Development and validation of a prognostic nomogram for ovarian clear cell carcinoma: a study based on the SEER database and a Chinese cohort.

In this study, patients diagnosed with OCCC between 2000 and 2018 in the SEER database were used. Patients diagnosed by autopsy or death certificate, unknown classification, unknown stage, and special classification of cause of death were excluded. Finally, 1855 patients with OCCC were selected. In addition, we collected 256 patients diagnosed with ECO from the Second Affiliated Hospital of Harbin Medical University between 2009 and 2021 and selected 101 patients with OCCC and uninterrupted follow-up (Fig.  1 ). Fig. 1 A flow diagram showing the screening process for the analysis of patients in the SEER and a Chinese cohort A flow diagram showing the screening process for the analysis of patients in the SEER and a Chinese cohort The X-tile software (Yale University, USA) was used to obtain the optimal age cut-off, which is shown in Fig.  2 A–B. In our study, the diagnosis age of OCCC patients in the training set was mostly between 50 and 70 years old (62.2%), while in the validation set, patients were mainly < 50 years old (53.5%) and 50–70 years old (41.6%). Patients with OCCC in the training set were mostly at grade III-IV (86.9%), while patients with OCCC in the validation set appeared to have more I-II than III-IV (86.1% vs. 13.9%). In terms of marital status at the time of diagnosis, 71.1% of OCCC patients were married, divorced, separated, or widowed at the time of diagnosis, compared to 25.4% of OCCC patients who were single. In the validation set, married patients were far more than unmarried patients (94.1% vs. 5.9%). In terms of tumor staging, 60.4% of tumors were found in the SEER database in Figo I stage, 11.9% in Figo II stage, 21.1% in the III stage, and 6.5% in the IV stage. Figo I, II, III, and IV patients were 73.3%, 11.9%, 6.9%, and 7.9%, respectively, in the Second Hospital of Harbin Medical University. Between 2000 and 2018, left-sided and right-sided cases were 45.7% and 41.5%, compared with bilateral cases (68.3%) in our data. In both the training and validation sets, almost all patients with OCCC were treated with surgery and chemotherapy. However, the number of people receiving radiotherapy is generally low (Table S1). Fig. 2 Identification of optimal cut-off values for age characteristics using X-tile software analysis. A Best cut-off value for age and B survival curves for different ages Identification of optimal cut-off values for age characteristics using X-tile software analysis. A Best cut-off value for age and B survival curves for different ages As shown in Table S2, 10 variables were included in the univariate Cox regression analysis. Then, multivariate Cox regression was used to analyze the significant variables. As shown in Table S3, age, marriage, Figo stage, site of disease, and radiotherapy ( p  < 0.05) were identified as independent prognostic factors for OS. However, only Figo was an independent prognostic factor for CSS (Fig.  3 A–D). Fig. 3 Forest plot of Hazard Ratio. A Overall survival of univariate Cox regression. B cancer-specific survival of univariate Cox regression. C Overall survival of multivariate Cox regression. D Cancer-specific survival of multivariate Cox regression Forest plot of Hazard Ratio. A Overall survival of univariate Cox regression. B cancer-specific survival of univariate Cox regression. C Overall survival of multivariate Cox regression. D Cancer-specific survival of multivariate Cox regression Based on the above variables, we construct a visual model to predict OS and CSS for OCCC patients at 1, 3, and 5-year survival probabilities. The total score predicted the OS or CSS probabilities through the nomograms and is obtained by adding the score of each variable (Fig.  4 A–B). Fig. 4 Constructed nomograms for prognostic prediction of overall survival and cancer-specific survival. A Nomogram for overall survival in OCCC patients. B Nomogram for cancer-specific survival in OCCC patients Constructed nomograms for prognostic prediction of overall survival and cancer-specific survival. A Nomogram for overall survival in OCCC patients. B Nomogram for cancer-specific survival in OCCC patients The nomograms from the SEER database and Chinese cohort were evaluated and validated by the C-index, ROC curve, calibration curves, and DCA curve. In the OCCC, the C-index of the nomogram for OS was 0.757 ( p  < 0.001), and for CSS was 0.600 ( p  < 0.001). In the training set, the ROC curve predicted an AUC of 0.803, 0.794, and 0.802 for OS at 1, 3, and 5-year respectively (Fig.  5 A), while the ROC curve predicted an AUC of 0.642, 0.624, and 0.651 for CSS at 1,3, and 5-year in OCCC respectively (Fig.  6 A). The 1, 3, and 5-year calibration curves in the training set for OS and CSS prediction proved to have a satisfying fit (Figs.  5 C, 6 C). As shown in Fig.  5 E and