conclusion is supported by an association between ependymoma grade and outcome in an extensive study of children of all ages, including infants, that were treated with radiotherapy soon after surgery . However, a major caveat from our study involves the lack of a similar association in children from the CNS9904 cohort. These discrepant findings might be related to different proportions of completely versus incompletely resected tumors in the AIEOP (69% vs. 31%) and CNS9904 (39% vs. 61%) cohorts. There are few fundamental differences between studied cohorts to explain our data; there was a significant association between extent of surgical resection and outcome in the AIEOP and SFOP but not the CNS9204 and CNS9904 series, and irradiation as a first line treatment was incorporated into the AIEOP, but not SFOP, trial.
Our data also suggest that a more robust assessment of ependymomas for therapeutic purposes may await the application of a more sophisticated combination of histological and molecular approaches, but while markers of tumor cell proliferation, e.g. Ki-67 immunolabeling [9, 26–28], and a few molecular abnormalities, e.g. copy number gain across chromosome 1q and ERBB2/4 receptor expression [29–32], have been proposed as prognostic indicators, insufficiently unambiguous data are available from large patient cohorts for the confident creation of such a scheme. It is also possible that continuing difficulties with finding prognostic markers for this tumor reflect a misplaced view of ependymomas as homogeneous across the neuraxis, a principle implicit in the WHO classification, which does not differentiate between supratentorial and posterior fossa tumors. However, both data showing that ependymomas at these two sites are characterized by distinct gene expression profiles and identification of a 'vascular' variant in a mainly supratentorial location make the case that ependymomas from these sites should probably be appraised separately [33, 34].