The Role of Molecular Diagnostics in the Management of Indeterminate Thyroid Nodules

Abstract

This paper is a commentary on the study by Livhits et al. addressing the relative utility of two molecular diagnostic techniques for thyroid nodules indeterminate on FNA cytology.

In a recent issue of the Journal, Livhits et al. (1) addressed the relative performance of two molecular diagnostic techniques for thyroid nodules indeterminate on fine-needle aspiration (FNA) cytology. The study was designed as a classic “real life” effectiveness trial, and as such was tailored toward external validation of the most commonly clinically applied molecular tests. Compared with strictly controlled efficacy trials characterized by higher internal validation achieved by strict design, e.g., blinding of the pathologist to the results of the molecular tests, the availability of the final pathology for all enrolled patients, the advantage of effectiveness—“real life” study design—is that it provides an evaluation of the intervention in a routine practice experience. Consequently, generalizability of its findings is increased, as the study accounts for patient-provider-system-level factors. The study aim was particularly important, since >500,000 FNAs are performed yearly in the United States, with around 100,000 nodules characterized by indeterminate cytology (2). Prior to the availability of molecular diagnostics, patients were referred for lobectomy to achieve a specific pathologic diagnosis. This approach is still followed widely throughout the world today and results in a large number of unnecessary surgeries, perhaps as many as 50% of surgeries for benign nodules. As a result, overtreatment is associated with substantial side effects and incremental cost contributing to the cost of treating and monitoring of patients with thyroid cancer, which reached $1.6 billion in the United States in 2013 (3).

Analysis of molecular signature of thyroid nodules with indeterminate cytology has been evolving over the past two decades. Initially, the only available test was PCR-based screening for BRAFV600E mutation, characterized by high specificity and positive predictive value (PPV) for detection of thyroid cancer, but very low sensitivity. The sensitivity was slightly improved by the introduction of a seven-gene mutation panel (ThyGenX) for testing for the most common genetic alterations present in up to 70% of thyroid cancers, namely BRAFV600E, NRAS, HRAS, and KRAS mutations and gene fusions RET/PTC1, RET/PTC3, and PAX8-PPARG (4–7). However, despite slightly increased sensitivity compared with single gene mutation testing, the performance of this assay was still characterized by very low negative predictive value (NPV), precluding its use as a rule-out test. Conversely, the high specificity and high PPV of this assay has been validated in several studies and the seven-gene mutation panel has been used as a rule-in test in clinical practice (4–7). To increase the accuracy and utility of this panel, combination with a panel of 10 miRNA markers was introduced (ThyGenX/ThyraMIR) (8). Based on a small multi-institutional study, the NPV of the assay was 94% and PPV was 74% (8). However, ThyGenX/ThyraMIR has not been appropriately validated in a large prospective study.

An alternative method applied to stratification of thyroid nodules into high vs low cancer risk used a combination of the ultrasonographic features with the molecular signature of the nodules. De Napoli et al. (9) documented as high as a 100% PPV for the combination of high risk ultrasonographic features such as hypoechogenicity, microcalcifications, irregular margins, and taller than wide shape of the nodules, with the presence of the most commonly observed thyroid cancer mutations in BRAF and NRAS genes (9). However, the diagnostic utility of this approach was tested on a population with a relatively high cancer prevalence of 35%, thus increasing the likelihood for a high PPV. The shortcoming of this approach was a low sensitivity with associated low NPV precluding its use as a rule out test.

The Afirma assay by Veracyte was introduced in 2012 as a response to a need for a good rule-out test (10). The assay was designed to analyze the expression profiles of 142 genes, based on a proprietary algorithm that classified the nodules as either benign or suspicious for malignancy. Based on several clinical validation studies performed to date, Afirma serves as a relatively good rule-out test with high sensitivity and NPV ranging from 75% to 100% (9–11). Unfortunately, the specificity is relatively low and as such is associated with a PPV ranging only between 14% and 44%, limiting its use as a rule-in test (10–12). Livhits et al. (1) consistently documented a sensitivity and NPV of 100%, but low specificity and a PPV of 15.8% and 38.5%, respectively. Of note, NPV and sensitivity of the assay described in this effectiveness trial might be lower. The estimate of false negative results in this study design was impossible, as most patients with benign or negative molecular tests were treated conservatively. Nevertheless, the results are valid and generalizable as they resemble a “real life experience,” where a nonsurgical approach is generally preferred and implemented in patients in whom a molecular signature is suggestive of a benign lesion.