Fig.  6 E, the DCA curves showed a better net clinical benefit. Moreover, DCA curves also showed that nomograms perform well in clinical practice. Therefore, 101 patients in the validation set were included in the nomogram constructed by the training set, and the C index of the nomogram for OS was 0.742 ( p  < 0.001) and that for CSS was 0.700 ( p  < 0.001). In the validation set, the ROC curve predicted the AUC of OS at 1, 3, and 5 years of 0.774, 0.800, and 0.923 respectively (Fig.  5 B), while the ROC curve predicted the AUC of CSS at OCCC at 1, 3, and 5 years of 0.700, 0.773, and 0.963 respectively (Fig.  6 B). The results of the DCA curve and calibration curve also indicated that the modeling of the nomogram is successful (Figs.  5 D, 6 D, F). Fig. 5 1, 3, and 5-year ROC curves for overall survival were compared between the training set ( A ) and validation set ( B ). The 1, 3, and 5-year calibration curves for overall survival in the training set ( C ) and validation set ( D ). The 1, 3, and 5-year DCA curves for overall survival in the training set ( E ) and validation set ( F ) Fig. 6 1, 3, and 5-year ROC curves for cancer-specific survival were compared between the training set ( A ) and validation set ( B ). The 1, 3, and 5-year calibration curves for cancer-specific survival in the training set ( C ) and validation set ( D ). The 1, 3, and 5-year DCA curves for cancer-specific survival in the training set ( E ) and validation set ( F ) 1, 3, and 5-year ROC curves for overall survival were compared between the training set ( A ) and validation set ( B ). The 1, 3, and 5-year calibration curves for overall survival in the training set ( C ) and validation set ( D ). The 1, 3, and 5-year DCA curves for overall survival in the training set ( E ) and validation set ( F ) 1, 3, and 5-year ROC curves for cancer-specific survival were compared between the training set ( A ) and validation set ( B ). The 1, 3, and 5-year calibration curves for cancer-specific survival in the training set ( C ) and validation set ( D ). The 1, 3, and 5-year DCA curves for cancer-specific survival in the training set ( E ) and validation set ( F )

We used data from the SEER Program of the National Cancer Institute ( http://seer.cancer.gov ). Data on patients diagnosed with OCCC from 2000 to 2018 were obtained from the SEER 18 Registries (with additional treatment fields) via the SEER*Stat software, version 8.4.1. We used ovarian cancer patients diagnosed between 2009 and 2021 at the Second Affiliated Hospital of Harbin Medical University as the validation set. The inclusion criteria for the training set are as follows: (1) Years of diagnosis: 2000–2018; (2) Site recode ICD-O- 3/WHO 2008: Ovarian cancer; (3) ICDO- 3 histology codes of histology subgroup were defined as follows: OCCC (8310/3). The exclusion criteria for the training set are as follows: (1) Patients diagnosed with autopsy or death certificate were excluded; (2) Unknown Grade; (3) Unknown stage; (4) Unknown cause special death classification. The clinical information in the validation set was approved by the Ethics Committee of the Second Affiliated Hospital of Harbin Medical University in accordance with the Helsinki Declaration. Clinicopathologic variables for each patient in this analysis included age, grade, marital status, CA125, Figo stage, lymph node metastasis, site of disease, surgery, radiotherapy, and chemotherapy. Overall survival (OS) is defined as the time between diagnosis and death from any cause. Cancer-specific survival (CSS) is defined as the time between ovarian cancer diagnosis and death. These two metrics are the endpoints of this study. First, X-tile software version 3.6.1 was used to calculate the best truncation value for converting continuous variables (such as age at diagnosis) into categorical variables. Univariate Cox regression analysis was used to determine the prognostic factors related to OS and CSS in OCCC patients, and significant variables with p  < 0.05 were included in multivariate Cox regression analysis. The R “rms” software package was used to integrate survival data from Cox regression analysis, and a nomogram was constructed to predict patients'OS and CSS at 1, 3, and 5 years. By integrating multiple predictive factors, multiple scales are drawn to calculate individual survival probability, and the R “pROC” software package is used to generate ROC curves and calculate the corresponding area under the curve (AUC) to evaluate the sensitivity and accuracy of this nomogram. The predictive power of the model is evaluated by using the C-index prior to being calibrated by using the calibration curve. In addition, the"ggDCA"software package was used to generate DCA to evaluate the clinical usefulness of the nomogram. Finally, ROC curves, calibration curves, and DCA curves were performed with data from the validation set to verify the accuracy and practicality of the model.

