Prognostic modeling of endometriosis-associated ovarian cancer based on molecular signatures: a retrospective study

This study is a retrospective clinical cohort study aimed at establishing a molecular signatures-based risk assessment system for the postoperative recurrence of EAOC, to provide a theoretical basis for its postoperative stratified management. This study complies with the Declaration of Helsinki and was approved by the Ethics Committee of the Beijing Obstetrics and Gynecology Hospital, Capital Medical University (approval number: 2024-KY-026-01). All the patients were resigned into three groups (mismatch repair deficient (MMRd/MSI-H), p53 expression abnormal (p53abn/CNV-H), and P53 wild type (NSMP/CNV-L) and POLE mutation subgroups) based on immunohistochemistry or genetic testing results. As the patients with POLE mutation usually occurred rarely and showed the best prognosis in EAOC [ 10 , 11 ], our study classified the patients with POLE mutation into P53 wide type subgroup. Further, the prognosis among three groups were compared and LASSO-COX method was adopted to construct a prediction model for EAOC recurrence based on molecular signatures. The study collected the medical records and pathological data of patients diagnosed with EAOC who underwent surgery in Beijing Obstetrics and Gynecology Hospital from January 2015 to July 2023. The inclusion criteria were as follows: (1) age 18 years or older; (2) patients with a pathologically confirmed diagnosis of EAOC after surgery; (3) underwent standardized ovarian cancer staging surgery or tumor cytoreductive surgery, (4) have complete clinical data and follow-up data, and (5) patients voluntarily signed informed consent. The exclusion criteria were (1) patients with comorbidities of malignant tumors of other sites or severe medical illnesses, (2) patients with no results of immunohistochemistry or genetic testing, or with other clinical data that were incomplete. The classification of the molecular classification were based on the expression of IHC. dMMR was defined by loss of expression of MSH2, MSH6, PMS2, or MLH1 via immunohistochemistry. P53abn was defined as missense or nonsense mutations in TP53 via IHC (evidenced by nuclear overexpression, complete loss of expression, or cytoplasmic staining). P53wt Defined by the absence of the aboving charateristic expression alteration.The subtyping procedure is conducted as follows: Firstly, immunohistochemical analysis of mismatch repair (MMR) proteins (MLH1, PMS2, MSH2, and MSH6) is performed. If no MMR deficiency is identified, the analysis proceeds to the next step. Secondly, p53 protein expression is assessed via IHC. Cases demonstrating complete absence of staining or strong (3+) nuclear expression in > 70% of tumor cells, or those with TP53 mutations detected by NGS, are classified as high-copy number or P53 abn subtype.Cases showing scattered 1 + positivity or wild-type p53 expression are designated as P53wt. We collected demographic and clinical pathological data from enrolled patients. The patient demographics mainly refer to age, menopause, metabolic disease, and family history of tumor. Clinicopathological data included the following: 1) tumor information: pathological classification, grade, tumor size, lymph node status, lymph-vascular invasion (LVSI), tumor envelope integrity, tumor marker (CA125 and/or HE4), FIGO staging, and unilateral or bilateral ovarian involvement. 2)immunohistochemistry or genetic testing: The mismatch repair (MMR) proteins (MLH1, PMS2, MSH2, MSH6) and p53 status. 3) treatment-related factors: Surgical access, surgical scope, surgical thoroughness, postoperative treatment patterns, and course. The primary outcomes were PFS and OS. PFS was calculated from the day of surgery until tumor recurrence or the end of follow-up. OS was calculated from the day of surgery until death or the end of follow-up.The analysis of the survival were using intention to treat analysis.For cases who were lost to follow-up after tumor recurrence, they were considered deceased according to the principle of intention to treat analysis. Statistical analyses were performed using SPSS 23.0 (IBM, Chicago, II, USA) and R software (version 4.0.2; http://www.Rproject.org ). Descriptive statistical methods were used to summarize the baseline characteristics of clinical cases, including percentage, mean ± standard deviation (SD), median, and interquartile range (IQR). The survival and recurrence rates were obtained by the Kaplan-Meier method and log-rank test. Univariate and multivariate analyses were conducted by regression test to obtain hazard ratio (HR) and 95% confidence interval (CI). A two-sided P  < 0.05 was considered statistically significant. The included clinicopathological characteristics were analyzed by LASSO regression to screen for variables associated with PFS. Multivariate Cox regression was used to select the prediction variables and construct the nomogram. A receiver operating characteristic (ROC) curve was subsequently drawn based on predictive factors, and the sensitivity and specificity were evaluated using the area under the curve (AUC) value. The concordance index (C-index) was used to evaluate the discrimination ability of the nomogram. Calibration curve and decision curve analysis (DCA) were performed to examine the performance characteristics and assess the clinical benefits. Finally, internal validation of the column-line diagram model using the bootstrap method (1,000 repeated samples).

