Real World Data To Enable Large-Scale Assessment of WHO CNS5 Glioma Classification

ASCO 2022 Presentation
Authors Kapilivsky, Joshuah, Hatanpaa, Kimmo J, Manghnani, Kabir, Taxter, Timothy, Stumpe, Martin C, Guinney, Justin, Beauchamp, Kyle, Buerki, Robin A, Breschi, Alessandra

Background: In 2021, the WHO revised its classification of central nervous system tumors (WHO CNS5) around IDH status and inclusion of key somatic alterations in addition to histopathological traits. While providing a more specific classification system for patients, the guidelines introduce new logistical challenges for pathologists, relying on multi-modal data for accurate classification. In this study, we use a combined clinical/molecular real world dataset to reclassify a cohort of adult diffuse gliomas and evaluate prognostic impact using real-world overall survival (rwOS).

Methods: We retrospectively analyzed a de-identified dataset of 2,703 adult diffuse glioma samples profiled with the Tempus xT assay (DNA-seq of 595-648 genes at 500x coverage, whole-exome capture RNA-seq). Original diagnoses were identified from sample pathology reports. We assessed mutation status for genes relevant to WHO CNS5 classification (IDH1/2, ATRX, and TERT) and copy number alterations for genes CDKN2A/B and EGFR, as well as arms 1p, 19q and chromosomes 7 and 10. Samples were excluded if the original diagnosis or molecular findings indicated a diffuse midline or pediatric glioma. Necrosis and microvascular proliferation were inferred from gene expression profiles using machine learning. rwOS was defined as time from original diagnosis until death. To account for left truncation, samples were only considered at risk of death after study entry (e.g., date of sequencing if sequenced as part of clinical care).

Results: Using relevant clinicopathological traits and genomic alterations, we assigned WHO CNS5 labels to all samples—512 astrocytoma, IDH-mutant; 186 oligodendroglioma, IDH-mutant, and 1p/19q co-deleted; and 2,005 glioblastoma, IDH-wildtype (IDH-wt). We further stratified astrocytoma, IDH-mutant samples by grade, resulting in 308 classified as grade 4. Of the samples with an original diagnosis of astrocytoma, oligodendroglioma, or glioblastoma, 13.9% changed under the WHO CNS5 guidelines, including 166 glioblastomas re-classified as grade 4 astrocytoma, IDH-mutant, and 125 astrocytomas re-classified as glioblastoma, IDH-wt. WHO CNS5 reclassification resulted in more accurate prognostic stratification (rwOS) than the original diagnosis (likelihood ratio test; P < 8e-32). For example, the observed hazard ratio (HR) of an original astrocytoma diagnosis (versus glioblastoma) (Cox PH Model; HR=0.42[CI:0.32,0.56]) was less extreme than the observed HR of an astrocytoma, IDH-mutant diagnosis (versus glioblastoma, IDH-wt) under WHO CNS5 (Cox PH Model; HR=0.08[CI:0.04,0.19] grades 2-3; HR=0.18[CI:0.12,0.26] grade 4).

Conclusions: This work highlights the utility of comprehensive molecular profiling in classifying patients with adult diffuse gliomas according to the newest WHO CNS5 guidelines, and confirms the improved prognostic stratification within a retrospective real-world dataset.