The Misclassification of Diffuse Gliomas: Rates and Outcomes

The molecular characterization for a spectrum of cancer types has revolutionized oncologic diagnostics and therapeutics and is increasingly becoming a critical component of clinical care for patients with cancer. In 2016, the World Health Organization (WHO) histologic classification of tumors of the central nervous system (CNS; 4th edition) was revised to synthesize our growing understanding of the molecular basis of CNS tumors, principally made possible by advances in next-generation sequencing (NGS) and gene expression analysis, with histology into a diagnostic and prognostic classification scheme that is more objective and accurate (1). The changes have been particularly groundbreaking for diffuse gliomas (i.e., gliomas with a diffusely infiltrative pattern of growth) for which IDH1/IDH2 gene mutational status and the presence or absence of chromosomal whole-arm codeletion of 1p and 19q has important prognostic and therapeutic implications (2). The accurate diagnosis of diffuse gliomas faces several challenges, including an overlapping spectrum of histomorphologies and substantial intratumoral histologic heterogeneity. These challenges can commonly be compounded by tumor under sampling and the limited specimen available for evaluation due to the surgical complexity of safely accessing many of these tumors. As a result, relying on histology alone for the classification of diffuse gliomas can lead to erroneous diagnoses. In addition to CNS tumors, molecular markers are now critical for the accurate diagnosis and treatment of hematopoietic and lymphoid, soft tissue and bone, renal, melanoma, lung, and various other malignancies; each accompanied by major recent revisions to their WHO classification schema (3–5).

In an effort to systematically collect cancer patient data in the United States, including that of patients with glioma, cancer registry reporting standards were developed and employed by North American Association of Central Cancer Registries, including the American College of Surgeons' National Cancer Database program (NCDB), the NCI's Surveillance, Epidemiology, and End Results program, and the Center for Disease Control's National Program for Cancer Registries and Central Brain Tumor Registry of the United States (6–8). These cancer registries, crucially, have vastly improved our understanding of cancer epidemiology, but have lagged in incorporation of key molecular drivers for a range of cancer types. U.S. cancer registries encode cancer diagnoses using WHO/IARC International Classification of Diseases for Oncology (ICD-O) v3, the codes and histologic definitions of which were defined in the WHO International Histological Classification of Tumours “Blue Books” published through 2010 (i.e., through the 4th edition; ref. 9). For CNS tumors, the 4th edition was published in 2007 and classified diffuse gliomas solely based on their histologic appearances (10). Consequently, cancer registry data cannot presently differentiate between isocitrate dehydrogenase (IDH)-mutant and IDH–wild-type diffuse gliomas. In addition, many of the key clinical trials of diffuse gliomas, as well as studies using historical cohorts, relied on older histology-only classification schemes. Nevertheless, the magnitude of potential coding errors and diagnostic misinterpretations in these databases are not fully appreciated or taken into consideration (11–27). Herein, we examine the accuracies of cancer registry–encoded diffuse gliomas by WHO grade and 1p/19q status, in addition to IDH, ATRX, and p53 status; we explore the associations between misclassification, tumor characteristics, and survival outcomes; and we demonstrate the important need for incorporating molecular data into registries, historical cohorts, and clinical trials of patients with cancer.

The NCDB, a hospital-based nationwide cancer database, comprises >70% of newly diagnosed cancers in the United States. Patients with newly diagnosed diffuse gliomas between 2010 and 2015 were identified by ICD-O-3 histologic codes coupled with a “/3” malignant behavior code as presented in Table 1, inclusive of all brain site codes (i.e., 71.0–71.9). Patients were excluded if previously diagnosed with cancer or if they were diagnosed at an index institution but received treatment entirely elsewhere. Primary brain-specific factors included the LOH/deletion of chromosome arms 1p and 19q (first encoded in 2010) and WHO grade. Cases were designated as 1p/19q-codeleted if both arms were coded for LOH/deletion, although the methodology utilized for this determination was not recorded in the NCDB.