Of 224 initially screened EAOC patients, 6 were excluded due to unavailable immunohistochemical test results, 4 for comorbidities with other malignancies, and 23 for incomplete clinical data, resulting in 191 patients included for final analysis. The mean age at diagnosis was 51.63 ± 9.14 years (range: 24–70 years), with 125 patients (65.4%) showing elevated serum CA125 levels. Histopathological distribution revealed 76 endometroid carcinomas (39.8%) and 115 clear cell carcinomas (60.2%). FIGO stages were distributed as follows: stage I ( n  = 135, 70.7%), stage II ( n  = 23, 12.0%), stage III ( n  = 32, 16.8%), and stage IV ( n  = 1, 0.5%). Molecular classification demonstrated p53wt ( n  = 163, 85.3%), MMRd ( n  = 8, 4.2%), and p53abn ( n  = 20, 10.5%) subgroups (Table  1 ). The 20 patients with p53abn were stage IA [3 cases], IC [8 cases], and stage IIC–IIIC [9 cases], respectively. All 8 cases with dMMR were stage IC to IIIC. (see Supplementary Table 1). Table 1 Baseline characteristics of EAOC patients ( n  = 191) Variables n % Age at diagnosis 51.63 ± 9.14 Tumor size  ≤4 cm 6 3.1  5–10 cm 75 39.3  >10 cm 110 57.6 Pathological Classfication  Endometrioid carcinoma 76 39.8  Clear cell carcinoma 115 60.2 Grade  G1 23 12.0  G2 30 15.7  G3 138 72.3 FIGO staging  I 135 70.7  II 23 12.0  III 32 16.8  IV 1 0.52 Lymph node metastasis  Yes 19 9.9  No 172 90.1 LVSI  Positive 11 5.8  Negative 180 94.2 Ovary side  unilateral 163 85.3  bilateral 28 14.7 Tumor rupture  Yes 146 76.4  No 45 23.6 Residual cancer  R0 190 99.5  R1-R2 1 0.5 CA125  Negative (< 35kU/L) 66 34.6  Positive (≥ 35kU/L) 125 65.4 Molecular typing  P53abn 20 10.5  MMRd 8 4.2  TP53wt 163 85.3 Adjuvant chemotherapy  Yes 177 92.7  No 14 7.3 EAOC endometriosis-associated ovarian cancer Baseline characteristics of EAOC patients ( n  = 191) EAOC endometriosis-associated ovarian cancer All 191 patients underwent comprehensive tumor staging or cytoreductive surgery: 172 via laparotomy ​and 19 via laparoscopy. Fertility-preserving procedures were performed in 8 patients, preserving the uterus and contralateral ovary. Complete surgical resection with no macroscopic residual disease was achieved in 190 patients (99.5%), while one stage IIIC patient had residual tumor < 1 cm. Adjuvant carboplatin-paclitaxel chemotherapy was administered to 177 patients (92.7%), with treatment cycles ranging from 3 to 8 courses. All patients with p53abn received chemotherapy after the surgery.The PARP inhibitors only used in 2 patients of this population. One patients with stage IIIC ovarian clear cell cancer and HRD negative, and after the cytoreductive surgery and adjuvant chemotherapy, the imaging evaluation showed no evidence of residual tumor and she received following nilaparib 200 mg once a day for 3 years. Another patients was stage IC1 ovarian endometoid cancer and underwent pelvic recurrence 11.6 months after surgery. The next generation examination showed she was MSH6 p.R1331* and BRCA2 p.N1784fs. She received recytoreductive surgery and oalaparib maintenance treatment. The median follow-up duration was 35.1 months (interquartile range [IQR] 20.87–63.5), with a median progression-free survival (PFS) of 12.13 months (IQR 8.05–25.82). Recurrence occurred in 29 patients (15.1%), distributed across molecular subtypes