Nomogram for predicting poor outcomes after hysteroscopic endometrial polypectomy in menstrual disorders

The risk model constructed in this study can effectively predict the risk of poor outcome of hysteroscopic endometrial polypectomy in treating menstrual disorders, especially for patients with longer menstrual duration and larger polyps. And it provides a feasible tool for clinical practice, and the constructed column chart also enhances the clinical application value of the risk model.

Endometrial polyps usually refer to local hyperplasia of the endometrium, which is common in women of childbearing age or postmenopausal women and is a benign lesion. [ 1 , 2 ] Endometrial polyps are usually composed of glands of the endometrium, stroma, and blood vessels, with varying sizes, and may be single or multiple. [ 3 ] Many patients with endometrial polyps do not have obvious symptoms in the early stages. However, as the condition worsens, it may lead to symptoms such as excessive menstruation, irregular menstruation, abnormal bleeding, and even affect fertility in severe cases. [ 4 , 5 ] The relationship between endometrial polyps and infertility has also received widespread attention. [ 6 ] Some studies have shown that endometrial polyps may affect embryo implantation and development, so timely treatment is particularly important. Hysteroscopic endometrial polypectomy is a minimally invasive surgery mainly used to treat endometrial polyps. Through hysteroscopy, doctors can directly observe the condition of the uterine cavity and remove polyps from the endometrium using small surgical instruments. [ 7 ] This surgical method has the advantages of less trauma, faster recovery, and shorter hospital stay compared to traditional open surgery. Hysteroscopic endometrial polypectomy, as an effective method for treating menstrual disorders caused by endometrial polyps, has been widely used in clinical practice. [ 8 ] Most patients have significantly improved their menstrual flow and cycle after surgery, especially in cases of excessive menstruation. [ 9 , 10 ] However, some patients may still experience irregular menstruation after surgery, mainly manifested as unstable menstrual cycles or abnormal menstrual flow. [ 11 ] The occurrence of irregular menstruation after surgery is related to multiple factors, including the size and location of polyps, whether the resection is complete during surgery, and preoperative hormone levels. At present, research on hysteroscopic endometrial polypectomy mainly focuses on comparing it with other treatment methods or analyzing the efficacy of combined drug therapy. [ 12 , 13 ] Research on complications mainly focuses on postoperative recurrence and infertility. [ 14 , 15 ] There is relatively little research on the effectiveness of treating menstrual disorders. This study aims to construct a risk model that can effectively predict poor outcome through retrospective analysis. We will analyze the clinical characteristics, surgical process, and changes in hormone levels before and after surgery of patients, screen important features that affect poor outcome, and use support vector machine (SVM) to construct a risk model. Further explore the impact of risk models on poor outcome through multivariate analysis, and construct a nomogram with clinical application value. I hope this study can provide effective tools for clinical practice to help doctors identify high-risk patients and intervene early, thereby improving the treatment effectiveness and quality of life of postoperative patients.

Data curation: Chunzhi Sheng, Zhaobo Yan, Fanrong Zhao. Methodology: Chunzhi Sheng, Zhaobo Yan, Fanrong Zhao. Resources: Chunzhi Sheng, Zhaobo Yan. Software: Chunzhi Sheng. Writing – original draft: Chunzhi Sheng. Writing – review & editing: Chunzhi Sheng, Fanrong Zhao.

