OR32-03 mRNA Expression Based Tumor Behavior Prediction Models In Thyroid Nodules

Abstract

Abstract Disclosure: A. Golding: None. G. Correa: None. E. Valenzuela Sheker: None. R. Jiang: Employee; Self; Veracyte, Inc. Stock Owner; Self; Veracyte, Inc. Y. Hao: Employee; Self; Veracyte, Inc. Stock Owner; Self; Veracyte, Inc. Y. Liu: Employee; Self; Veracyte, Inc. Stock Owner; Self; Veracyte, Inc. J. Huang: Employee; Self; Veracyte, Inc. Stock Owner; Self; Veracyte, Inc. J.P. Klopper: Employee; Self; Veracyte, Inc. Stock Owner; Self; Veracyte, Inc. R.T. Kloos: Employee; Self; Veracyte, Inc. Stock Owner; Self; Veracyte, Inc. G.C. Kennedy: Employee; Self; Veracyte, Inc. Stock Owner; Self; Veracyte, Inc. D. Bimston: None. The 2015 ATA thyroid cancer risk stratification system is driven primarily by the extent of vascular and extrathyroidal extension, as well as lymph node metastases. The objectives of this study were to evaluate the prevalence of invasion and lymph node metastases in a cohort enriched with thyroid malignancies and provide a risk signature with a high negative predictive value ruling out these tumor features. Histopathology reports from Afirma Genomic Sequencing Classifier (GSC) algorithm training and from thyroid cancer patients managed at an integrative endocrine surgery community care practice (total 697 and ∼50% from each) were reviewed for invasion and metastases. An integer from 0-3 was assigned based on the extent of invasion (ranging from none to extensive extra-thyroidal) or extent of lymph node metastases (ranging from none to lateral node metastases). To develop predictive signatures for invasion/metastases, over 400 literature-derived expression-based signatures were evaluated; they cover a broad spectrum of molecular processes, including cancer hallmarks such as endothelial to mesenchymal transition, tumor microenvironment, immune-oncology, metabolism, and treatment sensitivity. Further, acknowledging that multiple complex molecular processes are associated with invasion/metastases, over 200 machine-learning (ML) models were trained within the repeated and nested cross-validation (CV) framework. They utilized these literature-derived signatures as well as differentially expressed genes between high and low risk of invasion/metastases as the basic building blocks. The application of top performing signatures among Bethesda (B) V/VI nodules as well as indeterminate thyroid nodules (ITN - BIII/IV) that are GSC-suspicious, showed that 40-50% of these lesions were marked as low risk for invasion/metastases. With a starting prevalence of 10.0% of high-level invasion (scored 2 or 3), 41.3% of the cohort is ruled out for clinically relevant invasion with a negative predictive value (NPV) of 97.6%. With a starting prevalence of 11.3% of high-level metastases (scored 2 or 3), 49.8% of the cohort is ruled out for clinically relevant metastases with an NPV of 98.6%. The rule-out capability represents more than 4-fold or 7-fold reduction in high-level invasion/metastases from the baseline prevalence, respectively. Expression-based signatures that confidently rule out tumor invasion and metastasis may help personalize the surgical approach for individuals. This may reduce excess surgery and overtreatment, optimizing the initial response to therapy and decreasing the rates of surgical complications and post-operative hypothyroidism. Presentation: Sunday, June 18, 2023