as follows: 22 (11.5%) in the p53wt group, 4 (50.0%) in the MMRd group, and 3 (15.0%) in the p53abn group. Recurrence sites included the pelvic cavity (7 cases), abdominal cavity (9 cases), combined pelvic and abdominal cavities (6 cases), distant metastases (6 cases: 3 pulmonary, 2 osseous, 1 lymph node), and 1 case with unknown localization. Both 3-year and 5-year PFS rates for MMRd and p53abn subgroups were identical at 43.75% (95% CI: 10.14–74.19) and 81.25% (95% CI: 52.46–93), respectively. The p53wt group demonstrated 3-year and 5-year PFS rates of 84.96% (95% CI: 77.49–90.06) and 82.54% (95% CI: 74.37–88.31). MMRd patients exhibited significantly reduced PFS compared to p53wt (HR 19.5895% CI: 2.75–139.6; p  = 0.003), with a non-significant trend toward worse PFS versus p53abn (HR 5.461, 95% CI: 0.99–30.17; p  = 0.051). No significant PFS difference emerged between p53abn and p53wt groups ( p  > 0.05; Fig.  1 ). Fig. 1 The survival of EAOC patients with different molecular subgroups. A shows the Kaplan–Meier plot for progression-free survival among p53wt, MMRd, and p53abn subgroups in EAOC patients. B shows the Kaplan–Meier plot for overall survival among P53wt, MMRd, and p53abn subgroups in EAOC patients. A Cox proportional hazards model was used to determine the hazard ratio and 95% confidence interval. Tick marks indicate censored data. Note: p53abn, p53abnormal The survival of EAOC patients with different molecular subgroups. A shows the Kaplan–Meier plot for progression-free survival among p53wt, MMRd, and p53abn subgroups in EAOC patients. B shows the Kaplan–Meier plot for overall survival among P53wt, MMRd, and p53abn subgroups in EAOC patients. A Cox proportional hazards model was used to determine the hazard ratio and 95% confidence interval. Tick marks indicate censored data. Note: p53abn, p53abnormal One mortality occurred from postoperative complications at 0.77 months follow-up. MMRd patients showed 3-year and 5-year overall survival (OS) rates of 85.71% (95% CI: 33.41–97.86) and 42.86% (95% CI: 5.83–77.68), respectively. The p53abn group exhibited equivalent 3-year and 5-year OS rates of 76.19% (95% CI: 47.86–90.46). For p53wt patients, 3-year and 5-year OS rates were 90.27% (95% CI: 83.77–94.26) and 87.22% (95% CI: 79.09–92.34). Mirroring PFS patterns, MMRd patients demonstrated significantly shorter OS versus p53wt (HR 10.63, 95% CI: 1.31–86.04; p  = 0.027), with a non-significant trend toward inferior OS compared toabn (HR 1.686, 95% CI: 0.34–8.34; p  = 0.522). No OS difference was observed between p53abn and p53wt groups (HR 2.804, 95% CI: 0.67–11.79; p  = 0.159; Fig.  1 ). Subgroup analysis by histology revealed consistent patterns. Among endometrioid carcinoma patients, MMRd subtype showed significantly reduced PFS and OS versus p53wt ( p  < 0.05), while p53abn and p53wt groups exhibited comparable outcomes. Only one clear cell carcinoma patient with MMRd was identified, remaining recurrence-free during 5.5 months of follow-up. Similarly, in clear cell carcinoma patients, no significant PFS or OS differences emerged between p53abn and p53wt subgroups (Supplementary Fig. 1). Fifteen predictors were screened for possible association with prognosis, which included age at diagnosis, CA125 level, pathological classification, histological