Evaluation of A Lung Cancer Rna Expression Subtyping Panel and Comparison with Histologic Diagnosis in Lung Tumor Samples from Multiple Data Sets Including the Cancer Genome Atlas (Tcga).
Mark Robert Miglarese,Matthew Wilkerson,Cheng Fan,Hawazin Faruki,David N. Hayes,Charles M. Perou,Myla Lai-Goldman
DOI: https://doi.org/10.1200/jco.2014.32.15_suppl.7566
IF: 45.3
2014-01-01
Journal of Clinical Oncology
Abstract:7566 Background: Lung cancer subtypes are described by morphological and molecular characteristics. Differentiation between various subtypes is important for guiding patient management. Variability in morphology, limited tissue, and the need for performing a growing number of genomic tests pose challenges to the current diagnostic standard. Methods: The Lung Subtype Panel (LSP), a previously published 57 gene expression panel for prediction of lung tumor morphologic class (Wilkerson et al.JMD 2013), was investigated using 2,168 lung cancer samples from multiple publically available data sets (TCGA, NCI, UNC, Duke, Expo, Seoul, Tokyo, and France). Data sets with both gene expression data and morphologic classification were selected. Three platforms for gene expression were represented: Affymetrix U133+2 (n=883), Agilent 44K (n=334), and Illumina RNA-seq (n=951). A centroid predictor was used to assign an expression-defined histologic class. Predicted results were compared with tumor morphologic classification and percent agreement was calculated. Results: Ten lung tumor RNA expression datasets were combined into three platform specific data sets comprised of a diverse patient population, including smokers and nonsmokers with Stage I – Stage IV disease. Using the centroid predictor for two classes only (adenocarcinoma or squamous cell carcinoma), the agreement of the predictor with morphologic classification ranged from 79%-94% across the 3 data sets. Analysis of the 4-class predictor is underway and will be reported. To estimate an error rate in morphologic classification, a subset of tumors were reviewed by multiple pathologists, which showed an agreement rate of 65% (CI 49%-73%). Conclusions: In multiple datasets with over two thousand lung cancer samples, molecular profiling using the Lung Subtype Panel compared favorably to light microscopic derived diagnoses, and showed a higher level of agreement than seen by reassessment by multiple pathologists. RNA-based tumor subtyping can provide valuable information, especially when tissue is limiting and the overall morphologic diagnosis remains unclear.