Inter-observer Variability in the American College of Radiology Thyroid Imaging Reporting and Data System: In-Depth Analysis and Areas for Improvement

The American College of Radiology (ACR) introduced a new standardized system for management of thyroid nodules, the Thyroid Imaging Reporting and Data System (TI-RADS). The purpose of this retrospective study is to evaluate the inter-observer variability in applying TI-RADS in clinical practice without prior dedicated training. We evaluated a total of 180 nodules constituting all consecutive thyroid aspirations performed in the radiology department between January 1, 2014 and June 30, 2014, with exclusion of histologically inadequate samples. Four radiologists, blinded to each other's evaluation and to final pathology results, evaluated all of the nodules based on the TI-RADS lexicon. TI-RADS score and management recommendations were then deduced from the inputted features. Statistical analysis was performed to determine inter-observer agreement among all readers, as well as between each two readers, in all TI-RADS sonographic features and for recommended management per TI-RADS score with multi-user Cohen's κ (Light's κ) and percentage agreement using R. There was fair-to-moderate inter-observer agreement in nodule composition (two-reader κ range: 0.327-0.533) and presence of calcifications (κ range: 0.229-0.527), but poor-to-fair agreement in echogenicity (κ range: 0.141-0.355), shape (κ range: 0.0729-0.513) and margins (κ range: 0.176-0.283). There was fair inter-observer agreement regarding management recommendations (κ range: 0.242-0.359).