tumor size, FIGO staging, lymph node metastasis, LVSI,​​ tumor envelope integrity, residual cancer, unilateral or bilateral ovarian involvement, molecular signatures (MMRd, p53abn, or p53wt), and postoperative adjuvant chemotherapy. The Lasso regression was used to,​ and the coefficients of these variables are shown in Fig.  2 A. The 10-fold cross-validation method was applied during the iterative analysis, and a model with excellent performance using a minimum number of variables was obtained when λ = 0.044 (logλ = -1.36) (Fig.  2 B). The screened variables included pathological classification, FIGO staging, lymph node metastasis, LVSI, unilateral or bilateral ovarian involvement, and MMRd. These six retained variables were used for multivariate Cox proportional hazards analysis. Among these, pathological classification ( P  = 0.005), FIGO staging ( P  < 0.001),unilateral or bilateral ovarian involvement ( P  = 0.018), and MMRd ( P  = 0.003) were identified as independent predictors for PFS in patients with EAOC (Table  2 ). Therefore, these four variables were selected to construct the nomogram for predicting 1-, 3-, and 5-year recurrence (Fig.  2 C). The C-index of the model was 0.844 (95% CI 0.833–0.861; P  > 0.7), indicating good discrimination. The ROC curves for predicting 1-, 3-, and 5-year recurrence are shown in Suppl Fig. 2, with AUC values of 0.838, 0.912, and 0.898 ( P  > 0.5), respectively, demonstrating good discriminative ability (Suppl Fig. 2A-C). Internal validation of the nomogram through bootstrap resampling (1,000 replicates) showed a C-index of 0.831 (95% CI 0.803–0.834), confirming its robust discriminatory power. Fig. 2 The prediction modality for progression free survival in EAOC patients. A LASSO model was adjusted based on the minimum criteria (regularization parameter λ). B The optimal log value of lambda was indicated by the first black dotted line from the left. C . The prediction nomogram for survival in EAOC patients The prediction modality for progression free survival in EAOC patients. A LASSO model was adjusted based on the minimum criteria (regularization parameter λ). B The optimal log value of lambda was indicated by the first black dotted line from the left. C . The prediction nomogram for survival in EAOC patients Table 2 Cox proportional hazards regression to predict recurrence based on Lasso regression Variables β Z HR (95%CI) P -value Pathological classification(clear cell carcinoma vs. endometrioid) 1.733 2.824 5.66(1.70-18.84) 0.005 FIGO staging(III-IV VS. I-II) 1.025 3.633 2.79(1.60–4.85) < 0.001 Lymph node metastasis(positive and negative) -0.256 -0.417 0.77(0.23–2.57) 0.676 LVSI(positive VS.negative) 0.892 1.637 2.44(0.84–7.10) 0.102 Ovary side(bilateral vs. Unilateral ovary) 1.250 2.373 3.49(1.24–9.79) 0.018 MMRd vs. pMMR 2.192 2.937 8.95(2.07–38.67) 0.003 Cox proportional hazards regression to predict recurrence based on Lasso regression The calibration curves for 1-, 3-, and 5-year recurrence prediction demonstrated excellent agreement between predicted and observed outcomes (Suppl Fig. 2D-F). As shown in Suppl Fig. 3, the decision curve analysis revealed that the model’s clinical utility curve was positioned above both the “none” and “all” reference lines, indicating favorable clinical applicability.

