RESUMO
CONTEXT: Ultrasound (US) risk-stratification systems for investigation of thyroid nodules may not be as useful as anticipated. OBJECTIVE: We aimed to assess the performance and costs of the American College of Radiology Thyroid Image Reporting And Data System (ACR-TIRADS). DESIGN SETTINGS AND PARTICIPANTS: We examined the data set upon which ACR-TIRADS was developed, and applied TR1 or TR2 as a rule-out test, TR5 as a rule-in test, or applied ACR-TIRADS across all nodule categories. We assessed a hypothetical clinical comparator where 1 in 10 nodules are randomly selected for fine needle aspiration (FNA), assuming a pretest probability of clinically important thyroid cancer of 5%. RESULTS: The gender bias (92% female) and cancer prevalence (10%) of the data set suggests it may not accurately reflect the intended test population. Applying ACR-TIRADS across all nodule categories did not perform well, with sensitivity and specificity between 60% and 80% and overall accuracy worse than random selection (65% vs 85%). Test performance in the TR3 and TR4 categories had an accuracy of less than 60%. Using TR5 as a rule-in test was similar to random selection (specificity 89% vs 90%). Using TR1 and TR2 as a rule-out test had excellent sensitivity (97%), but for every additional person that ACR-TIRADS correctly reassures, this requires >100 ultrasound scans, resulting in 6 unnecessary operations and significant financial cost. CONCLUSIONS: Perhaps surprisingly, the performance ACR-TIRADS may often be no better than random selection. The management guidelines may be difficult to justify from a cost/benefit perspective. A prospective validation study that determines the true performance of TIRADS in the real-world is needed.