To evaluate the accuracy of diagnoses based on additional molecular alterations, 150 patients with diffuse glioma (25/year, from 2010 to 2015) were randomly queried from the cancer registry–submitted data from each of three tertiary care institutions (Brigham and Women's Hospital, Boston, MA; Dana-Farber Cancer Center, Boston, MA; and Massachusetts General Hospital, Boston, MA) using ICD-O-3 codes. The neuropathologist-assigned integrated diagnosis and WHO grade were evaluated for each of the 450 patients, along with the status of 1p/19q, IDH, ATRX, and p53, as determined by IHC and/or molecular assays (e.g., NGS panels, array comparative genomic hybridization, and fluorescence in situ hybridization).

The clinicopathologic factors of age at diagnosis, histology, tumor location, 1p/19q status, and WHO grade were summarized and then compared by χ² test, t test, and ANOVA as appropriate. The primary outcome was concordance rate between the histologic classification of diffuse gliomas and their encoded WHO grade and 1p/19q molecular status. Unadjusted differences in overall survival (OS) were estimated by Kaplan–Meier methods and compared with log-rank tests. The NCDB excludes survival data for patients diagnosed in the final year of the dataset due to limited follow-up, which for this release was 2015. OS was measured from date of diagnosis with the endpoint assigned as date of death, with patients censored at the date of most recent follow-up. The end date for follow-up in this release was December 31, 2015. A secondary outcome was the concordance rate between the histologic classification of institutional cases included in the cancer registry and their corresponding neuropathologist-assigned integrated histologic and molecular diagnoses. Statistical analyses were conducted using STATA (v. 14.2, StataCorp), with two-sided P < 0.05 denoted as significant. This study was approved by the Partners HealthCare institutional review board (2015P002352).

From 2010 to 2015 NCDB data, there were 74,718 patients newly diagnosed with diffuse gliomas following exclusion (Table 1). For ICD-O-3–encoded diffuse oligodendrogliomas and astrocytomas (i.e., grade II), only 86.3% (n = 2,742) and 65.7% (n = 2,748) were coded as WHO grade II (Table 2). In anaplastic oligodendrogliomas and astrocytomas (i.e., grade III), the concordance rates between ICD-O-3 coding and WHO grade improved to 90.0% (n = 1,361) and 90.4% (n = 5,090), respectively. The age at diagnosis and OS significantly differed between WHO grades for each ICD-O-3–encoded oligodendroglioma and astrocytoma group: diffuse oligodendrogliomas and astrocytomas that were in fact coded as WHO grade II demonstrated significantly improved OS and younger age at diagnosis than those that were coded as WHO grade III (both P < 0.001). The highest concordance rate (98.7%, n = 39,828) between ICD-O-3 coding and WHO grade was observed in glioblastomas (GBMs, grade IV).

The 2016 WHO recognized that 1p/19q-codeletion in combination with IDH mutation is a pathognomonic feature of oligodendrogliomas, with clinical guidelines therefore recommending molecular testing for all diffuse gliomas. For cases with known 1p/19q status, only 74.4% (n = 1,001) and 78.8% (n = 523) of WHO grade II diffuse oligodendrogliomas and WHO grade III anaplastic oligodendrogliomas (as encoded by both ICD-O-3 histology and WHO grade), respectively, were truly 1p/19q-codeleted; suggesting that the remainder were likely inappropriately encoded astrocytomas or nondiffuse glioma subtypes (Table 3). Correspondingly, the histologically encoded oligodendrogliomas with confirmed 1p/19q-codeletion were associated with significantly younger age at diagnosis, greater predilection for the frontal lobe, and improved OS than histologically encoded “oligodendrogliomas” with retained 1p/19q (all P < 0.05). In our multi-institutional cohort, when integrating 1p/19q, IDH, ATRX, and p53 molecular statuses, 23.1% (n = 6) of registry-encoded WHO grade II diffuse oligodendrogliomas were reclassified: 3.8% as WHO grade II diffuse astrocytoma IDH-mutant and 19.2% as WHO grade IV glioblastoma IDH–wild-type, often with an oligodendroglioma-like histologic component (Table 4). Likewise, 35.0% (n = 7) of registry-encoded WHO grade III anaplastic oligodendrogliomas were also reclassified: 15.0% as WHO grade III anaplastic astrocytoma IDH-mutant, 5.0% as WHO grade III anaplastic astrocytoma IDH–wild-type, 10.0% as WHO grade IV glioblastoma IDH–wild-type (also commonly with an oligodendroglioma-like component), and 5.0% as a high-grade glioma without 1p/19q or IDH status.

