3.1 Retrospective design, data sources, and population selection

The data for this study were obtained from the latest Surveillance, Epidemiology, and End Results (SEER) Plus database, specifically the SEER 8, 12, and 17 Registries with additional treatment information submitted in November 2021. Additionally, one female patient diagnosed with MC at the Guangzhou Red Cross Hospital of Jinan University was included in the study. The SEER database is a well-regarded and comprehensive source of cancer-related information, covering approximately 50% of the U.S. population.²³ For this study, records of patients diagnosed with MC between 1990 and 2019 were extracted using the SEER*Stat software (version 8.4.1), with a minimum follow-up period of 4 years.

MC was identified based on the International Classification of Diseases for Oncology third edition (ICD-O-3) code 9110, which encompasses subtypes 9110/0, 9110/1, and 9110/3. Male patients were excluded from the analysis due to the rarity of MC in males, with only four male patients identified in the SEER database with incomplete information.

The inclusion criteria for the study were as follows: (1) MC as the primary malignancy; (2) confirmation of MC through histological diagnosis; (3) availability of complete follow-up data during the course of treatment. On the other hand, the exclusion criteria were as follows: (1) male patients; (2) cases with missing information on important variables; (3) exclusion of repeated cases within the three databases with the same patient ID in the SEER database. The flowchart outlining the selection process of cases is provided in Figure 1.

3.2 Variable declaration and outcome

A total of 64 female MC patients from the SEER database and one patient from the Guangzhou Red Cross Hospital of Jinan University were included. Various variables were extracted for analysis, including age, race, year of diagnosis, primary site, tumor differentiation grade, TMN stage, surgery information, radiotherapy information, chemotherapy information, time from diagnosis to treatment, SEER stage, number of lymph nodes examined/removed/positive, tumor size, site of distant metastasis, International Federation of Gynecology and Obstetrics (FIGO) stage, survival status, cancer-specific death, cause of death, and survival months.

Age was categorized into six groups: ≤40, 41–49, 50–59, 60–69, and 70+. Race was grouped as white, black, and other. Tumor differentiation grade was classified as grade I (well differentiated), grade II (moderately differentiated), grade III (poorly differentiated), and grade IV (undifferentiated or anaplastic). The primary site was divided into cervix uteri, corpus uteri, ovary, other female genital organs, kidney/renal pelvis/urinary bladder, and vagina.

The SEER stage was categorized into localized, regional, and distant based on the SEER database classification. The duration from diagnosis to treatment was classified as less than or more than 1 month. The number of regional lymph nodes examined was regrouped as none, ≤10, 11–20, and ≥21. Tumor size was categorized as ≤5 and >5 cm. Detailed FIGO stages were regrouped into FIGO I, FIGO II, FIGO III, and FIGO IV. Overall survival (OS) in this study referred to the time from the initial cancer diagnosis to cancer-specific death.

3.3 Study design and methods

The entire cohort was randomly divided into a development cohort (70%) and a validation cohort (30%) for the development and validation of the predicting model nomogram. The patient from the Guangzhou Red Cross Hospital of Jinan University was included in the validation cohort. Standardized mean differences (SMDs) were used to assess distributional differences in the baseline variables between the development and validation cohorts.

3.4 The Kaplan–Meier method

The Kaplan–Meier method was constructed to estimate the difference in OS between the development cohort and the validation cohort. Meanwhile, the Kaplan–Meier method was also applied to estimate the difference in OS among different groups of those variables. To avoid bias, patients' age at the year of diagnosis was enrolled as the adjusted variable when analyzing those baseline characteristics such as race, primary site, tumor differentiated grade, the SEER stage, and the FIGO stage. Meanwhile, the SEER stage also served as the adjusted variable, along with age, when analyzing the treatment variables such as surgery/chemotherapy/radiotherapy performed and regional lymph node status.

3.5 Variable selection

The univariate and multivariate Cox regression analyses were applied to identify variables that significantly affected the MC-specific survival.

3.6 Nomogram construction and evaluation

In order to help the clinicians predict the survival probability of MC patients, two nomograms were established. The first nomogram model 1 was established based on the variables selected by the univariate and multivariate Cox regression analyses. Considering distant metastasis has been included in the clinical stage, lung metastasis was excluded when constructing nomogram models. Another simplified nomogram model 2 was established based on several key variables that clinicians would pay attention to. Next, we identified low-risk and high-risk patients by calculating the dichotomous quantiles of total points of the nomogram in the cohort and compared the difference in their survival time using the Kaplan–Meier method. Validation of the nomogram was performed by calculating the concordance index (C-index) and plotting calibration curves by a bootstrapping method with 1000 resamples. Furthermore, the receiver operating characteristic (ROC) curves were drawn to estimate the predictive value by calculating the area under the ROC curve (AUC). Meanwhile, the decision curve analyses (DCA) were conducted to show the clinical effectiveness of the nomogram model.