This study is a retrospective analysis, and the research subjects are patients who underwent hysteroscopic endometrial polypectomy for the treatment of menstrual disorders caused by endometrial polyps at our hospital between June 2021 and June 2024. The inclusion criteria were: (1) premenopausal women aged between 18 and 50 years; (2) diagnosed with endometrial polyps confirmed by ultrasound or hysteroscopy; (3) successful completion of hysteroscopic polyp resection. The exclusion criteria included: (1) history of malignant tumors; (2) presence of other uterine diseases, such as uterine fibroids or adenomyosis; (3) simultaneous surgery; (4) incomplete clinical data. A total of 450 patients were included. On the day before surgery, oral misoprostol (200 μg-400 μg) was taken to soften the cervix, and fasting and water deprivation were performed. Combined lumbar epidural anesthesia was used, with the patient in the lithotomy position. 0.9% sodium chloride injection with a flow rate of 200 to 400 mL/min was used as the dilating agent, and the intrauterine pressure was maintained between 80 and 100 mm Hg. A hysteroscope was inserted to conduct a detailed examination of the structure inside the uterine cavity, confirming the number, size, location, and morphology of polyps. Use an electrosurgical loop to remove the base of the polyp, with the cutting mode power of 60 to 120 W and the coagulation mode power of 30 to 60 W. After polypectomy, a hysteroscope is used for further examination to ensure complete removal of the polyp, and residual tissue or blood in the uterine cavity is cleared. The surgery is completed. Collect basic demographic and clinical data of all participants and record the characteristics of polyps. In addition, surgical related indicators were also recorded. The patient’s hormone levels, including estradiol (E2), follicle stimulating hormone (FSH), luteinizing hormone (LH), and progesterone (P4), were measured during the last follicular phase before surgery and the first follicular phase after surgery to evaluate changes in hormone levels. Based on the criteria of the International Federation of Gynecology and Obstetrics (FIGO) and the World Health Organization. For menstrual cycle, normal (0 points): 21 to 35 days; mild disorder (1 point): 18 to 21 days or 35 to 40 days; severe disorder (2 points):  40 days. For menstrual volume, normal (0 points): 30 to 80 mL; mild disorder (1 point): 80 to 100 mL or  100 mL or amenorrhea. For menstrual duration, normal (0 points): 3 to 7 days; mild disorder (1 point): menstrual duration between 2 and 3 days or 8 to 10 days; severe disorder (2 points): menstrual duration <2 days or more than 10 days. The total score of these 3 indicators was used to evaluate the severity of menstrual irregularities. Preoperative and 6-month postoperative assessments of the degree of menstrual disorders in patients will be performed. If the total score of the patient increases after surgery, they will be categorized into the menstrual disorders group; otherwise, they will be categorized into the menstrual regularization group. In this study, menstrual disorders refer to the condition where the degree of menstrual disorders remains unchanged or worsens postoperatively, while menstrual regularization refers to the condition where the degree of menstrual disorders improves after surgery. Categorical variables were compared using chi-square tests and presented as frequency (percentage). Continuous variables were analyzed using either independent t-tests or Mann-Whitney U tests, depending on data distribution, and were presented as median (range). The hormonal change was calculated by subtracting the preoperative hormone levels from the postoperative hormone levels. These change variables and other significant indicators were used as independent variables in the SVM analysis. Data were divided into training and test sets with a 7:3 ratio, and the SVM parameters were tuned using the “tune” function in the training set. The model performance was validated in the test set using ROC(receiver operating characteristic) curves, F1 Score, Accuracy, Recall, and Precision. Visualization of the SVM results was performed using the “iml” package, and feature importance was ranked. Based on the top 5 most important features and their corresponding expression levels, a risk model was constructed, with the specific formula as follows: The multivariable analysis risk model and the interaction effects on poor outcomes were evaluated. A nomogram was constructed based on the risk model and significant factors from the multivariable analysis in both the training and testing sets. Calibration curves, ROC curves, and DCA(decision curve analysis) curves were drawn to assess the accuracy of the nomogram.

Postoperatively, 95 patients experienced menstrual disorders, while 355 patients had regular menstruation. The results showed that the menstrual disorders group was affected by age BMI, The duration of menstruation, endometrial thickness, and polyp size were significantly higher in the group with regular menstruation (P values were 0.0471, 0.0404, 0.00891, 0.00612, and 0.00144, respectively). In addition, the proportion of stem polyps in the menstrual disorders group was significantly higher than that in the menstrual regularization group ( P  = .0131), and the proportion of polyps located on the posterior wall of the uterus was also higher ( P  = .0182). In addition, there were no significant differences between the 2 groups in terms of age at menarche, degree of dysmenorrhea, history of miscarriage, delivery history, number of polyps, as well as surgical time, intraoperative blood loss, first activity time, and length of hospital stay (Table 1 ). Difference in demographic characteristics, clinical and surgical features between postoperative menstrual regularization and menstrual disorders groups. The difference in preoperative estradiol levels between the 2 groups was relatively small (110.44 pmol/L in the menstrual regularization group and 119.00 pmol/L in the menstrual disorders group, P  = .039). The postoperative E2 level significantly decreased, and the decrease in the menstrual regularization group was higher than that in the menstrual disorders group (95.98 pmol/L in the menstrual regularization group and 113.13 pmol/L in the menstrual disorders group, P  = .00831). There was no significant difference in preoperative FSH levels between the 2 groups (4.94 mIU/mL in the menstrual regularization group