In 2014, the Cancer Genome Atlas, based on DNA and RNA sequencing data, identified genetic alterations in 97% of papillary thyroid cancers (13). The Cancer Genome Atlas formed a basis for the construction of thyroid cancer-tailored next generation sequencing panels designed by Dr. Yuri Nikiforov and colleagues at the University of Pittsburgh. The rationale for utilizing the next generation sequencing panels platforms was to improve diagnostic accuracy. The increased number of mutations examined should lead to a higher assay sensitivity and be associated with a higher NPV, whereas an anticipated detection of cancer-specific mutations would result in maintaining relatively high specificity and PPV. An initial custom panel (ThyroSeq) was designed to target 12 cancer genes with 284 mutational hot spots (14). The second panel, ThyroSeq v2, included 56 thyroid-related genes and analyzed for point mutations and small insertions/deletions in 14 genes and 42 types of gene fusions found in 90% of thyroid cancers. In addition, ThyroSeq v2 evaluated the expression levels of 16 genes to provide an internal control for an adequacy of cellularity in FNA material, as well as to assure appropriate screening for medullary thyroid cancer by measurement of calcitonin expression and parathyroid adenomas by assessing parathyroid hormone expression (15, 16). ThyroSeq v2 has been validated as characterized by high sensitivity and NPV of 96% to 97% with reasonably high specificity and a PPV of 77% to 83% (15–17), enabling its use as both a rule-out and a rule-in test. And Livhits et al. (1) reported that ThyroSeq v2 outperformed Afirma in its specificity and PPV, thereby affecting clinical management and leading to conservative rather than a surgical therapeutic approach in a larger number of patients. However, the study is limited by its analysis of two different cohorts of patients, thus rendering impossible the direct comparison of the accuracy of Afirma and ThyroSeq v2 in the same nodules. Moreover, it is important to recognize that although the sensitivity and specificity of any diagnostic test depends only on test performance, the NPV and PPV depend on the prevalence of disease in the tested population. Thus, in a recently published study analyzing a population with cancer prevalence of 16%, the NPV of ThyroSeq v2 remained high at 96%, but the PPV was only 22% (18). One of the reasons for low PPV in mutation positive samples is the well-known phenomenon of the presence of certain mutations in benign lesions. Indeed, there are mutations associated with close to 100% cancer risk such as BRAFV600E mutation or PPARG, NTRK1, NTRK3, and ALK fusions, whereas the presence of RAS, PTEN, and EIF1AX mutations or THADA fusions entails a significantly lower risk of cancer, and GNAS mutation is almost exclusively associated with benign lesions (15, 16). This phenomenon has been recognized in the newest version of the ThyroSeq v3 (112-gene test). ThyroSeq v3 was designed to increase sensitivity by adding recently discovered genetic markers related to thyroid nodules and cancer, including the copy number alteration observed in up to 7% of cancers (19). Most importantly, this version of the test introduces a scoring system to account for molecular signatures typical for cancer and for benign adenomas. Another advantage of ThyroSeq v3 is the inclusion of benign and malignant Hurthle cell lesions thereby increasing the accuracy of classification of Hurthle cell tumors (19). Based on the analysis of a training set and validation set, Nikiforov et al. (19) found that the test is characterized by a sensitivity of 94% to 98% and specificity of 81.8% to 89.4% (19). A new version of Afirma has also been recently introduced and this genomic sequencing classifier demonstrated a sensitivity of 91% and specificity of 68% (20). Both of the latter tests need to be validated in prospective cohort studies.

Finally, the recently introduced Rosetta GX test is based exclusively on detection of 24 miRNA markers. Based on a retrospective study involving 150 nodules with indeterminate cytology, its NPV of 92% and PPV of only 43% suggests its potential utility as a rule-out test with limited utility as a rule-in test. Moreover, its performance in Hurthle cell lesions remains unknown (21). It is worthwhile to underscore that the performance of all of the above mentioned molecular tests is also unknown for noninvasive follicular thyroid neoplasm with papillary-like nuclear features—NIFTP, as it has been categorized previously by pathologists as either benign or malignant noninvasive encapsulated follicular variant of papillary thyroid cancer.

Given the above-mentioned limitations and the lack of appropriate independent validation of the most recently developed molecular tests, current American Thyroid Association guidelines suggest patient-tailored individualized decision making for patients with thyroid nodules with indeterminate cytology. Recommendation 13 states: “If molecular testing is being considered, patients should be counseled regarding the potential benefits and limitations of testing, and about the possible uncertainties in the therapeutic and long-term clinical implications of results” (22).

Another potential limitation of the routine utilization of molecular diagnostics is their relatively high cost. However, that cost needs to be balanced against the cost of unnecessary surgery and potential postsurgical complications, which are likely, given that the majority of thyroidectomies in the United States are performed by low volume surgeons, with 51% of surgeons performing just one thyroidectomy per year (23). Lee et al. (24), in an analysis published 4 years ago, and as such not including the newest molecular tools, concluded that Afirma gene expression profiling combined with a seven-gene mutation panel was associated with beneficial cost-effectiveness in the US health care system, but not in the Canadian system (24).

It is clear that the molecular signature of thyroid nodules may serve as one of the useful tools in guiding decision making regarding a conservative vs surgical approach to management. Interpretation of the results of molecular screening of cytologically indeterminate nodules needs to account for cancer prevalence at a given institution, as low cancer prevalence increases NPV of the test, whereas high cancer prevalence increases the PPV of the test. Appropriate clinical judgment and individualized risk stratification is best based on patient age, presence of comorbidities, ultrasonographic features of the nodules, and determination of molecular signature to identify candidates for active surveillance or surgery.

Glossary

Abbreviations:

FNA

fine-needle aspiration

NPV

negative predictive value

PPV

positive predictive value