In histologically encoded “mixed gliomas” (i.e., oligoastrocytoma), 28.9% (n = 163) of WHO grade II and 36.8% (n = 168) of WHO grade III tumors, respectively, had 1p/19q-codeletion, therefore likely representing oligodendrogliomas; whereas the remaining 1p/19q-retained cases were likely astrocytomas (Table 3). Within our multi-institutional cohort, when incorporating the integrated molecular status, 92.9% (n = 39) of registry-encoded “mixed gliomas” were reclassified as either WHO grade II oligodendroglioma 1p/19-codeleted and IDH-mutant (9.5%), diffuse astrocytoma IDH-mutant (21.4%), and diffuse astrocytoma IDH–wild-type (4.8%), or WHO grade III anaplastic oligodendroglioma 1p/19q-codeleted and IDH-mutant (14.3%), anaplastic astrocytoma IDH-mutant (23.8%), and anaplastic astrocytoma IDH–wild-type (19.0%). Similarly, 47.8% (n = 11) of registry-encoded “malignant gliomas” (i.e., high-grade gliomas) could be reassigned to a 2016 CNS WHO subtype when integrating molecular status.

1p/19q status was reported in only a minority of histologically encoded WHO grade II diffuse astrocytomas and WHO grade III anaplastic astrocytomas, but of these cases, 6.3% (n = 34) and 8.8% (n = 72), respectively, demonstrated 1p/19q-codeletion and would now be reclassified as oligodendrogliomas in the presence of an IDH mutation (Table 3). In addition, applying integrated molecular approaches to our multi-institutional registry-encoded WHO grade II diffuse astrocytomas revealed that 41.7% (n = 15) were IDH–wild-type and 36.1% (n = 13) were IDH-mutant, and 2.8% (n = 1) were in fact BRAFV600E-mutant (IDH–wild-type) pilocytic astrocytomas; whereas for the WHO grade III anaplastic astrocytomas, 60.0% (n = 45) were reclassified as IDH–wild-type and 26.7% (n = 20) were IDH-mutant. Of the encoded GBMs with reported 1p/19q status, 9.4% (n = 227) displayed 1p/19q-codeletion (and would be reclassified as anaplastic oligodendrogliomas if IDH-mutant); these were associated with significantly improved OS compared with the encoded GBMs with retained 1p/19q. Among the multi-institutional registry-encoded GBMs, 3.7% were found to be IDH-mutant, with another 0.9% and 0.4% found to be the giant cell and gliosarcoma variants of GBM IDH–wild-type.

The 2016 WHO made several significant changes to the classification of diffuse gliomas, primarily by integrating IDH1/IDH2 mutation and 1p/19q-codeletion status with histology into a diagnostically and prognostically more accurate scheme (Table 5). The most frequent IDH driver mutation in diffuse glioma, occurring in approximately 90% of cases, is an arginine to histidine missense mutation at codon 132 of the IDH1 gene, which can be reliably detected with an IDH1 R132H IHC stain (28, 29). In IDH-mutant gliomas, 1p/19q-codeletion is pathognomonic for oligodendroglioma. This biomarker and MGMT promoter methylation status are the only brain cancer–specific molecular data encoded in cancer registry data as of 2010. Data on IDH mutation status are notably not yet reported in the U.S. cancer registries, but have started to be collected in 2018. Grading of diffuse gliomas was made on the basis of histologic features such as mitotic activity, tumor necrosis, and microvascular proliferation, with these conventions retained in the 2016 revised 4th edition. Herein, we examine the accuracies of cancer registry–encoded diffuse gliomas by WHO grade and 1p/19q status, and demonstrate how misclassification may impact survival outcome estimates.

