The validity and reproducibility of the thyroid imaging reporting and data system (TI-RADS) in categorization of thyroid nodules: Multicentre prospective study

•ACR TI-RADS predicts malignant thyroid nodules with high diagnostic validity.•Combined TR4/TR5 increases sensitivity of ACR TI-RADS for malignant thyroid nodules.•The referring clinicians accept ACR TI-RADS as a valuable reporting system. To assess diagnostic validity and reproducibility of Thyroid Imaging Reporting and Data System (TI-RADS) for interpretation of thyroid nodules by thyroid ultrasonography (US). A prospective multicentre study initially included 557 patients with clinically suspected thyroid nodules. After exclusion, a final cohort of 380 patients with 948 thyroid nodules detected by US were enrolled. Based on American College of Radiology (ACR) TI-RADS, three radiologists analysed all US examinations independently and assigned a TI-RADS category to each thyroid nodule. The final diagnosis was based on cytology which was used as reference standard for calculating diagnostic performance of TI-RADS for predicting malignant thyroid nodules. The Fleiss and weighted kappa (κ) statistics were applied to assess inter-observer agreement of morphological features and TI-RADS scoring results for thyroid nodules. Additionally, we made a simple screening among referring clinicians to assess the clinical response to application of TI-RADS. A total of 948 thyroid nodules were evaluated; 136 (14.3%) were malignant, and 812 (85.7%) were benign. The papillary carcinoma was the most common malignant thyroid nodules (81.6%). The best cut-off value for predicting malignant thyroid nodules was > TR3. On a lesion-based analysis, the TI-RADS had a sensitivity, specificity, and an accuracy of 98.3%, 90.9%, and 92.1%, respectively when regarding those thyroid nodules classified as > TR3 for predicting malignancy. The inter-observer agreement of the TI-RADS category was good (κ = 0.636). Ninety percent of referring clinicians accept TI-RADS. TI-RADS improves diagnostic performance of US for predicting malignant thyroid nodules with high validity and high reproducibility.