Interobserver variability and positive predictive value for ultrasonographic BI-RADS categories requiring pathohistological evaluation

The objective of this study was an analysis of interobserver variability and positive predictive value (PPV) for BI-RADS categories requiring pathohistological evaluation: 4A, 4B, 4C, and 5. Interobserver variability for each of descriptors as well as PPV for final BI-RADS categories requiring pathohistological evaluation was measured in a retrospective study which included 30 ultrasonographic reports, with pathohistological verification, randomly selected from ultrasonographic reports from Institute for Oncology and Radiology of Serbia where about 1,100 breast cancers are verified every year. Ten observers, seven gynecologists, and three radiologists, independently rated each ultrasonographic report according to the fourth edition of BI-RADS atlas. Interobserver variability was measured with k coefficient. There was most conformity for a category of orientation (k = 0.79). Substantial degree of conformity was also present for both boundary (k = 0.71) and shape (k = 0.65) categories. Moderate degree of conformity was achieved for posterior features (k = 0.54) and margins (k = 0.41) descriptors, while there was poor conformity in echogenicity (k = 0.38). In case of a final score, common conformity for all BI-RADS 4A, 4B, 4C, and 5 categories was (k = 0.51); it was the greatest for category 5 (k = 0.50), and it was less for categories 4C (k = 0.37), 4B (k = 0.32), and 4A (k = 0.29). Interobserver conformity for ultrasonographic descriptors and final evaluation of BI-RADS 4A, 4B, 4C, and 5 categories is good. PPV implies that not only division into categories 4 and 5, but also classification into categories 4 and subcategories 4A, 4B, and 4C are justified and clinically applicable.