Observer agreement in the TI-RADS classification of thyroid nodules using ultrasound at Mulago Specialized Hospital
Abstract
Introduction: Ultrasonography is the modality of choice in the initial assessment of thyroid nodules. It is readily available in most parts of the country and can easily be performed. Thyroid Imaging Reporting and Data System (TI-RADS) is used to standardize the risk stratification system for assessment of thyroid nodules. Ultrasonography is operator dependent as it is real time and thus the need to determine observer agreement.
General objective: To determine the observer agreement of TI-RADS categorization of thyroid nodules at Mulago National Referral Hospital.
Methods: It was a cross-sectional study carried out at Mulago National Referral Hospital. Three radiologists with 13, 10 and 4 years of practice described 102 images that were presented to them. They described the thyroid lesions with TI-RADS terminology and assigned a final TI-RADS category. After a period of one month the same images were presented to the same raters to assess intra-observer agreement. Inter-observer and intra-observer agreement in TI- RADS terminology descriptors and final categorization were assessed with the Fleiss kappa and Cohen’s kappa statistics respectively and interpreted according to Landis and Koch guidelines.
Results: The inter-observer agreement among all the raters for composition was substantial (k=0.67), echogenicity was substantial (k=0.67), shape was substantial (k=0.67), margin was slight (k=0.1), and echogenic foci was moderate (k=0.48). Agreement for overall TIRADS categorization was substantial (k=0.62).For intra-observer agreement, rater 1 had moderate to near perfect agreement (k=0.42-1.00), rater 2 had fair to near perfect agreement (k=0.38-1.00) and rater 3 had slight to near perfect agreement (k=0.10-1.00) for the terminology descriptors. The level of agreement for intra-observer assessment in final TIRADS categorization was substantial in all the three raters.
Conclusion: The study showed moderate to substantial level of agreement in most of the terminology descriptors; margin showed slight agreement and there was substantial agreement in the overall TIRADS categorization for inter-observer agreement. For the intra-observer agreement, there was moderate to substantial agreement in terminology descriptors and substantial agreement for the overall TIRADS categorization. This study showed that ACR TI- RADS has little reproducibility especially for the terminology descriptors hence the need for further training of ultrasound performers in ACR TIRADS guidelines.