3.7 Statistical analyses

All the analyses were performed using R software (version 4.21, https://www.r-project.org/). The significance level is set as p < .1 in the Kaplan–Meier analyses and p < .05 in univariate and multivariate Cox regression analyses.

4.1 Case report

A 53-year-old woman who has been postmenopausal for 1 year presented with no clinical symptoms. During her annual routine check-up, elevated levels of CA-125 (44.99 U/mL) and HE-4 (173.3 pmol/L) were detected. Gynecological ultrasound revealed heterogeneous echogenicity within the uterine cavity, with an irregularly shaped mass measuring approximately 55 × 40 × 30 mm and displaying low echogenicity. No significant abnormalities were observed in the bilateral adnexa. Hysteroscopy indicated adhesions in the upper segment of the cervical canal, abnormal uterine cavity morphology, thickened endometrium, and vascular engorgement. Pathological examination following uterine segmental curettage revealed small round cell malignancy in the endocervical and endometrial epithelium. Pelvic magnetic resonance imaging (MRI) confirmed a uterine mass (31 × 36 × 44 mm) protruding into the myometrium at the right anterior margin of the uterus, raising suspicion of endometrial carcinoma(Figure 2A–E). Multiple small lymph nodes were identified in the pelvic region and inguinal area, along with multiple nabothian cysts. Chest CT scan showed minimal fibrous lesions in the right middle lung and a few bullae in the right middle and lower lung. The patient's obstetric history included one vaginal delivery and one miscarriage, and she had no previous history of tumors or genetic diseases.

The patient underwent a laparoscopic total hysterectomy with bilateral salpingo-oophorectomy and pelvic lymph node dissection. During surgery, low-grade carcinoma was discovered in the superficial myometrial layer of the uterus via frozen section analysis. The final pathology report indicated a primary MC originating from the cervix primarily located at the junction of the lower segment of the uterus and cervix. The tumor did not breach the serosal surface, but extensive infiltration was observed with glandular differentiation. Immunohistochemistry studies revealed positive results for Syn, cgA, CD56, and Vimentin, while negative for D2-40, Calretinin, Inhibin-α, CD99, AR, P16, CEA, MC, CK5/6, W-1. Additionally, the Ki-67 index was 20%–30%. Elastic fiber staining was negative. Typical microscopic pathological photos are shown in Figure 2F,G. No significant tumor was found in the bilateral adnexa. The lymph nodes on the left side showed no evidence of metastasis (0/7), as well as the lymph nodes on the right side (0/3). Microscopic examination of the peritoneal deposit in the rectovesical pouch revealed fibrous tissue with crystal deposits and foreign body giant cell reaction, without evidence of tumor. The patient was staged as FIGO IB2. She received 6 cycles of neoadjuvant chemotherapy with docetaxel and carboplatin. Regular follow-up for 4 years has shown no signs of tumor recurrence.

4.2 Patient characteristics

A total of 65 MC patients, who were initially diagnosed between 2005 and 2019 were finally enrolled in the present study. Among those MC patients, 15 (23.08%) patients died from MC, and 6 (9.23%) patients died from other causes, with the average survival time at 84.22 ± 50.66 months, up to date. Among those patients, the most common primary site was cervix uteri (53.85%), followed by corpus uteri (21.54%), other female genital organs (12.31%), ovary (4.62%), kidney/renal pelvis/urinary bladder(4.62%), and vagina(3.08%). 37 (56.92%) patients were diagnosed at stage FIGO I, while 17 (26.15%) at FIGO II, 7 (10.77%) at FIGO III, and 3 (4.62%) at FIGO IV. While being staged according to the SEER stage standard, 33 (50.77%) patients were staged localized, 24 (36.92%) were regional, and 8 (12.31%) were distant. The distribution of the primary site of MC is shown in Figure 3.

The detailed information on those MC patients as well as the difference between the development and the validation cohorts were summarized in Table 1. Tables S1, S2, and S3 showed detailed information on those patients who underwent surgery, radiotherapy, and chemotherapy.

