Improving predicted risk of recurrence using molecular profiling in papillary thyroid cancer

Molecular testing can refine the prediction of cancer recurrence. We sought to compare patterns of gene expression in patients with and without recurrence of well-differentiated thyroid cancer to identify pathways associated with recurrence and develop a predictive model based on gene expression. RNA was extracted and sequenced from archival tumor samples of patients well-differentiated thyroid cancer with (n = 8) and without (n = 8) recurrence, all of whom appear clinically at high risk for recurrence. A predictive model was developed using machine learning (ML) with the Thyroid Carcinoma TCGA PanCancer Atlas dataset and externally validated using archival samples. RNA-seq analysis from archival patient samples demonstrated gene expression patterns with striking sex-dependent differences. In tumors from female patients, the TNFα pathway was activated whereas tumors from males showed inhibition of TNFα and estradiol pathways, with findings externally validated through analysis of TCGA data. A prediction model based on TCGA data in female patients was developed that demonstrated an AUC of 0.88 in an external validation cohort for predicting recurrence, providing prognostic information that improves predictions beyond standard clinical parameters. Sex-dependent differences, specifically in TNFα and estrogen response pathways, in thyroid cancer recurrence have important implications for prognosis and treatment.