- Academic Editor
†These authors contributed equally.
Background: Ovarian clear cell carcinoma (OCCC) is a special pathological type of epithelial ovarian cancer (EOC). Due to its low incidence rate, there is a lack of real-world studies at present. The purpose of the study is to construct a nomogram model for predicting postoperative cancer-specific survival (CSS) of patients with OCCC and analyze in detail the risk factors associated with OCCC. To construct a nomogram model for predicting postoperative CSS of patients with OCCC and analyze in detail the risk factors associated with OCCC. Methods: The clinical pathological data of 596 OCCC patients were collected from the surveillance, epidemiology, and end results (SEER) database from 2010 to 2015. Of these patients, 420 were allocated to the training group and 176 patients to the validation group using bootstrap resampling. The nomogram was developed based on the Cox regression model for predicting the cancer-specific survival probability of patients at 3 and 5 years after the operation. The model was evaluated in both the training and validation groups using consistency index, receiver operating characteristic (ROC), and calibration plots. Results: The independent risk factors for CSS in OCCC patients included International Federation of Gynecology and Obstetrics (FIGO) stage, race, age, tumor laterality, and the log odds of positive lymph nodes (LODDS). The nomograms were established for predicting the 3-year and 5-year CSS of patients after operation. The c-index of the nomogram for CSS was 0.786 in the training group and 0.742 in the verification group. Area under the curve (AUCs) of the 3-year and 5-year ROC curves were 0.818, 0.824 in the training group; and 0.816, 0.808 in the verification group, respectively. Conclusions: Based on the real population data, the construction of the CSS prediction model after OCCC surgery has high prediction efficiency, can identify postoperative high-risk OCCC patients, and can be a valuable aid for the tumor staging system.
Epithelial ovarian cancer (EOC) is the second most common cancer of the female reproductive system and the leading cause of death associated with gynecologic cancers in developed countries [1]. EOC has four main histological types: serous, mucinous, endometrioid, and ovarian clear cell carcinoma (OCCC). The latter is a special tissue type characterized by a relatively young age of onset compared with other EOC subtypes. The average age of onset reported in foreign literature is 55 years old. Moreover, the incidence of OCCC has been reported to be around 11% in the majority of Asian populations, up to 10% in Caucasian women, and even as high as 29.1% in Japan. Although the clinical diagnosis is mostly established early, prognosis evaluation is still controversial [2]. There has been a study suggest that the tumor stage of the International Federation of Gynecology and Obstetrics (FIGO) is the main independent factor affecting the prognosis of OCCC [3]. The 5-year overall survival rate (OS) of FIGO stage I and II is 80%–89%, and that of FIGO stage III and IV is reduced to 52% [4]. Currently, the standard treatment for OCCC includes comprehensive staging or tumor reduction surgery, which requires total hysterectomy, bilateral appendages, and platinum-based combined chemotherapy [5].
FIGO staging system, which is widely used in clinical practice, has been established as a relatively important indicator for evaluating tumor prognosis. However, most researchers and clinicians believe that other important, relevant factors, such as demographic characteristics, residual tumor, venous thrombosis, and surgical methods, should be considered when predicting cancer survival [2]. In recent years, different prediction models after surgery, which are more advantageous than the existing FIGO staging, have emerged, helping clinicians to better predict disease recurrence and the benefit of adjuvant therapy [6]. However, nomograms are rarely used in OCCC. Most previous studies only considered the overall survival (OS) based on Kaplan Meier and COX survival analysis, rarely focusing on cancer-specific survival (CSS).
