Interobserver agreement and histologic analysis of atypical ductal hyperplasia bordering on ductal carcinoma in situ: A multi-institutional study.

Objective: Atypical ductal hyperplasia (ADH) shares histologic features with low-grade ductal carcinoma in situ (DCIS). "ADH bordering on DCIS" represents a diagnostic gray zone with variable interobserver agreement, complicating clinical management.

Methods: We retrospectively analyzed 54 cases of ADH bordering on DCIS between 2010 and 2023. Each case underwent independent histologic review by multiple breast pathologists from different institutions. Histologic features, radiologic findings, clinical follow-up data, and interobserver agreement were analyzed.

Results: While pathologists showed moderate to substantial agreement on individual histologic features, agreement in distinguishing ADH from DCIS was poor (κ = 0.16). Lesion extent (47.7%) was the most frequently cited diagnostic factor, followed by nuclear features (24.9%) and duct involvement (18.5%). Among biopsy cases, those with carcinoma (DCIS or invasive) on subsequent excision (n = 22) were compared to those without (n = 16). Nuclear size more than 2-fold of background epithelial cells (P = .02), spindle-shaped nuclei (P = .006), and necrosis (P = .048) were significantly associated with carcinoma on excision. The presence of any 1 feature had 36.4% sensitivity and 72.2% specificity.

Conclusions: Breast pathologists demonstrated substantial agreement on individual histologic features but poor agreement on final diagnoses, likely due to differences in weighting histologic parameters. While lesion extent was frequently cited, it did not significantly differ between cases with and without carcinoma on excision. Instead, nuclear enlargement, necrosis, and spindle-shaped nuclei were significantly associated with carcinoma in subsequent excision. We propose that biopsy cases exhibiting a nuclear size more than 2-fold of background epithelial cells, necrosis, or spindle-shaped nuclei should be suggestive of DCIS.