and 4.64 mIU/mL in the menstrual disorders group, P  = .135). The postoperative FSH level significantly increased, and the postoperative FSH value in the menstrual disorders group was higher than that in the menstrual regularization group (5.44 mIU/mL in the menstrual regularization group and 6.45 mIU/mL in the menstrual disorders group, P  = 8.50E-06). There was a slight difference in preoperative LH levels between the 2 groups, with the menstrual disorders group having slightly higher preoperative LH levels (13.13 mIU/mL vs 11.08 mIU/mL, P  = .0872), but the difference was not significant. The postoperative LH level decreased, and the postoperative LH level in the menstrual disorders group was significantly higher than that in the menstrual regularization group (12.82 mIU/mL vs 9.55 mIU/mL, P  = 2.14E-05). The hormone levels in the group with preoperative menstrual disorders were slightly higher (1.39 ng/mL vs 1.27 ng/mL, P  = .0499). Postoperative P4 levels decreased, with a greater decrease observed in the menstrual regularization group (0.87 ng/mL vs 1.07 ng/mL, P  = .000233) (Table 2 ). The changes in sex hormones before and after hysteroscopic endometrial polypectomy. FSH = follicle stimulating hormone, LH = luteinizing hormone. The optimal parameter combination is best cost = 2, best gamma = 0.0625, build SVM model using these parameters, The AUC value is 0.877, indicating high predictive ability. The confidence interval for AUC is between 0.811 and 0.944, indicating high stability (Fig. 1 A). The other parameters also indicate the accuracy and comprehensiveness of the SVM model. The SVM results were visualized and analyzed using the IML package, and sorted by feature importance. The results showed that P4 Change, LH Change, The morphology of the polyp (sessile polyp), The morphology of the polyp (pedunculated polyp), The location of the polyp (Anterior wall of the uterine body) For the first 5 features (Table 3 ) (Fig. 1 B). Summary of ROC curve parameters. CI= confidence interval, ROC = receiver operating characteristic, SVM = support vector machine. (A) ROC curve of SVM model. (B) Visualization results of SVM model. ROC = receiver operating characteristic, SVM = support vector machine. A risk model was constructed based on the importance values and corresponding expression levels of the first 5 features. The AUC value of the model was 0.701, indicating its good ability to distinguish different categories, and the confidence interval was 0.645 to 0.758, indicating that the results were relatively stable. The threshold is −7.933, Indicating a high risk of poor outcome. On the contrary, it indicates a lower risk of poor outcome. The sensitivity is 0.747, indicating that the model can effectively identify patients with poor outcomes (Fig. 2 ). ROC curve of risk model. ROC = receiver operating characteristic. The results showed that age had a significant impact on the research outcomes ( P -value = .015), with an OR of 1.008, indicating that for every year of age increase, the risk of poor outcome increased by 0.8%. BMI has a significant impact on the results ( P -value = .020), with an OR of 1.009, indicating that for every 1kg/m ² increase in BMI, the risk of poor outcome increases by 0.9%. The duration of menstruation was significantly correlated with the outcome ( P -value = .003), with an OR of 1.027, indicating that for every additional day of menstrual duration, the risk of poor outcome increased by 2.7%. Endometrial thickness has a significant impact on the results ( P -value = .002), with an OR of 1.026, indicating that for every 1mm increase in endometrial thickness, the risk of poor outcome increases by 2.6%. The size of polyps was significantly correlated with the outcome ( P -value = .007), with an OR of 1.011, indicating that for every 1mm increase in polyp size, the risk of poor treatment increased by 1.1%. The changes in FSH (FSH) were significantly correlated with the results ( P -value = .023), with an OR of 0.970, indicating that for every 1 mIU/mL increase in FSH changes, the risk of poor outcome decreased by 3%. The P -value of the risk model is .000, indicating its significant predictive effect on the results. The OR is 1.020, indicating that for every 1 unit increase in risk score, the risk of poor outcome increases by 2%. The interaction analysis of the risk model shows that the risk model is related to age, BMI, There is no interaction between endometrial thickness and FSH changes, which means that the risk model is not affected by these factors. There is a significant positive correlation with menstrual duration, and the OR value is greater than that of the risk model alone, indicating that the risk model is more effective in patients with longer menstrual duration. The risk model also showed a significant positive correlation with polyp size, with an OR value greater than that of the risk model alone, indicating better performance in patients with larger polyps (Table 4 ). Multivariable regression analysis risk model and the predictive effect of the interaction of the risk model. BMI = body mass index, CI = confidence interval, FSH = follicle stimulating hormone, OR = odds ratio. The results show that the risk model significantly predicted poor outcomes in both the training and testing sets, and the patient’s age, menstrual duration, polyp size, and FSH changes were significant in the training set. In the test set, the duration of menstruation and polyp size were significant (Fig. 3 A and B). By adding up these scores, we can obtain a total score that is associated with the probability of poor outcome, for example, the probability of poor outcome in the training set is 17.7%, and the probability of occurrence in the test set is 8.87%. According to the calibration curve, ROC curve, and DCA curve of the model, it is shown that the predicted risk is consistent with the actual occurrence, indicating that the column chart has high accuracy and clinical application value (Fig. 4 A–F, Table 3 ). (A) Nomogram model of the training set. (B) Nomogram model of test set. (A) Calibration curve of the nomogram model in the training set. (B) ROC curve of the nomogram model in the training set. (C) DCA curve of the nomogram model in the training set. (D) Calibration curve of the nomogram model in the test set. (E) ROC curve of the nomogram model in the test set. (F) Test the DCA curve of the nomogram model in the test set. DCA = decision curve analysis, ROC = receiver operating characteristic.