In this study, we investigated the distribution of molecular classifications of endometrial cancer in patients with EAOC and developed a prognostic model based on these molecular signatures. Our findings provide a theoretical foundation for predicting recurrence and supporting clinical decision-making in the individualized management of this population with immunohistochemistry.After the introduction of the TCGA molecular classification system, numerous studies have investigated its prognostic value and utility in guiding postoperative treatment for endometrial cancer [ 4 – 8 ]. Antonio Raffone et al. conducted a meta-analysis of six studies involving 2,818 endometrial cancer patients across TCGA subgroups. Compared with the NSMP group, the pooled hazard ratios (HRs) for mortality were 1.986 (p53abn), 1.192 (MSI), and 0.795 (POLEmut), respectively. For disease-free survival, the corresponding HRs were 2.133 (p53abn), 1.068 (MSI), and 0.325 (POLEmut) [ 10 ].In a secondary analysis of the PORTEC-3 trial, León-Castillo et al. retrospectively evaluated patients and found significant survival benefits from chemotherapy in the p53abn subgroup (5-year recurrence-free survival: 59% with chemoradiotherapy vs. 36% with radiotherapy alone, P = 0.019), whereas no benefit was observed in other subgroups [ 8 ].Similä-Maarala J et al. extended these findings to ovarian carcinomas, comparing molecular classifications in 115 ovarian clear cell carcinomas (OCCs) and 158 ovarian endometrioid carcinomas (OECs). While POLEmut and MMR-deficient (MMRd) OCCs demonstrated excellent prognosis, MMRd OECs were associated with poorer outcomes. The p53abn subgroup showed the worst prognosis across both histotypes, particularly in OCC [ 11 ]. Our EAOC cohort revealed that the MMRd subgroup had the worst prognosis, with 3- and 5-year progression-free survival (PFS) rates of 43.75% and 3- and 5-year overall survival (OS) rates of 85.71% and 42.86%, respectively. Although the p53abn group showed numerically unfavorable outcomes compared to p53 wild-type (p53wt), this difference did not reach statistical significance. In the traditional treatment of ovarian clear cell carcinoma and ovarian endometrioid carcinoma, patients with stage IA/B G1/2 endometrioid cancer, stage IC G1 ovarian endometrioid cancer, and stage IA–IC1 ovarian clear cell carcinoma could choose observation and did not receive chemotherapy. However, in endometrial cancer, patients with p53abn status and myometrial invasion were classified as high-risk and treated similarly to stage III patients. In our study, all 20 patients with p53abn received chemotherapy and only three cases with stage IIIA experienced recurrence. Previous studies have indicated that endometrioid cancers with dMMR and clear cell carcinomas benefit less from chemotherapy, whereas cancers with p53abn derive greater benefit due to genomic instability [ 12 ]. This explains why dMMR and clear cell carcinoma are high-risk factors in this population. In previous studies, the rates of MSI-H/dMMR status were observed in 2/217 OCCC (0.9%), 10/115 OEC (8.7%), and 1/4 mixed cases (25%) [ 13 ]. Another study reported that the distribution among OCCC/OEC was as follows: POLEmut 0.9%/3.2%, MMRd 3.5%/6.3%, p53abn 20%/30%, and NSMP 76%/60% in ovarian clear cell carcinomas and ovarian endometrioid carcinomas [ 14 ]. In a meta-analysis of ovarian cancer, the prevalence of MMR deficiency was highest in the endometrioid subtype (12%), followed by non-serous non-mucinous carcinomas (9%). The rates of dMMR in our study were in accordance with previous reports [ 15 ].All eight cases with dMMR in our study were stage IC to IIIC. Even in stage IC ovarian clear cell carcinoma, the recurrence rate is as high as 36% [ 16 ]. The phase 2 trial KEYNOTE-158 ( NCT02628067 ) evaluated pembrolizumab in microsatellite instability-high and mismatch repair-deficient (MSI-H/dMMR) noncolorectal tumors. Among 373 participants (95% with baseline MSI/dMMR documentation) and after 4.5 years of follow-up, the primary endpoint of overall response rate was 33.8%. Secondary endpoints included duration of response, overall survival, and progression-free survival, which were 63.2, 19.8, and 4.0 months, respectively. Grade ≥ 3 treatment-related adverse events occurred in 50 (13%) participants. With over 5 years of follow-up, responses to pembrolizumab remained durable. Median overall survival was more than twice as long in patients treated with pembrolizumab versus chemotherapy in the first line, despite an effective crossover rate of 62% [ 17 ]. Pembrolizumab remains a standard of care for MSI-H/dMMR metastatic