Interobserver reproducibility of macroscopic classification of breast cancer from representative cut surface of resected specimens and hematoxylin-eosin-stained slides.

BACKGROUND

Characteristics in shape and margins detected with diagnostic imaging are key features of breast cancers, but consensus on the criteria for macroscopic classification of breast cancer has yet to be established. The General Rules for Clinical and Pathological Recording of Breast Cancer, 19th edition, of Japan has adopted a macroscopic classification of breast cancer.

METHODS

Four breast pathologists evaluated the interobserver agreement level of the macroscopic classification for breast cancer cases using macroscopic photographs of cut surfaces of resected specimens and their overview images of hematoxylin-eosin-stained slides, which were provided from the pathology database of three facilities. From macroscopic shape and margins, these cases were classified into four types: non-mass, expansive, infiltrative, and mixed. Criteria for each type were established for the training set (n = 60), and then four pathologists independently classified 105 cases as the validation set. Interobserver agreement levels were calculated with kappa statistics.

RESULTS

In the validation set, consensus types were non-mass, expansive, infiltrative, and mixed in 18, 22, 25, and 35 cases, respectively. The four observers gave unanimous types for 47 cases (45%), and three or more observers gave concordant types for 78 cases (74%). An agreement level among the four observers was moderate (kappa = 0.561) in total, but the agreement levels between consensus types and observer-assigned types were always substantial (kappa = 0.690-0.799).

CONCLUSION

Although the interobserver agreement level of the classification was still moderate in total, setting more specific and quantitative criteria and further studies may enhance the reproducibility of the macroscopic classification.