4.3 Kaplan–Meier analysis

As is shown in Figure 4A, there is no significant difference in the OS in those patients between the development and the validation cohorts (p = .9807). Different groups of race, primary site, tumor differentiated grade, and the FIGO stage also showed no different OS in the Kaplan–Meier analysis (Figure 4B–E). However, a significant difference was found in patients' survival time when MC patients were staged by the SEER stage standards as localized, regional, or distant, which indicates that the SEER stage standards might be a better staging system in the MC patients to distinguish their prognosis (Figure 4F, p = .0835).

Additional Kaplan–Meier analyses were applied to the effect of different treatments on the survival time in those patients. We first analyzed the influence of the waiting time from diagnosis to begin treatment. As is shown in Figure 4G, patients benefited from less than 1 month's waiting to begin their treatment (p = .0320). Figure 4H showed that those who accepted surgery had a better survival time (p = .0678). We also compared the influence of different solutions of regional lymph nodes and found that MC patients benefited from regional lymph node examination (Figure 4I, p = .0608). Moreover, patients who found more than 1 regional lymph node positive (Figure 4J, p = .0662), or received regional lymph node resection (Figure 4K, p = .0807) had a poorer survival time. Radiotherapy (Figure 4L, p = .0431), as well as chemotherapy (Figure 4M, p = .0488), showed a significant influence on the survival time in the MC patients.

4.4 Univariate and multivariate Cox regression analysis with cancer-specific death

Univariate and multivariate analyses were applied to estimate the effect of those variables on cancer-specific death. As is shown in Table 2, the SEER stage, more than 1-month waiting from diagnosis to begin treatment, surgery performed, ≤10 lymph nodes examined, regional lymph positive, lung metastasis, and tumor size larger than 5 cm showed significant influence on cancer-specific death through univariate analysis. As for the multivariate Cox regression analysis, age at the year of diagnosis, race, primary site, tumor differentiated grade, the SEER stage, more than 1-month waiting from diagnosis to begin treatment, surgery, chemotherapy, radiotherapy, lymph node resection/examination/positive, lung metastasis, the FIGO stage shown significant influence to the cancer-specific death in MC patients.

4.5 Nomogram construction and validation

The first nomogram model 1 was established based on those significant variables selected by the multivariate Cox regression analysis, which were age, race, primary site, tumor differentiated grade, the SEER stage, surgery, radiotherapy, chemotherapy, and the FIGO stage to predict the 3-to-8-year survival probability of MC patients (Figure 5A). Meanwhile, a simplified nomogram model 2 was established based on age, the primary site of the tumor, tumor differentiated grade, the SEER stage, and the FIGO stage for patients' estimation before treatment (Figure 5B). The C-index of model 1 was 0.921 in the development cohort and 0.720 in the validation cohort, while model 2 had a C-index of 0.848 in the development cohort and 0.640 in the validation cohort. Model 1 achieved a good predictive value in the ROC analyses to predict a 3-year (Figure 6A, AUC = 0.936), 5-year (Figure 6B, AUC = 0.903), and 8-year (Figure 6C, AUC = 0.875). Moreover, model 2 also achieved a good predictive value in the ROC analyses to predict a 3-year (Figure 6D, AUC = 0.843), 5-year (Figure 6E, AUC = 0.821), and 8-year (Figure 6F, AUC = 0.796) survival probability. The calibration curves showed both model 1 (Figure 6G) and model 2 (Figure 6H) had good agreement between predicted probability and the observed outcome. The DCA analyses also showed that model 1 (Figure 7A,B) and model 2 (Figure 7C,D) had good discrimination both in the development and validation cohort.

Furthermore, we identified the patients into low-risk and high-risk groups at the dichotomous quantiles of total points of the nomogram total points in the whole cohort, which cut-off points were 181 in model 1 and 182 in model 2, respectively, and compared the difference of the survival time between them. Figure 8A showed that there was a significant difference in survival time between the high-risk and low-risk patients (p = .00019), based on the nomogram model 1. Meanwhile, Figure 8B also showed the significant survival time difference between the high-risk and low-risk patients (p = .00058), based on the nomogram model 2.

Integrated discrimination improvement (IDI) and net reclassification improvement (NRI) were enrolled to compare the predicting and discrimination capability between model 1 and model 2. As is shown in Table 3, there was no significant difference between model 1 and model 2 in predicting the 3-year survival probability (IDI = −0.261, p = .897; NRI = −0.479, p = .226), the 5-year survival probability (IDI = −0.285, p = .837; NRI = −0.634, p = .286), the 8-year survival probability (IDI = −0.255, p = .844; NRI = −0.524, p = .166). The detailed points of these two nomograms are shown in Figures S1 and S2.