[Predictive value of a combined prediction model for pathologic upgrade to endometrial carcinoma after hysterectomy in patients with endometrial hyperplasia].

Objective: To explore the predictive value of combined prediction model for pathological upgrading to endometrial carcinoma (EC) after hysterectomy in patients with endometrial hyperplasia (EH). Methods: A total of 109 EH patients who underwent hysterectomy in Tianjin Medical University General Hospital from March 2017 to January 2023 were retrospectively enrolled. According to the postoperative pathological results, they were divided into EC group (63 cases) and EH group (46 cases). The clinical data, hysteroscopic manifestations, ultrasound, pelvic magnetic resonance imaging (MRI) and tumor markers were collected. Binary logistic regression was used to analyze the independent risk factors and establish a combined prediction model, which was internally validated by Bootstrap repeated sampling method to evaluate the discrimination and calibration of the combined prediction model (hysteroscopy performance, ultrasound, MRI, tumor markers). The receiver operating characteristic curve and decision curve were drawn to analyze the predictive efficacy and net clinical benefit. Results: (1) Univariate analysis showed that compared with the EH group, the EC group had a higher proportion of diabetes mellitus, a younger age at menarche, a longer duration of irregular menstruation, and a higher proportion of nulliparous and non-pregnant patients (all P<0.05). Under hysteroscopy, the proportion of endometrial polypoid change, irregular polypoid change, tumor necrosis and rupture, brain circuit or dendritic growth and hysteroscopy score in EC group were significantly higher than those in EH group (all P<0.05). Imaging examination showed that the endometrial thickness, detection rate of the mass, the proportion of enhanced endometrial echo, the abundance of endometrial blood flow, the proportion of unclear boundary between the endometrium and the muscle layer, the proportion of increased diffusion-weighted imaging (DWI) signal of the endometrium or the mass, and the proportion of interrupted combined zone in the EC group were higher than those in the EH group (all P<0.05). The results of tumor marker detection showed that the serum human epididymis protein 4 (HE4) level in EC group was significantly higher than that in EH group (P=0.005). (2) Multivariate analysis showed that diabetes mellitus (OR=4.884), enhanced intimal echo under ultrasound (OR=11.234), heterogeneous high signal on DWI (OR=7.031), and elevated serum HE4 level (OR=1.032) were independent risk factors for postoperative pathological upgrading of EH patients to EC (all P<0.05). (3) The combined prediction model was established by combining the above indicators, and the area under the corrected curve was 0.811 (95%CI: 0.720-0.902). The sensitivity of the combined prediction model in predicting postoperative pathological upgrade to EC in patients with EH was higher than that of hysteroscopy score (82.1% vs 44.6%), and the specificity was lower than that of hysteroscopy score (71.0% vs 96.8%). Conclusion: The combined prediction model has a high predictive sensitivity, which provides a reference for predicting the postoperative pathological upgrade of EH patients who plan to undergo total hysterectomy, helps to reduce the risk of missed diagnosis of EC, and is expected to become a supplementary tool for preoperative risk assessment of EH patients who plan to undergo total hysterectomy.