Confluent glands drive the risk of concurrent carcinoma in atypical endometrial hyperplasia: a multi-center external validation study

Background: Atypical endometrial hyperplasia/endometrioid intra-epithelial neoplasia is the main precursor of endometrioid endometrial carcinoma. Its histopath- ological spectrum is heterogeneous, and inter-observer reproducibility remains sub- optimal. An integrated histological model combining cytological atypia and architectural complexity has been proposed to improve risk stratification for concurrent carcinoma. Objective: To externally validate the prognostic performance and reproducibility of an integrated histological risk stratification model of atypical endometrial hyperplasia/ endometrioid intra-epithelial neoplasia in a multi-center cohort. Methods: This multi-center retrospective cohort study included 201 patients with a pre-operative diagnosis of atypical endometrial hyperplasia/endometrioid intra-epithelial neoplasia who underwent hysterectomy between January 2020 and December 2024 at 3 tertiary referral centers. Index biopsies were independently reviewed by 2 blinded pathologist panels and classified into low-grade atypia, high-grade atypia, and confluent glands. The primary outcome was the rate of concurrent endometrial carcinoma at hysterectomy. Inter-observer agreement and multi-variable logistic regression were used to assess reproducibility and identify independent predictors of carcinoma. Results: Among 201 patients (mean age 57 years), 83 (41.3%) were classified as low-grade, 87 (43.3%) as high-grade, and 31 (15.4%) as confluent glands. Endometrial carcinoma was identified in 42 patients (20.9%). Carcinoma rates were 9.6%, 14.9%, and 64.5% in the low-grade, high-grade, and confluent glands groups, respectively. High-grade atypia was not independently associated with carcinoma compared with low- grade (adjusted odds ratio 1.6, 95% confidence interval 0.6 to 4.0, p � .30), whereas confluent glands showed a strong independent association (adjusted odds ratio 17.0, 95% confidence interval 6.5 to 45.0, p < .001). Inter-observer agreement was sub- stantial for low-grade versus high-grade classification (kappa � 0.76) and excellent for confluent gland identification (kappa � 0.95) Background: Atypical endometrial hyperplasia/endometrioid intra-epithelial neoplasia is the main precursor of endometrioid endometrial carcinoma. Its histopath- ological spectrum is heterogeneous, and inter-observer reproducibility remains sub- optimal. An integrated histological model combining cytological atypia and architectural complexity has been proposed to improve risk stratification for concurrent carcinoma. Objective: To externally validate the prognostic performance and reproducibility of an integrated histological risk stratification model of atypical endometrial hyperplasia/ endometrioid intra-epithelial neoplasia in a multi-center cohort. Methods: This multi-center retrospective cohort study included 201 patients with a pre-operative diagnosis of atypical endometrial hyperplasia/endometrioid intra-epithelial neoplasia who underwent hysterectomy between January 2020 and December 2024 at 3 tertiary referral centers. Index biopsies were independently reviewed by 2 blinded pathologist panels and classified into low-grade atypia, high-grade atypia, and confluent glands. The primary outcome was the rate of concurrent endometrial carcinoma at hysterectomy. Inter-observer agreement and multi-variable logistic regression were used to assess reproducibility and identify independent predictors of carcinoma. Results: Among 201 patients (mean age 57 years), 83 (41.3%) were classified as low-grade, 87 (43.3%) as high-grade, and 31 (15.4%) as confluent glands. Endometrial carcinoma was identified in 42 patients (20.9%). Carcinoma rates were 9.6%, 14.9%, and 64.5% in the low-grade, high-grade, and confluent glands groups, respectively. High-grade atypia was not independently associated with carcinoma compared with low- grade (adjusted odds ratio 1.6, 95% confidence interval 0.6 to 4.0, p � .30), whereas confluent glands showed a strong independent association (adjusted odds ratio 17.0, 95% confidence interval 6.5 to 45.0, p < .001). Inter-observer agreement was sub- stantial for low-grade versus high-grade classification (kappa � 0.76) and excellent for confluent gland identification (kappa � 0.95)