Variability in Diagnostic Accuracy of Thyroid Nodules in a Southern United States Clinical Cohort.

INTRODUCTION: Thyroid nodules are common, with malignancy identified in 5%-15% of cases. Ultrasound-based Thyroid Imaging Reporting and Data System (TI-RADS) and fine-needle aspiration biopsy-derived Bethesda classifications are central to risk stratification and clinical decision-making. However, both systems were developed without explicit consideration of racial diversity, raising concerns regarding their generalizability. Given known racial disparities in thyroid cancer presentation and outcomes, this study evaluated whether the diagnostic performance of TI-RADS and Bethesda varies across racial groups. METHODS: A retrospective cohort study was conducted at a single quaternary academic center. Adult patients who underwent thyroidectomy with available final pathology and preoperative TI-RADS scoring, Bethesda classification, or both were included. Test-positive thresholds were defined as TI-RADS ≥4 and Bethesda III-VI. Sensitivity and specificity were calculated using surgical histopathology as the reference standard. Diagnostic performance was stratified by race, and chi-square analyses were performed. Receiver operating characteristic curves were generated to compare overall discriminatory ability. RESULTS: Of 512 patients, 64.7% were White, 32.4% Black, and 2.9% other racial/ethnic groups. Malignancy was identified in 31.4%. TI-RADS sensitivity was 73.2% overall and did not differ significantly by race (76.9% Black versus 68.8% White; P = 0.432). Specificity was 49.0% overall and was higher in Black patients (56.8%) than White patients (45.7%), although this difference was not statistically significant (P = 0.093). Bethesda demonstrated higher overall sensitivity (89.3%) and specificity (60.6%), with similar sensitivity across groups (89.5% Black versus 88.1% White; P = 0.823). Specificity was marginally higher in Black patients (68.5%) than White patients (56.6%), although this difference was also not statistically significant (P = 0.056). Receiver operating characteristic analysis showed superior discriminatory performance of Bethesda compared with TI-RADS. CONCLUSIONS: TI-RADS exhibited moderate sensitivity and low specificity, whereas Bethesda demonstrated stronger diagnostic accuracy. Importantly, performance did not significantly differ by race for either classification system, supporting their applicability in this Southern US cohort. The consistently higher specificity observed among Black patients, although not statistically significant, warrants further investigation. These findings highlight opportunities for further refinement, including biology-informed threshold adjustments and prospective validation in more diverse cohorts, to ensure continued equitable diagnostic performance.