Risk-stratification machine learning model using demographic factors, gynaecological symptoms and β-catenin for endometrial hyperplasia and carcinoma: a cross-sectional study

Abstract

Background: Demographic features, suggestive gynaecological symptoms, and immunohistochemical expression of endometrial β-catenin have a prognostic capacity for endometrial hyperplasia and carcinoma. This study assessed the interaction of all variables and developed risk stratification for endometrial hyperplasia and carcinoma. Methods: This cross-sectional study was conducted from January 2023 to July 2023 at two teaching hospitals in Makassar Indonesia. Patients (< 70 years old) with suggestive symptoms of endometrial hyperplasia or carcinoma or being referred with disease code N.85 who underwent curettage and/or surgery for pathology assessment except those receiving radiotherapy, or chemotherapy, presence of another carcinoma, coagulation disorder, and history of anti-inflammatory drug use and unreadable samples. Demographic, and clinical symptoms were collected from medical records. Immunohistochemistry staining using mouse-monoclonal antibodies determined the β-catenin expression (percentage, intensity, and H-score) in endometrial tissues. Ordinal and Binary Logistic regression identified the potential predictors to be included in neural networks and decision tree models of histopathological grading according to the World Health Organization/WHO grading classification. Results: Abdominal enlargement was associated with worse pathological grading (adjusted odds ratio/aOR 6.7 95% CI 1.8–24.8). Increasing age (aOR 1.1 95% CI 1.03–1.2) and uterus bleeding (aOR 5.3 95% CI 1.3–21.6) were associated with carcinoma but not with %β-catenin and H-Score. However, adjusted by vaginal bleeding and body mass index, lower %β-catenin (aOR 1.03 95% 1.01–1.05) was associated with non-atypical hyperplasia, as well as H-Score (aOR 1.01 95% CI 1.01–1.02). Neural networks and Decision tree risk stratification showed a sensitivity of 80-94.8% and a specificity of 40.6–60% in differentiating non-atypical from atypical and carcinoma. A cutoff of 55% β-catenin area and H-Score of 110, along with other predictors could distinguish non-atypical samples from atypical and carcinoma. Conclusion: Risk stratification based on demographics, clinical symptoms, and β-catenin possesses a good performance in differentiating non-atypical hyperplasia with later stages.