A Retrospective Evaluation of the Diagnostic Performance of an Interdependent Pairwise MicroRNA Expression Analysis with a Mutation Panel in Indeterminate Thyroid Nodules.

Abstract

Background: The addition of genetic analysis to the evaluation of thyroid nodule fine-needle aspiration biopsy samples improves diagnostic accuracy of cytologically indeterminate thyroid nodules (ITNs) with Bethesda III or IV cytopathology. We previously reported the performance of a multiplatform molecular test, referred to in this study as MPTXv1, that includes a mutation panel (ThyGeNEXT®) plus an algorithmic microRNA (miRNA) risk classifier (ThyraMIR®). Complex interactions of growth-promoting and -suppressing miRNAs affect the phenotype. We previously demonstrated that accounting for these interactions with pairwise miRNA expression analysis improves the diagnosis of medullary thyroid carcinoma. In this study, we assess the impact of pairwise miRNA expression analysis on risk stratification of ITNs. Methods: Pairwise expression analysis of 11 miRNAs was performed on a training cohort of histopathology-proven benign nodules (n = 50) to define the mean and standard deviation of each pairwise analysis and create a Benign/Malignant Profiler (MPTXv2), deviations from which predicted the malignancy risk. Clinical validation of MPTXv2 was assessed using a cohort of 178 ITN (Bethesda III and IV) samples from a multicentered, blinded retrospective study, previously evaluated by MPTXv1. Results: Compared with MPTXv1, MPTXv2 significantly improved the test performance. The receiver operating characteristic (ROC) areas under the curve (AUC) increased from 0.85 to 0.97 (p < 0.001), and the diagnostic accuracy at the positive threshold increased significantly (p < 0.05) from 83% [95% confidence interval (CI) = 76-88] to 93% [CI = 89-96]. The significant improvement in the ROC AUC and the diagnostic accuracy was due to a strong statistical trend for improvement in specificity at the positive threshold. At the positive threshold, the specificity for MPTXv1 was 90% [CI = 84-95] and improved to 98% [CI = 94-99] for MPTXv2. Using the MPTXv2, the Moderate-Risk cohort decreased from 50 samples (28% of the cohort) to 24 samples (13% of the cohort). This 52% decrease is statistically significant (p < 0.001) and clinically meaningful. Conclusion: As compared with MPTXv1, pairwise miRNA expression analysis used in MPTXv2 significantly improved the diagnostic accuracy of ITN risk stratification and reduced the size of the Moderate-Risk group. Prospective trials are indicated to confirm these findings in a clinical practice setting.