colorectal cance [ 18 ]. These results support the use of pembrolizumab in MSI-H/dMMR tumors. Although arising from different origins, endometrial cancer and EAOC (endometriosis-associated ovarian cancer) share similar histologic subtypes. Studies on endometrial cancer, such as NRG GY018, RUBY, and DUO-E, mainly enrolled patients with recurrent disease or advanced stage (FIGO 2009 stage III or IVA with measurable disease post-surgery, or stage IVB with or without measurable disease). For patients with stage IA/B G1/2 endometrioid cancer, stage IC G1 ovarian endometrioid cancer, and stage IA–IC1 ovarian clear cell carcinoma, chemotherapy rather than observation is recommended for those with p53abn subtypes. The endometrial cancer patients who benefited from ICIs in the above trials were also those with advanced or recurrent disease, where lesions are often located in the peritoneum, ovaries, or distant sites. ICIs have shown efficacy even in pMMR clear cell carcinoma patients [ 19 ].These findings collectively suggest that immunotherapy and endocrine therapy may represent preferable options for EAOC patients with dMMR or p53wt signatures [ 8 , 20 , 21 ]. In addition to the aforementioned molecular signatures, previous studies have identified older age, higher FIGO stage, grade 3 tumors, and suboptimal surgical resection as risk factors for ovarian endometrioid carcinoma [ 22 ], whereas advanced tumor burden, ascites exceeding 400 mL, lymph node metastasis, and bilateral ovarian involvement have been recognized as prognostic determinants for ovarian clear cell carcinoma [ 23 , 24 ]. Consistent with these findings, our study confirmed FIGO stage, histopathological subtype, and bilateral ovarian involvement as independent risk factors for EAOC recurrence.No statistical significance between lymph node metastasis and recurrence was found in this study. One reason is that lymph node metastasis is relatively uncommon in ovarian clear cell carcinoma [ 25 , 26 ] and ovarian endometrioid carcinoma [ 27 ], which is consistent with our finding of 9.9%. As patients with LNM was enrolled into FIGO staging IIIA, its role as a prognostic indicator may be confounded by advanced stage. In this study, 10 of 28 patients with bilateral ovarian tumors had lymph node metastasis and 17 of 28 patients with bilateral involvement were classified as stage IIA-IVB, bilateral ovarian tumor may serve as a potential indicator of metastasis.To stratify high-risk populations, we established a prognostic model incorporating four key variables: histopathological classification, FIGO stage, bilateral ovarian involvement, and MMRd status. The model demonstrated strong discriminative ability with a C-index of 0.844 (95% CI 0.833–0.861) and AUC values of 0.838, 0.912, and 0.898 for predicting 1-, 3-, and 5-year recurrence, respectively. There are also several limitations in this study. First, except for endometrial cancer molecular classification, ARID1A, PI3KCA, and PTEN are also frequently mutated genes and may be involved in the carcinogenesis of clear cell carcinoma and endometrioid carcinoma in both the endometrium and ovaries [ 28 ]. However, the detailed mechanisms of these molecular events in this population remain unclear, and treatments targeting ARID1A, PI3KCA, and PTEN are currently immature. Further studies to elucidate the roles of ARID1A, PI3KCA, PTEN, CTNNB1, and others are important. Second, this study did not investigate the role of PARPi maintenance therapy in this population. Until now, there has been only one study investigating histotype-specific BRCA1/2 or HRD detection in ovarian cancer. Alsop et al. indicated that in their Australian cohort, 8 out of 10 designated endometrioid cancer patients and 3 of 4 clear cell cancer patients identified with BRCA1 or BRCA2 mutations were reclassified as high-grade serous carcinoma following IHC review. Consequently, considerably fewer than 5% of patients with either endometrioid or clear cell carcinoma were found to have a germline BRCA mutation [ 29 , 30 ]. Third, POLE exonuclease domain mutations were not included in the analysis of this manuscript. Although patients with POLE mutations have a better prognosis than the other three subtypes, this omission does not affect the application of our findings, especially in developing countries where universal genetic testing is not yet promoted. In conclusion, based on molecular signatures, we established a validated postoperative recurrence prediction model for endometriosis-associated ovarian This tool enables individualized risk quantification to guide clinical management after primary treatment.