The 8 prognostic factors we selected were used for constructing models, which is of great significance for predicting survival. Our model helps doctors predict the survival status of patients and assist them in making more accurate choices, enabling the use of different treatment strategies for patients with varying levels of risk. In addition, accurate survival predictions can be used to comfort patients and their families, thereby having a positive impact on treatment.

Ovarian cancer is one of the malignant tumors that seriously harm women's physical and mental health. Ovarian clear cell carcinoma is a highly malignant type of ovarian cancer with a specific histological type and a poor prognosis at late stages [ 21 , 22 ]. Traditional single clinical indicators such as TNM classification and tumor grading are no longer sufficient to meet our requirements for evaluating and predicting patient survival. Meanwhile, nomogram have been widely used as a prognostic device in oncology and medicine. Compared to traditional staging, a nomogram has higher accuracy and a more understandable prognosis. The digital interface can also aid clinical decision-making [ 23 ]. Ma et al. constructed a nomogram based on the SEER database and a Chinese cohort to predict the prognosis of metastatic breast cancer patients of reproductive age for personalized treatment [ 24 ]. Last year, some scholars used the Seer database and their own data to build a prognostic model for vaginal cancer, and through this model, they concluded that surgery combined with external beam radiation plus brachial therapy may be the most recommended treatment choice and constructed a relevant column chart to guide treatment [ 25 ]. For childhood tumors, cancer-specific survival predictions have also been made for Wilms tumors in children, and the reliability of nomogram construction has been verified by internal and external validation [ 26 ]. Zhang et al. explored the risk factors for early death in pancreatic cancer patients with liver metastasis and constructed a nomogram that could predict early death in patients with liver metastasis [ 27 ]. In recent years, due to the rarity of OCCC, a large number of cases have been obtained for review and classification by many scholars, which has become a major difficulty in research. Several clinical retrospective studies have been conducted to establish a nomogram for prognosis prediction. Chen et al. used a nomogram to predict the overall and specific survival of OCCC using SEER data from earlier years, but the prediction was not perfect due to the lack of validation of external data [ 20 ]. Last year, Li et al. collected recent cases of OCCC patients at Union Hospital and constructed a nomogram [ 28 ]. However, there is a lack of generality and reliability due to insufficient patients and the fact that the trained and validated populations are from the same hospital. ​Based on this study, many OCCC patients were collected from the SEER database to explore their pathological features and model the factors that are significant for the prognosis. In addition, we also collected 101 samples from the Second Affiliated Hospital of Harbin Medical University for external validation of the model. To the best of our knowledge, we are the first retrospective study with a large sample to predict overall and specific survival outcomes in OCCC and to validate with external data. In this study, we explore the prognosis of OCCC patients using Figo staging, which is more effective than TNM staging in gynecologic tumors and is more scientifically plausible than previous studies. In addition, we adopted data from the SEER database, which is more recent and has more samples for modeling, so that our model is more consistent with the current diagnosis and treatment of OCCC. In our modeling of overall and specific survival, eight important prognostic factors of overall survival were identified by univariate Cox regression analysis, age, grade, marriage, CA125, Figo stage, site of disease, lymph node metastasis, and radiotherapy. We put 101 patients in the validation set into the nomogram and the predicted survival time showed a high degree of agreement with the actual survival time from our follow-up. The reliability of the results of this study and the level of evidence in existing clinical studies are relatively high, confirming a high level of clinical practicability that can guide clinicians in making clinical decisions. There are several prognostic factors that deserve our attention. For example, married women accounted for the majority of ovarian cancer diagnoses in our cohort, but our univariate analysis showed that unmarried women appeared to have a higher risk than married women (HR = 1.19, p  < 0.05). We hypothesized that it might be related to married women's higher marital happiness. CA125 is a commonly used tumor marker for ovarian cancer and is highly sensitive to the occurrence of the cancer. It has been reported in the literature that 75% of OCCC patients are accompanied by an increase in CA125 [ 29 , 30 ]. This was also verified in our study, but we also encountered some difficulties. For example, the reason that the unknown CA125 state in the group is better than the CA125 negative is unclear. ​The most important point to note here is that our analysis found radiation therapy to be a risk factor for outcomes in OCCC patients, and it is contrary to our understanding. According to a