Development and validation of an individualized diagnostic signature in thyroid cancer

Abstract

New molecular signatures are needed to improve the diagnosis of thyroid cancer (TC) and avoid unnecessary surgeries. In this study, we aimed to develop a robust and individualized diagnostic signature in TC. Gene expression profiles of tumor and nontumor samples were from 13 microarray datasets of Gene Expression Omnibus (GEO) database and one RNA-sequencing dataset of The Cancer Genome Atlas (TCGA). A total of 1246 samples were divided into a training set (N = 435), a test set (N = 247), and one independent validation set (N = 564). In the training set, 115 most frequent differentially expressed genes (DEGs) among the included datasets were used to construct 6555 gene pairs, and 19 significant pairs were detected to further construct the diagnostic signature by a penalized generalized linear model. The signature showed a good diagnostic ability for TC in the training set (area under receiver operating characteristic curve (AUC) = 0.976), test set (AUC = 0.960), and TCGA dataset (AUC = 0.979). Subgroup analyses showed consistent results when considering the type of nontumor samples and microarray platforms. When compared with two existing molecular signatures in the diagnosis of thyroid nodules, the signature (AUC = 0.933) also showed a higher diagnostic ability (AUC = 0.886 for a 7-gene signature and AUC = 0.892 for a 10-gene signature). In conclusion, our study developed and validated an individualized diagnostic signature in TC. Large-scale prospective studies were needed to further validate its diagnostic ability.

We develop and validate an individualized diagnostic signature in thyroid cancer.

http://ift.tt/2G9w92T