A large-scale consensus study of a diagnostic algorithm for canine and feline mammary tumor histopathology: An...

Histological classification of mammary proliferative lesions in dogs and cats is essential for prognostic and predictive studies, as well as for comparative oncology; however, standardization of diagnoses is highly affected by substantial interobserver variability. In this study, a structured diagnostic algorithm based on the Davis-Thompson Foundation classification was assessed for the ability to improve interobserver agreement in the morphological diagnoses of mammary gland lesions in dogs and cats. Consensus diagnoses of 114 selected mammary lesions (91 canine, 23 feline) made by 16 veterinary pathologists with mammary tumor expertise were used as the gold standard, and 79 veterinary pathologists from 21 countries evaluated digitized whole slides in 2 rounds: a first reading without the algorithm and a second reading with the algorithm. Interobserver concordance was measured using unweighted kappa statistics for lesion category (hyperplastic/dysplastic, benign, malignant) and histological subtype. Across all lesions, agreement was higher for lesion category than for histological subtype. A slight improvement in concordance was observed when applying the algorithm, both for category (k = 0.72, before, and k = 0.73, after) and histotype classifications (k = 0.40, before to k = 0.42, after). Also, by using the algorithm, 73/79 raters (92%) modified at least 1 diagnosis and, among these, approximately two-thirds reached agreement with the gold standard in at least 1 case. Considering histotypes, malignant tumors with mixed epithelial and myoepithelial components showed persistently low agreement. Overall, the algorithm provided a reproducible diagnostic framework that may support training and improve standardization of canine and feline mammary tumor classification.