Diffuse astrocytoma (WHO grade II) and anaplastic astrocytoma (WHO grade III) are 1p/19q-retained and can be further subclassified on the basis of their IDH status: mutant, wild-type, or not otherwise specified (NOS; Table 4). The IDH status has important prognostic value, as IDH-mutant infiltrating astrocytomas are associated with a better survival, as well as increased sensitivity to chemoradiation (30, 31). The majority of WHO grade II and III infiltrating astrocytomas harbor IDH mutations. These IDH-mutant astrocytomas almost always have concurrent inactivating TP53 and ATRX mutations. The Consortium to Inform Molecular and Practical Approaches to CNS Tumor Taxonomy—Not Official WHO (cIMPACT-NOW) was established in 2016 to translate the 2016 CNS WHO criteria into diagnostic guidelines and recommends that for diffuse gliomas with astrocytoma-like histology, demonstrating IDH1 (R132H) mutation with loss of nuclear ATRX expression and/or strong diffuse p53 expression by IHC is sufficient for rendering a diagnosis of IDH-mutant astrocytoma, without necessitating 1p/19q testing (32–34). If the IHC results or clinicopathologic features are noncanonical for IDH-mutant astrocytomas, then additional testing for alternative IDH mutations, other molecular hallmarks, or chromosomal alterations may be indicated. A subset of tumors that were initially diagnosed as grade II or grade III IDH–wild-type infiltrating astrocytomas can be reclassified after molecular testing as other glioma subtypes, including pilocytic astrocytomas, gangliogliomas, glioblastomas, and diffuse midline gliomas (H3 K27M-mutant; ref. 35). We found that 1p/19q status was only reported in a minority of astrocytomas, but that a significant subset had 1p/19q-codeletion, likely representing miscoded oligodendrogliomas in most examples (i.e., if IDH-mutant). In addition, the integrated molecular testing of multi-institutional registry-encoded astrocytomas demonstrated that the prognostically relevant IDH status varied substantially.

Oligodendroglioma (WHO grade II) and anaplastic oligodendroglioma (WHO grade III) are IDH-mutant and 1p/19q-codeleted (Table 4; refs. 36, 37). In contrast to IDH-mutant astrocytomas, oligodendrogliomas often have TERT promoter mutations, and p53 and ATRX genetic aberrations are not characteristic. The 2016 CNS WHO and cIMPACT-NOW recommend that all diffuse gliomas with oligodendroglioma-like histology or mixed oligodendroglioma–astrocytoma histology be evaluated for 1p/19q-codeletion and IDH status. In clinical trials, IDH-mutant 1p/19q-codeleted oligodendrogliomas are associated with significantly improved response to chemotherapy, as compared with IDH-mutant and IDH–wild-type astrocytomas (38–40). Herein, only 74% and 79% of grade II and III oligodendrogliomas, respectively, were in fact coded as 1p/19q-codeleted. These codeleted cases were associated with significantly younger age, more frequent localization to the frontal lobe, and more favorable OS as compared with the 1p/19q-retained subset of miscoded astrocytomas or other glial tumors. The differences in clinical characteristics and outcome between 1p/19-codeleted and 1p/19q-retained gliomas in this cohort highlight the importance of accurate integration of molecular data with histology. Furthermore, when 1p/19q, IDH, ATRX, and p53 statuses were incorporated, we observed similar rates of reclassification in our multi-institutional cases of registry-encoded oligodendrogliomas.

