Analysis of the positive predictive value of the subcategories of BI-RADS (R) 4 lesions: preliminary results in 880 lesions

Objectives: The positive predictive values (PPV) of the subcategories of BI-RADS (R) 4 lesions (A/B/C) vary widely, and their correlation with specific descriptors has yet to be defined. We aimed to analyze the PPV of the subcategories and of the mammographic and ultrasonographic descriptors assigned to each. Material and methods: We analyzed 880 histologically confirmed lesions prospectively classified as BI-RADS (R) 4 A/B/C between 2003 and 2010. The statistical analysis included significance tests, contingency tables, and relative risk (RR) ratios, calculated for 545 mammographic lesions and 627 ultrasonographic lesions. Results: The PPV was 8.8% for subcategory 4A, 18.9% for subcategory 4B, and 58.3% for subcategory 4C. The correlation between PPV and lesions was what we expected, with three exceptions: a) the PPV of 4A was greater than that of 4B in nodules that were irregular or had uncircumscribed margins on ultrasonography and in microcalcifications with segmental distribution on mammography, b) BI-RADS (R) 3 lesions classified as BI-RADS (R) 4, and c) identical lesions classified in distinct subcategories. In the contingency table analysis, the mammographic lesions were 4B/C and the ultrasonographic lesions were 4B. On mammography, the RR was significant for nodules with irregular shape (RR = 3.205) and for those with spiculated margins (RR = 2.469), as well as for microcalcifications that were pleomorphic (RR = 2.531) or amorphous (RR = 0.334), and for those with segmental (RR = 1.895). On ultrasonography, the RR were significant for all the descriptors, with values greater than 1 for irregular shape (RR = 1.977) and uncircumscribed margins (RR = 2.277). Conclusions: Our results corroborate previous reports. The exceptions can be explained by aspects related to variability and nonradiological factors that might influence the classification and PPV. Mathematical models should be developed to enable the objective classification and these should include factors not related to imaging. (C) 2011 SERAM. Published by Elsevier Espana, S.L. All rights reserved.