An Agreement Study of Stromal Cellularity, Stromal Cell Atypia, and Tumor Border of Phyllodes Tumors of the Breast.

Accurate evaluations of stromal cellularity, stromal cell atypia, and tumor border are critical for the classification of fibroepithelial lesions. This study evaluated the agreement among 19 subspecialized academic breast pathologists for evaluation of these morphologic features. Nineteen pathologists evaluated 54 images for both stromal cellularity and stromal cell atypia and 23 images for tumor border. Virtual training sessions were provided to the participants. The training sessions included a separate set of 57 images for stromal cellularity and stromal cell atypia and 16 images for tumor border; these images were different from the images used in evaluations. After the training sessions and at least 3 weeks of washout period, the 19 pathologists re-evaluated the same images. We used Krippendorff's alpha (α) to assess agreement in ordinal evaluations (mild vs moderate vs marked stromal cellularity and stromal cell atypia); and Fleiss's kappa (κ) for binary evaluations (mild/moderate vs marked stromal cellularity and stromal cell atypia, and infiltrating vs well-circumscribed tumor border). Before training, agreement on stromal cellularity was moderate in both ordinal (α=0.727) and binary evaluations (κ=0.675); agreement on stromal cell atypia was poor in ordinal (α=0.570) but moderate in binary evaluation (κ=0.446); agreement on tumor border was moderate (κ=0.532). After training, agreement on stromal cellularity remained moderate in both ordinal (α=0.747) and binary evaluations (κ=0.675); agreement on stromal cell atypia remained poor in ordinal (α=0.554) and moderate in binary evaluation (κ=0.427); agreement on border status was improved but remained moderate (κ=0.615). After training sessions, 15 pathologists became more conservative in evaluating stromal cellularity (assigning lower cellularity to the same cases); and 13 became more conservative in evaluating stromal cell atypia. To our knowledge, this is the first systematic study specifically evaluating interobserver agreement in the assessment of key morphologic criteria used in the classification of fibroepithelial lesions. There was moderate agreement in evaluating stromal cellularity and stromal cell atypia using a two-tier system (mild/moderate vs. marked) and moderate agreement in evaluating tumor border. Structured training had limited impact on improving agreement among highly subspecialized breast pathologists but promoted more conservative interpretations. The agreement benchmarks and reference images established from this study may help build consensus on diagnostic criteria for fibroepithelial lesions.