Importantly, under the 2016 WHO scheme, a diagnosis of oligoastrocytoma should be exceedingly rare, reserved for instances in which the tumor demonstrates distinct areas with histologic and genetic features of astrocytoma and oligodendroglioma, respectively (41, 42). Instead, nearly all diffuse gliomas with mixed astrocytic and oligodendroglial morphology can be reclassified as either astrocytomas or oligodendrogliomas when incorporating molecular and cytogenetic data (43). For example, we found that 29% and 37% of WHO grade II and III histologically encoded “oligoastrocytomas” had 1p/19q-codeletion, therefore likely representing grade II and III oligodendrogliomas, assuming that they also had IDH mutations. In our multi-institutional series, 93% and 48% of registry-encoded “oligoastrocytomas” and “malignant gliomas”, respectively, could be reassigned to a 2016 CNS WHO subtype when integrating molecular status.

In key trials of chemotherapy [e.g., RTOG9802 trial of radiotherapy combined with procarbazine, lomustine, and vincristine (NCT00003375) and RTOG0424 trial of radiotherapy combined with temozolomide (NCT00114140)] and radiotherapy (e.g., EORTC22845 timing trial and NCCTG/RTOG/ECOG dose trial) for grade II diffuse gliomas, tumors were classified only histologically as oligodendroglioma, oligoastrocytoma, or astrocytoma, and 1p/19q status was not addressed in survival analyses (44–47). On the basis of our results, up to 29% of “oligoastrocytomas”, approximately 6% of “astrocytomas”, and only 74% of “oligodendrogliomas” in these trials would have likely represented WHO grade II 1p/19q-codeleted oligodendrogliomas, if also IDH mutant, and thus would have been associated with improved OS, as also suggested by the improved progression-free survival by 1p/19q-codeleted status in EORTC22033-26033 of temozolomide versus radiotherapy (NCT00182819; ref. 38). In addition, our findings suggest that grade II diffuse “astrocytomas” in these trials actually comprised a combination of prognostically distinct WHO grade II IDH-mutant and IDH–wild-type diffuse astrocytomas, as well as suggesting that up to 3% of the diffuse “astrocytomas” may have in fact represent pilocytic astrocytomas. Correspondingly, updated subgroup analyses by 1p/19q status in clinical trials for grade III diffuse gliomas [e.g., EORTC26951 and RTOG9402 (NCT00002569) trials of radiotherapy combined with procarbazine, lomustine, and vincristine] demonstrated improved OS with 1p/19q-codeletion, which was reflected in our results: 79% and 49% 5-year OS for 1p/19q-codeleted and noncodeleted WHO grade III gliomas, respectively (39, 40).

Glioblastoma (WHO grade IV) can also be subclassified by IDH status (Table 4). However, unlike the lower grade infiltrating astrocytomas, most GBMs (i.e., >90%) are IDH–wild-type. IDH–wild-type GBMs frequently have TERT promoter mutations, recurrent copy number alterations including polysomy 7 and monosomy 10 (with loss of PTEN), amplification of EGFR, and loss of tumor suppressors CDKN2A and CDKN2B. IDH-mutant GBMs, representing approximately 4% of registry-encoded GBMs in our cohort, may progress from grade II or III diffuse astrocytomas and are associated with a better prognosis and younger age at time of diagnosis compared with IDH–wild-type GBMs (28, 48, 49). The distinction between IDH-mutant and IDH–wild-type GBMs is not permitted by current cancer registry encoding. We found that only a fraction of encoded “GBMs” included 1p/19q status, of which 9% were 1p/19q-codeleted and were associated with significantly improved OS. These cases may have been previously interpreted within the now-defunct “GBM with oligodendroglial component” category (i.e., the grade IV counterpart of oligoastrocytoma), but would now be more appropriately reclassified as WHO grade III anaplastic oligodendrogliomas when IDH-mutant (50).

In 2018, NAACCR began incorporating all of the 2016 WHO ICD-O coding and definitions for diffuse gliomas, including the new entity diffuse midline glioma characterized by the H3 K27M mutation (Table 4); although incorporation of registry data into databases for public analysis usually lags by approximately 3 years. In addition, IDH status will be collected as a site-specific variable. Together, these crucial updates will permit cancer registries to more accurately stratify patients with diffuse glioma, which in turn will enable clinically relevant evaluations of contemporary glioma epidemiology and management.

