FNA Analysis Could Help Predict Thyroid Cancer Recurrence Risk

Use of transcriptional data taken from the FNA of thyroid nodules could help in predicting recurrence post surgery in certain thyroid cancer patients.

The use of transcriptional data taken from the FNA of thyroid nodules may be useful in improving the preoperative prediction of postoperative recurrence among patients with differentiated thyroid carcinoma, according to the results of a small study presented as a poster at the 2015 American Society of Clinical Oncology (ASCO) Annual Meeting (abstract 6044).

“If independently validated in a sufficiently large number of patients, such molecular classifiers may augment initial risk stratification and individualization of patient care,” wrote researchers led by Steven I. Sherman, MD, of the University of Texas MD Anderson Cancer Center.

According to background information of the study, patients with thyroid cancer are currently classified according to the 2009 American Thyroid Association (ATA) system as high-, intermediate-, or low-risk for recurrence after undergoing thyroid surgery. According to the researchers, this staging system currently lacks any information on molecular predictors of outcome and is limited because it does not allow for pre-thyroidectomy staging.

In this study, Sherman and colleagues used FNA material from 79 preoperative patient samples and classified each sample as low-risk or intermediate/high-risk. Microarray expression data were obtained on all samples and supervised learning was used to train classifiers (support vector machine [SVM], random forest [RF], penalized logistic regression [PLR], and an ensemble of the three). 

The researchers built classifiers using 320 genes and DESeq models that controlled for BRAF status. Genes used in the classification included COX6C, FANCA, KCTD17MPRIP, TUBA1B, DCAKD, ICE2, MCM3AP, WSB2, and TNFRSF14.

According to the poster results, a maximum classification performance for low- compared with intermediate/high-risk was found for an SVM classifier with a maximal area under the ROC curve (AUC) of 0.86; however, all classifiers were able to achieve similar AUC. RF achieved an AUC of 0.82; PLR achieved an AUC of 0.82, and the ensemble of the three achieved an AUC of 0.84.

The researchers found that genes that were useful in the classification of patients belong to a “variety of transmembrane signaling pathways including ECM-receptor interaction, focal adhesion, and cell adhesion molecules.”

When the researchers applied the SVM classifier to the 79-patient cohort in this study, it correctly identified 79.3% of ATA low-risk tumors and 82% of the ATA intermediate/high-risk tumors.