Use of Predictive Modeling to Tailor Molecular Testing Utilization for Thyroid Nodules

Objective: Various risk stratification systems for cytologically indeterminate thyroid nodules are available. However, malignancy risk assessment data, such as ultrasound features, are not always used when the decision is to order molecular testing or not. Our aim was to investigate the utility of molecular testing after incorporating an algorithm with ultrasound-based risk of malignancy (ROM) estimation. Study Design: Diagnostic/prognostic study. Setting: Single-institution urban tertiary care center. Methods: We performed a single-institution retrospective chart review of all thyroid nodules that had undergone molecular testing. A web-based Malignancy Risk Estimation System for Thyroid Nodules was utilized with ultrasound findings to stratify malignancy risk according to the Korean Thyroid Imaging Reporting and Data System (TI-RADS), French TI-RADS, American Association of Clinical Endocrinology guideline, and American Thyroid Association guideline. A novel algorithm for utilizing molecular testing at our institution was developed with the Korean TI-RADS and with recommendations from the American Thyroid Association and National Comprehensive Cancer Network. Results: The Korean TI-RADS performed best in our population (area under the curve = 0.83). A positive molecular test result had a positive association with a higher ROM according to all 4 models (P <.05). Use of our algorithm prior to molecular testing would have prevented 38% of benign/low-ROM negative nodules (n = 28) from being tested. Conclusion: In patients with indeterminate thyroid nodules, an algorithm built on pre- and posttest probability to guide molecular testing might reduce unnecessary testing of benign and low-risk nodules.