Thyroid Imaging Reporting and Data System(TI-RADS): Interobserver variability and positive predictive value for assessment of thyroid nodules

Abstract

Objective To retrospectively evaluate interobserver variability between radiologists by using lexicon of the Thyroid Imaging Reporting and Data System (TI-RADS) to classify lesions on ultrasonography and retrospectively determine the positive predictive value (PPV) of each TI-RADS level. Methods Five radiologists retrospectively reviewed 188 consecutive thyroid nodules with known pathologic diagnosis of 172 patients from June, 2016 to November, 2016. Each observer described each lesion with TI-RADS terminology and final TI-RADS level were assigned by adding the points from all categories. Cohen κ statistic was used to assessed the interobserver variability for each ultrasonographic(US) descriptors and final TI-RADS level. PPV for all TI-RADS levels were determined for all readers combined. Results For each US descriptor, composition, echogenicity, shape, margin, echogenic foci, their κ were 0.743 (0.713-0.772), 0.418 (0.319-0.517), 0.468 (0.389-0.547), 0.397 (0.291-0.503) and 0.566 (0.514-0.617) respectively (data in parentheses are 95% confidential intervals). The κ for TI-RADS level was 0.782 (0.749-0.81). PPV for TI-RADS level 1 to 5 were 0(0/8), 0(0/23), 14.3%(4/28), 29.7%(11/37) and 85.7%(78/91). Conclusions Interobserver agreement with the TI-RADS terminology is substantial for composition, moderate for echogenicity, shape and echogenic foci, fair for margin.Interobserver agreement is substantial for TI-RADS level. Consequently different understanding in US images has little influence in patient management, justifying the use of ACR-TIRADS in clinical practice. Key words: Ultrasonography; Thyroid nodule; Observer variation; Positive predictive value