INTER-OBSERVER AGREEMENT IN BI-RADS CLASSIFICATION OF BREAST MASSES AMONG ULTRASONOGRAPHY PERFORMERS AT MULAGO NATIONAL REFERRAL HOSPITAL

Abstract

Background: Breast ultrasound is an invaluable tool in breast imaging for detection and characterization of breast masses. Using the Breast Imaging Reporting And Data System (BI-RADS) lexicon for lesion differentiation demands high rate of inter-observer agreement. Inter-observer agreement is thus a matter of strong concern in clinical radiological practice. Objective: To evaluate inter-observer agreement in BI-RADS classification of breast masses among ultrasonography performers at Mulago National Referral Hospital. Research methods: This was a cross-sectional study carried out at Mulago National Hospital. Three radiologists, three radiographers and three radiology residents described 124 images that were displayed on a large screen in front of an auditorium. Two minutes were given for each image to watch and to describe a lesion with BI-RADS terminology and assign a final BI-RADS category. Inter-observer agreement in BI-RADS terminology descriptors and final categorization was assessed with Fleiss kappa statistic and interpreted using Landis and Koch guidelines. Results: Among the radiologists, the inter-observer agreement was fair for echopattern (k-0.38), moderate for shape and posterior acoustic features (k-0.57 and 0.51 respectively) and good for orientation and margin (k-0.68 and 0.62 respectively). Among the radiographers, the agreement was fair for echopattern (k-0.38), moderate for posterior acoustic features (k-0.42) and good for shape, margin and orientation (k-0.64, 0.65 and 0.73 respectively). Among the residents, the agreement was good for margin (k- 0.64), moderate for shape, orientation and posterior acoustic features (k-0.58, 0.50 and 0.45 respectively) and poor for echo pattern (k-0.19).The reproducibility of the final assessment after grouping the final categories into negative (BI-RADS 0,2), imaging follow up (BI-RADS 3) and positive (BI-RADS 4,5) the level of agreement was moderate for radiologists and radiographers, while residents had fair agreement (k- 0.45, 0.49 and 0.36 respectively). Conclusion and recommendations: The radiologists and radiographers had fair to good agreement for terminology descriptors and final categorization while residents had poor to good agreement. Classification into subgroups of non-circumscribed margins was poorly reproducible, thus we recommend disseminated use of the condensed ultrasound BI-RADS lexicon in Uganda and the attending radiologist should review the residents’ preliminary report.