Endometrial abnormalities in breast cancer patients receiving tamoxifen: a retrospective study with a cautionary note on diagnostic thresholds and verification bias

Objective To evaluate endometrial abnormalities in breast cancer patients receiving tamoxifen and assess clinical symptoms and endometrial thickness (ET) while examining verification bias in a real-world retrospective cohort. Methods This single-center retrospective study included 120 breast cancer patients treated with tamoxifen for ≥6 months (January 2018–December 2023). Clinical, ultrasound, and pathological data were collected. Verification-bias analyses used the full cohort (n=120) with assumption-based sensitivity analysis for unverified patients and the pathology-confirmed sub-cohort (n=76) for diagnostic evaluation. Univariate and Firth-penalized multivariable logistic regression, including abnormal uterine bleeding (AUB) and ET ≥9 mm, identified predictors of non-benign lesions (atypical hyperplasia or carcinoma). Unverified patients were not assumed disease-free. Results Among 76 pathologically verified patients, the pathological abnormality rate was 81.58%, mostly polyps (36/76, 47.37%); 12 patients (15.79%) had non-benign lesions. In univariable analysis, AUB was associated with non-benign lesions (26.32% vs. 5.26%, P = 0.009). ET yielded an ROC-AUC of 0.845 (95% CI 0.752–0.938). Although 10 mm had the highest Youden index (0.563), 9 mm was selected as sensitivity-prioritized (sensitivity 83.33% vs. 75.00%). Begg-Greenes correction for the 9-mm threshold gave sensitivity 81.1%, specificity 74.9%, PPV 35.7%, and NPV 95.8% versus naive estimates of 83.33%, 71.88%, 35.7%, and 95.8%. A conservative analysis assuming all unbiopsied patients were benign yielded PPV 25.0%. In the Firth-penalized multivariable model, ET ≥ 9 mm was associated with higher odds of non-benign lesions (OR = 9.62, 95% CI 1.78–52.03, P = 0.009), although the confidence interval was wide; AUB was not statistically significant (OR = 3.85, 95% CI 0.71–20.87, P = 0.122). Conclusions ET ≥9 mm and AUB were associated with non-benign endometrial lesions, but the 9-mm threshold had a low PPV and should serve as a trigger for further work-up rather than a definitive surgical indication. Diagnostic accuracy based only on biopsied patients may be affected by selective verification bias, and dynamic ET monitoring requires prospective validation before clinical use.