The cancer registry–based databases, although extensive in their scope, are constrained by several limitations. For CNS tumors, molecular data are limited only to MGMT promoter methylation status and 1p/19q status, with IDH status notably lacking despite now being essential for accurate diagnosis. To help address this challenge, we included the diagnostic accuracy results for 450 registry-encoded diffuse gliomas from three tertiary care institutions, based on canonical 1p/19q, IDH, ATRX, and p53 statuses. Also, although 1p and 19q LOH variables are included, registry data lack details about the diagnostic methodologies used: for instance, it is known that only whole-arm deletions correlate with the biologically favorable oligodendrogliomas; however, smaller deletions are seen in a subset of astrocytic tumors, which might be construed as a false-positive in FISH assays that only assess one probe per arm. The databases additionally only include data from a patient's initial presentation and initial treatment courses, precluding evaluation of subsequent treatments, progression, recurrence, or metastasis.

Cancer registries are critical to our understanding of the epidemiology, oncogenesis, and treatment of a wide spectrum of tumor types. However, many diffuse gliomas are inappropriately classified, with significant implications for OS estimates. On the basis of 1p/19q, IDH, ATRX, and p53 statuses, the overall rates of misclassified oligodendrogliomas, astrocytomas, and GBMs based on histology-only classification are approximately 21%–35%, 6%–9%, and 9%, respectively. Our rapidly expanding understanding of the genetic landscape of diffuse gliomas highlights the importance of integrating histology with molecular data for diagnosis. Our findings suggest that when compared with historical histology-only classified data, there is the potential for false-positive results in contemporary therapeutic trials of molecularly classified diffuse gliomas. These issues may be true for institutional historical databases, as well as national registry–based sources, and could also contribute to a seemingly positive phase II trial (based on historical comparison) failing at the phase III stage. Critically, the findings from diffuse glioma clinical trials, registry-based studies, and historical cohorts using prior histology-based WHO classification schemes must be cautiously reinterpreted and tempered in light of the integrated diagnoses defined in the revised 2016 WHO. Particularly for institutional historical comparison groups, we recommend that investigators consider using IHC evaluation (and chromosomal/molecular testing if indicated) of archival formalin-fixed, paraffin-embedded samples to appropriately reclassify historical cohorts. Our results highlight the pressing need to design preclinical investigations, clinical trials, and national cancer registries moving forward that incorporate key molecular data across all cancer types.

D. A. Reardon is a consultant/advisory board member for Abbvie, Advantagene, Agenus, Bristol-Myers Squibb, Celldex, EMD Serono, Genentech/Roche, Inovio, Merck, Merck KGaA, Monteris, Novocure, Oncorus, Oxigene, Regeneron, Stemline, and Taiho Oncology, Inc. No potential conflicts of interest were disclosed by the other authors.

Conception and design: J.B. Iorgulescu, T.R. Smith

Development of methodology: J.B. Iorgulescu, T.R. Smith, A.A. Aizer

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.B. Iorgulescu

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J.B. Iorgulescu, M. Harary, T.R. Smith, A.A. Aizer, D.A. Reardon, J.S. Barnholtz-Sloan

Writing, review, and/or revision of the manuscript: J.B. Iorgulescu, M. Torre, M. Harary, T.R. Smith, A.A. Aizer, D.A. Reardon, J.S. Barnholtz-Sloan, A. Perry

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J.B. Iorgulescu, M. Harary

Study supervision: J.B. Iorgulescu, T.R. Smith, A. Perry

J.B. Iorgulescu is supported by an NIH 5T32HL007627-34 award and an Arthur Purdy Stout award. The authors are grateful to Dr. David Louis for his editorial assistance and insights and to the cancer registrars of Brigham and Women's Hospital, Dana-Farber Cancer Center, and Massachusetts General Hospital for their assistance.