Xpression Atlas, the largest gene and fusion panel, is now a part of Afirma GSC. TERT DNA analysis is now also available
Learn More

Development and Validation of Classifiers to Enhance the Afirma Genomic Sequencing Classifier Performance Among Hürthle Cell Specimens

 

Interview with Quan-Yang Duh, M.D., University of California San Francisco

 
Dr. Quan-Yang Duh is professor of surgery and chief, Section of Endocrine Surgery, at the University of California San Francisco. He presented data at the 2017 ATA Annual Meeting showing that the Afirma Genomic Sequencing Classifier’s (GSC) ensemble machine learning algorithms effectively distinguished benign from cancerous Hürthle cells in thyroid cancer diagnosis. Here he discussed his findings.
 

Q: Why are Hürthle cells so hard to distinguish? 

Dr. Duh: Hürthle cells have a different appearance than other thyroid cells, largely due to their high amounts of mitochondria. When Hürthle cells are found in a thyroid patient, they could be one of three things: inflammation that is not a tumor; a benign tumor; or a cancerous tumor. The challenge is that distinguishing between these three possibilities is quite difficult using cytopathology. In fact, Hürthle cells account for approximately 20 percent of indeterminate FNAs. Traditionally, when Hürthle cells were found, the only way to determine what type of Hürthle cell they were was to remove the thyroid.

The original Afirma Gene Expression Classifier did a great job at distinguishing between benign and malignant thyroid nodules when cytopathology was indeterminate. However, Hürthle cells remained a challenge because, even genomically, they are hard to distinguish.

Q: How does the Afirma Genomic Sequencing Classifier address this challenge?

Dr. Duh: The next-generation Afirma GSC uses RNA sequencing to extract as much genomic information as possible from the FNA sample. The test then uses multiple classifiers – which is called an “ensemble” approach – to interpret the data. To improve the classification of Hürthle cells, the Afirma GSC first determines whether the FNA sample contains Hürthle cells or not. It uses over 1,400 genes to do this. If yes, then a second classifier looks at additional criteria, including increased loss of heterozygosity – which is largely specific to Hürthle cell neoplasms – to distinguish Hürthle neoplasms from Hürthle non-neoplasms. This second classifier uses over 2,000 genes and nearly 190,000 genomic variants. Samples that are found to be Hürthle, but not neoplastic, are adjudicated by a more tolerant GSC threshold because their risk of cancer is very low. This allows the GSC to report more benign Hürthle cell samples as benign. FNA samples that are found to be Hürthle neoplasms are evaluated as any other indeterminate samples by the overall classifier to determine if they are benign.

Q: Can you please explain the study that you are presenting at the ATA Annual Meeting? 

Dr. Duh: We validated the Afirma GSC’s ability to distinguish benign from malignant Hürthle cells by using it blindly on 191 indeterminate thyroid nodule FNA samples, of which 26 contained Hürthle cells. Among the Hürthle cell group, the test showed a sensitivity of 89 percent for malignancy and a specificity of 59 percent for benign nodules. This is a 47 percent improvement over the original Afirma classifier.

Q: Why are these findings important? 

Dr. Duh: Our findings suggest that Hürthle cells that used to be called suspicious by the original Afirma classifier will now be called benign by the Afirma GSC with a high degree of accuracy. Our findings should give physicians confidence to trust the results of the next-generation test. This means that more patients with benign nodules should be able to avoid unnecessary surgery.

Read the Abstract Here