literature review, there is a different academic discussion on whether OCCC performs radiotherapy. A previous report from British Columbia, Canada, suggested that radiotherapy for early-stage ovarian cancer may benefit survival outcomes [ 31 ]. A study by Japanese academics also suggested that local radiotherapy may be one of the treatment options for refractory or recurrent ovarian cancer, with improved survival rates for patients who receive it [ 32 ]. In Yutaka Nagai's report, it was suggested that post-operative radiation therapy may have had a positive effect on specific OCCC patients in recent years, while the long-term prognosis is not yet known [ 33 ]. However, it has been reported in recent years that total abdominal radiotherapy or pelvic lymph node radiotherapy in the early stages of ovarian clear cell carcinoma does not affect the survival prognosis [ 34 , 35 ]. The large disparity in the number of people in our study who received radiotherapy versus those who did not may also be a result of false positives in the data. All in all, this suggests that we need to explore further whether radiotherapy is appropriate for OCCC patients. We have to admit that there are some shortcomings in this study. First, this is a retrospective study, which may have sample bias. Selection bias like this seems inevitable in most studies, but we’ve done a lot of work on it. For example, we included patients from a wider range of years, ensuring that the samples were large enough to reduce selection bias. Second, we set the validation set for validation in our nomograms, but the number of validation sets is small, which may result in some false positive results. Last of all, the SEER database includes most of the resident population of the United States, while the OCCC is a regional disease. The single-center Chinese population is not very applicable as a validation set, and more prospective multicenter studies are needed to further validate the reliability of the model. Therefore, our team will try to fill these gaps in future studies. The nomogram is a theoretical tool, and although our results are very good and the predicted results of the model are consistent with the actual trend, they can only serve as a temporary reference until more research appears.

Epithelial Ovarian Carcinoma (EOC) has the highest mortality rate of any malignancy of the female reproductive system [ 1 ]. EOC has multiple tissue types and varies in origin, pathogenesis, risk factors, and prognosis [ 2 ]. EOC is heterogeneous tumors each histological type has a specific genetic defect that causes the tumor cells to mutate constantly, making it more difficult to explore [ 3 ]. Ovarian clear cell carcinoma, a unique and rare subtype of EOC, accounts for 5 to 10% of ovarian cancers. WHO defines OCCC as a malignancy of the ovaries consisting of cytoplasmic clear, eosinophilic, and nail cells arranged in papillary, tubular, and solid structures [ 4 ]. The incidence of OCCC also varies widely by race and region [ 5 ]. Recent epidemiological and genomic studies have shown that ovarian endometriosis is significantly associated with the formation of OCCC [ 6 , 7 ]. High-Grade Serous Ovarian Carcinoma (HGSOC) is one of the most common and aggressive epithelial tumors of the ovary [ 8 ]. Compared to HGSOC, OCCC has a younger disease age and is often found at an early stage due to its large volume, thus it has a better prognosis than HGSOC at an early stage [ 9 ]. However, once OCCC enters the advanced stage, its prognosis is extremely poor due to its specific treatment regimen, unclear carcinogenic mechanism, and high resistance to chemotherapy drugs [ 10 – 12 ]. OCCC is also an ovarian tumor with a high recurrence rate. First, owing to the lack of a specific molecular target, there is no effective targeted therapy for OCCC with a specific license [ 13 , 14 ]. Meanwhile, surgical treatment remains the primary treatment for OCCC. However, due to its rarity, specific surgical principles cannot be formulated and can only be consistent with the majority of EOC patients, which is also a significant reason for the high recurrence rate in OCCC patients [ 15 , 16 ]. We selected OCCC as our research object because OCCC is a special cancer, and its treatment and prediction methods are quite limited compared to other gynecological tumors [ 17 ]. Due to the unique biological and clinical behavior of OCCC, there is deficient evidence to specify specific diagnosis strategies [ 18 ]. Therefore, there is an urgent need to explore its pathological features, identify prognostic factors, and develop specific interference measures for this malignant tumor. The nomogram is a valid tool to predict cancer prognosis by quantifying individual risk based on clinicopathological variables [ 19 ]. Although Chen et al. developed prognostic nomograms of OCCC in 2020 [ 20 ], our objective in this study was to develop scientific nomograms using the SEER database and OCCC patients from the Second Affiliated Hospital of Harbin Medical University for efficient assessment of overall and specific survival of 1, 3, and 5 years in patients with OCCC under the condition of expanding the sample and updating the data. The model could help clinicians assess the condition and choose appropriate treatment.

Additional file1 Additional file2 Additional file3 Additional file4 Additional file5 Additional file1 Additional file2 Additional file3 Additional file4 Additional file5