This study used a SVM model for feature screening and constructed a risk model for predicting the poor outcome of hysteroscopic endometrial polypectomy in treating menstrual disorders. The occurrence of poor outcome is closely related to multiple factors. In addition to risk models, FSH changes, polyp size, patient age, body mass index (BMI), preoperative menstrual duration, and endometrial thickness are significant predictive factors. And there is a significant positive correlation between the risk model and polyp size, as well as preoperative menstrual duration. The important feature we selected is P4 Change, LH Change, The morphology of the polyp (sessile polyp), The morphology of the polyp (pedunculated polyp), The location of the polyp (Anterior wall of the uterine body). When postoperative progesterone and LH levels return to normal from elevated levels, the risk of poor outcomes is relatively low. Progesterone is one of the key hormones that maintain the normal functioning of the menstrual cycle. [ 16 ] When progesterone levels decrease from elevated levels, it may prevent excessive endometrial hyperplasia or prolonged secretory phase, thereby reducing the occurrence of menstrual disorders. [ 17 ] LH mainly regulates the process of ovulation. [ 18 ] If LH returns to normal after surgery, it indicates that the hormone balance in the body tends to stabilize after surgery, and menstruation is more likely to become regular, as excessive LH levels may lead to ovulation disorders and menstrual irregularities. [ 19 ] The different shapes of polyps may affect the difficulty of surgery. pedunculated polyps are usually easier to remove than sessile polyps, which may cause endometrial damage and increase the risk of poor treatment outcomes. [ 20 ] We also found that the risk model had better predictive performance in patients with longer preoperative menstrual duration and larger polyps. This may be because the duration of menstruation is often closely related to the growth of the endometrium and fluctuations in hormone levels. The risk model includes hormone changes such as progesterone and LH, which makes it more sensitive and specific in predicting menstrual duration in patients with longer menstrual periods. The size of polyps may affect the endometrial reactions and bleeding they cause. Larger polyps may lead to local hyperplasia or irregular shedding of the endometrium, and can cause more severe endocrine imbalances, thereby affecting sex hormone levels. The risk model includes hormonal changes, which can better identify and predict postoperative menstrual problems in patients with larger polyps. The nomogram we constructed in both the training and test sets showed significant predictive performance, indicating that nomogram can provide personalized predictions based on the specific situation of patients. This provides more accurate risk prediction for clinical treatment, helping doctors choose appropriate treatment plans based on the actual situation of patients. Knowing the risk of occurrence can reduce the risk of poor efficacy through early intervention measures such as hormone therapy, further examination, etc It provides a reliable tool for clinical decision-making. The risk model constructed in this study can accurately predict high-risk patients by integrating multiple clinical and preoperative/postoperative hormone changes. The predictive ability of the model is strong, with an AUC value of 0.701, indicating its good clinical application potential. As a visualization tool, nomograms facilitate doctors to quickly assess patient risks and take corresponding preventive or intervention measures. However, there are also certain limitations. This study is a retrospective analysis with a relatively limited sample size, and there may be some selection bias in the data sources. Future research can further validate the effectiveness and stability of this model through prospective cohort studies, and explore how to further optimize the predictive model.