Afirma GSC with Hurthle classifiers
In a second study, investigators led by Quan-Yang Duh, MD, professor of surgery, division of general surgery, and chief, section of endocrine surgery, University of California, San Francisco, developed and validated a pair of classifiers to enhance performance of the Afirma platform among Hurthle cell specimens.
“The Hurthle cell lesions tend to give us trouble,” Dr. Duh said. On molecular analysis, those that are malignant seldom harbor mutations that would aid diagnosis, whereas those that are benign are usually classified as suspicious by the original Afirma Gene Expression Classifier (GEC).
“The specific group that is causing trouble are those that are Hurthle cell but not neoplasm, because they are the ones that give you the false positives,” Dr. Duh said. Therefore, it makes sense to stratify lesions on both of these factors, and then subject that specific subset to a more stringent threshold.
Dr. Quan-Yang Duh
The investigators developed two classifiers that work with the Afirma core Genomic Sequencing Classifier (GSC), which uses RNA sequencing and machine learning algorithms.
The first classifier uses differential expression of 1,408 genes to determine whether a sample contains Hurthle cells. The second classifier, applied only to lesions containing Hurthle cells, uses differential expression of 2,041 genes and assesses loss of heterozygosity – which is prevalent in Hurthle cell neoplasms – to determine whether a Hurthle cell lesion is a neoplasm.
The ensemble model then makes a final classification, using a higher threshold for suspicious lesions determined to be Hurthle cell but not neoplasm, and a normal threshold for all the rest.
The investigators validated the Afirma GSC with the two classifiers in blinded fashion using 186 thyroid lesion samples having Bethesda III or IV cytology that had been part of the overall multicenter validation of the original Afirma GEC (N Engl J Med. 2012 Aug 23;367[8]:705-15).
Among the 26 Hurthle cell lesions, specificity for identifying benign lesions improved from 11.8% with the original Afirma GEC to 58.8% with the Afirma GSC and new classifiers. That was an absolute gain of 47% (P = .012), Dr. Duh reported. Sensitivity for identifying cancer was 88.9%.
There were also smaller absolute gains in specificity of 18% among all lesions in the cohort (P = .0028) and 14% among non-Hurthle lesions (P = .028).
“The new GSC test has significantly improved specificity in the patients with Bethesda III and IV specimens with Hurthle cells, and this may reduce unnecessary diagnostic surgery,” said Dr. Duh. “Basically, there are fewer false positives and more patients who can be called benign in the Hurthle cell group who would not need an operation.”
Further validation is needed, he acknowledged. “For a while, I wouldn’t send my Hurthle cell aspirate patients for Afirma, because I knew it was going to come back suspicious. I think I will start to do it now, but we need to see what the answers look like” with additional validation.
Afirma GSC with medullary thyroid cancer classifier
In a third study, investigators developed and validated a classifier for medullary thyroid cancer to be used with the Afirma GSC. They were led by Gregory Randolph, MD, professor of otolaryngology and the Claire and John Bertucci Endowed Chair in Thyroid Surgical Oncology at Harvard Medical School, and division chief of the general and thyroid/parathyroid endocrine surgical divisions at the Massachusetts Eye and Ear Infirmary, Boston.
Better preoperative identification of this cancer is key for several reasons, he maintained.
Dr. Gregory Randolph
Establishing the diagnosis from needle biopsy is challenging, because some features overlap with those of other thyroid lesions, according to Dr. Randolph. In about a third of patients with medullary thyroid cancer brought to the operating room, the diagnosis is unknown at the time, and that often results in inadequate initial surgery.
The investigators developed a medullary thyroid cancer classifier cassette that assesses differential expression of 108 genes. They then performed blinded, independent validation in a cohort of 211 fine-needle aspiration samples from thyroid nodules: 21 medullary thyroid cancers and 190 other benign and malignant neoplasms.
Results showed that the Afirma GSC with the medullary thyroid cancer classifier had sensitivity of 100% and specificity of 100%, reported Dr. Randolph.
“The Afirma GSC medullary thyroid cancer testing cassette, within the larger GSC system, uses RNA sequencing and advanced machine learning to improve the diagnostic detection of medullary thyroid cancer, which currently misses approximately a third of medullary thyroid cancer patients,” he said.
Session attendees wondered which patients are appropriate candidates and how much the test will cost.
“We have to have a discussion about that, because the missed medullaries are, frankly, widely distributed – they can be in any of the Bethesda categories, basically,” Dr. Randolph said. “So, there are cytopathologic mistakes made uniformly, including in the suspicious and frankly malignant Bethesda categories. In terms of cost, this is embedded in the GSC classifier; so, if you order that test, you will obtain this medullary cassette.”
Actual sensitivity of the classifier may ultimately be less than 100% with use in larger samples, he acknowledged. “I think a greater number of validation tests is absolutely in order. I imagine this classifier may not be perfect, but it is way better than the third we miss with just cytopathology.”
Dr. Nikiforov disclosed that he is owner of an IP for ThyroSeq, and that his laboratory has a contract to offer the test commercially. Dr. Duh disclosed that he had no relevant conflicts of interest. Dr. Randolph disclosed that he had no relevant conflicts of interest.