Endometriosis-associated ovarian cancer (EAOC), arising from ectopic endometrium, predominantly manifests as ovarian endometrioid (OEC) and clear cell carcinomas (OCCC), accounting for 10% and 5–25% of epithelial ovarian cancers respectively [ 1 , 2 ]. The management of endometriosis-associated ovarian cancer (EAOC) is primarily based on therapeutic protocols established for serous ovarian carcinoma. However, EAOC exhibits distinct clinical characteristics and tumor biology from serous ovarian cancer [ 3 ]. For example, notably low recurrence rates have been observed in EAOC patients diagnosed with stage IA ovarian clear cell carcinoma (OCCC), whereas significantly higher recurrence rates occur in those with stage IC or advanced disease [ 3 ]. Additionally, EAOC demonstrates poor response to carboplatin-paclitaxel chemotherapy regimens, while hormone therapy may serve as an alternative treatment for stage I grade 1 endometrioid carcinoma. Currently, there is no validated assessment system to predict postoperative recurrence of EAOC or guide subsequent treatment strategies. In 2013, the Cancer Genome Atlas (TCGA) Research Network classified endometrial cancer into four molecular subtypes based on tumor mutation burden and copy number alterations: POLE-mutated (ultramutated), MSI-H (hypermutated), copy-number low (endometrioid; CNV-L), and copy-number high (serous-like; CNV-H) [ 4 ]. To enhance clinical applicability, this molecular classification was subsequently simplified by the Proactive Molecular Risk Classifier for Endometrial Cancer (ProMisE) system, which utilizes​ immunohistochemistry (IHC) to assess mismatch repair (MMR) protein expression and p53 status, combined with sequencing for POLE exonuclease domain mutations [ 4 – 7 ]. Extensive evidence has validated the ​prognostic significance and predictive value for treatment response of these molecular classifications in endometrial cancer [ 4 – 8 ]. Specifically, molecular subtyping enables prognosis prediction in high-risk patients and guides therapeutic decisions. For example, patients with POLE-mutated tumors exhibit an excellent prognosis, while those with p53abnormal (p53abn) subtype derive significant survival benefits from chemotherapy. Notably, the 2020 ESGO/ESTRO/ESP guideline incorporated molecular classification into the risk stratification system for endometrial cancer management [ 9 ]. Until now, there are no recommended molecular markers or predictive models for the recurrence of EAOC and its postoperative treatment selection. EAOC showed a similar morphological and molecular feature to eutopic endometrial cancer. Therefore, we aim to establish a recurrence prediction model for EAOC by incorporating clinical characteristics, tumor information, and molecular mutation signature, which may provide a theoretical foundation for the individualized and precise management of EAOC.

Supplementary Material 1: Supplementary Figure 1. The survival of EAOC patients with different molecular subgroups in different histology subtypes. (A)/(B) The comparison of overall survival and progression-free survival among P53wt, MMRd, and p53abn subgroups in endometroid cancer patients. (C)/(D) The comparison of overall survival and progression-free survival among P53wt, MMRd, and p53abn subgroups in clear cell cancer patients. A Cox proportional hazards model was used to determine the hazard ratio and 95% confidence interval. Tick marks indicate censored data. Supplementary FigureSuppl Fig. 2. The areas under the curve and calibration curves of the nomogram model. (A-C) Receiver operating characteristic (ROC) curves of 1-, 3-, and 5-year progression-free survival for patients with EAOC. (D-F) Calibration plots of predicted 1-, 3-, and 5-year PFS of the established modeling. Supplementary FigureSuppl Fig. 3. Decision curve analysis (DCA) of the nomogram prediction. Model includes MMRd predictors, and model 1 does not include MMRd predictors. Supplementary Material 1: Supplementary Figure 1. The survival of EAOC patients with different molecular subgroups in different histology subtypes. (A)/(B) The comparison of overall survival and progression-free survival among P53wt, MMRd, and p53abn subgroups in endometroid cancer patients. (C)/(D) The comparison of overall survival and progression-free survival among P53wt, MMRd, and p53abn subgroups in clear cell cancer patients. A Cox proportional hazards model was used to determine the hazard ratio and 95% confidence interval. Tick marks indicate censored data. Supplementary FigureSuppl Fig. 2. The areas under the curve and calibration curves of the nomogram model. (A-C) Receiver operating characteristic (ROC) curves of 1-, 3-, and 5-year progression-free survival for patients with EAOC. (D-F) Calibration plots of predicted 1-, 3-, and 5-year PFS of the established modeling. Supplementary FigureSuppl Fig. 3. Decision curve analysis (DCA) of the nomogram prediction. Model includes MMRd predictors, and model 1 does not include MMRd predictors.