[1888] Multisite Analytical Validation of a 92-Gene Molecular Classifier for Cancers of Uncertain Primary

Sarah E Kerr, Catherine A Schnabel, Peggy S Sullivan, Yi Zhang, Veena Singh, Brittany Carey, Mark G Erlander, W Edward Highsmith, Sarah M Dry, Elena F Brachtel. Mayo Clinic, Rochester, MN; bioTheranostics, Inc., San Diego, CA; University of California Los Angeles, Los Angeles, CA; Massachusetts General Hospital, Boston, MA

Background: With increasing use of site- and subtype-specific cancer therapies, accurate tumor classification is of paramount importance to optimize treatment and improve patient outcomes. Diagnosis remains uncertain in 2-5% of solid malignancies after standard histopathologic evaluation. Gene expression-based molecular classifiers have been proposed as a diagnostic aid. This study evaluates the performance of a 92-gene molecular classifier (CancerTYPE ID, bioTheranostics Inc.) for identification of tumor type and subtype in a large multi-institution cohort.
Design: Formalin-fixed paraffin-embedded tumors representing 28 tumor types and 50 subtypes predicted by the molecular classifier were selected and passed pathologist-adjudicated review at 3 institutions. Blinded tumor sections were submitted and tested using a prespecified classification model that reports computational algorithm results as rank probabilities. The top-ranking tumor type predicted by the classifier was compared to the adjudicated reference diagnosis.
Results: The study cohort was comprised of 790 tumors with at least 25 and 10 cases for each classifier type and subtype, respectively. Forty-seven (5.9%) cases were considered unclassified by the assay. The overall concordance rate of the assay prediction with the reference diagnosis was 87% at the main type level and 82% at the subtype level. The reference diagnosis was within the top two rank predicted types in 91% of cases. No statistically significant differences in performance were observed in subset analyses of primary versus metastatic tumors, by grade, or by biopsy type.

Performance by Tumor Type

Type	Concordance %	Type	Concordance %
Adrenal	96	Melanoma	88
Brain	96	Meningioma	100
Breast	80	Mesothelioma	87
Cervix-adeno	72	Neuroendocrine	98
Endometrium	48	Ovary	86
Gastroesophageal	65	Pancreatobiliary	88
GIST	92	Prostate	100
Germ-cell	83	Sarcoma	95
HeadNeck-salivary	88	Sex-cord-stromal	80
Intestine	85	Skin-basal-cell	100
Kidney	97	Squamous	86
Liver	96	Thymus	72
Lung	65	Thyroid	96
Lymphoma	84	UrinaryBladder	64

Conclusions: The 92-gene assay demonstrated excellent overall performance for classification of a diverse set of tumor histologies. Variable positive predictive values indicate that the classifier may be more useful for some differential diagnoses than others. Results from this large scale validation study support the clinical utility of the 92-gene assay as a standardized adjunct to clinicopathologic findings.
Category: Special Category - Pan-genomic/Pan-proteomic approaches to Cancer

Tuesday, March 20, 2012 9:30 AM

Poster Session III # 274, Tuesday Morning