At present, initial tumor debulking surgery combined with platinum-based chemotherapy is considered the standard treatment for epithelial ovarian cancer. However, the general direction for diagnosing and treating ovarian clear cell carcinoma is still based on high-grade serous ovarian cancer. As this form of cancer is rare, the efficacy of various surgical modalities has not yet been prospectively evaluated [7]. In a previous study that compared the clinicopathological features and survival of OCCC patients with other EOCs, OCCC patients had worse 5-year CSS than serous carcinomas [8]. In a Japanese multicenter retrospective study, Sugiyama et al. [9] also reported that 48.5% of OCCCs were diagnosed as stage I, but only 16.6% of serous carcinomas were diagnosed as stage I. The recurrence rate of IC stage OCCC was as high as 37%, and the survival rate was lower than IC stage serous carcinoma [9]. After entering the advanced stage, OCCC disease progresses rapidly, the chemotherapy resistance rate is high, and the prognosis is worse [4]. Therefore, synchronizing the treatment for OCCC with other epithelial ovarian cancers has certain limitations. Hence, it is necessary to summarize the clinical characteristics of OCCC and identify more accurate solutions for diagnosis, treatment, and follow-up. Retrospective analyses through the surveillance, epidemiology, and end results (SEER) database and real-world research are effective methods for studying low-incidence diseases such as OCCC.
This study selected 11 indicators. FIGO stage is the most clearly prognostic risk factor. Studies have shown that age, race, preoperative carbohydrate antigen 125 (CA125) level, postoperative chemotherapy, and surgical method have a certain correlation with OS, while tumor laterality has rarely been studied. Log odds of positive lymph nodes (LODDS) is a new indicator that can better reflect the condition of lymph nodes. It has not been seen in OCCC prognostic studies.
SEER database (https://seer.cancer.gov/) is a US population-based cancer registry. In the present study, SEER*Stat software version 8.4.0.1 (IMS Inc., Calverton, MD, USA) was used to extract information on patients diagnosed with OCCC between 2010 and 2015.
The inclusion criteria were the following: (1) patients diagnosed with ovarian cancer between 2010 and 2015; (2) patients whose mucinous ovarian cancer was confirmed by pathology and was identified using the site recode ICD-O-3/WHO 2008 (International Classification of Diseases for Oncology, 3rd edition); (3) with morphological codes were C56.9 (ovary); (4) with morphological codes: 8005/3, 8290/3, 8310/3, 8313/3. The exclusion criteria were the following: (1) those with unknown tumor stage, race, laterality, and marital status; (2) no surgical treatment (Rx sum surgprim site field code is 0); (3) with unknown tumor size (CS tumor size was coded as 989, 990, 991, 999); (4) the cause of death (COD) was not ovarian cancer (COD to site record non-ovarian); (5) lymph node test and positive data were unknown (regional nodes examined and positive codes are 96, 97, 98, 99); (6) with unknown CA125 and grade (Fig. 1).
Flowchart of patient selection from the surveillance, epidemiology, and end results (SEER) database (“UNK Stage” means unknown stage, “Stage: I NOS” means unknown Stage IA, IB or IC). CA125, carbohydrate antigen 125; LNE/P, lymph node test and positive data were unknown; COD, the cause of death.
Risk factors for analysis included FIGO stage, race, age, tumor laterality, the log odds of positive lymph nodes (LODDS), surgery, postoperative chemotherapy, preoperative CA125 level, grade, tumor size, and marital status. The outcome variable was cancer-specific survival at the end of follow-up.
LODDS is log (number of positive lymph nodes + 0.05)/(total number of biopsy lymph nodes – number of positive lymph nodes + 0.05). The LODDS range of the modeling group in this study was –0.6~2.38, the tumor size range was 5~800 mm, and the age range was 15~85 years. The cutoff value was selected by X-tile software (version3.6.1; https://medicine.yale.edu/lab/rimm/research/software/).
LODDS was divided into three grades (–0.6~–0.02, –0.01~0.01, 0.01~2.38), tumor size was divided into two categories (5~80 mm, 82~800 mm), and age was divided into two stages (15~49 years, 50~85 years). Fertility-Sparing Surgery (FSS) included unilateral adnexectomy (preservation of the uterus and contralateral ovary) and bilateral adnexectomy (preservation of uterus). Radical surgery (RS) was defined as a complete hysterectomy and bilateral appendages. Codes of FSS in the SEER database were 17, 27, 36, 51, and 56. Meanwhile, codes of RS in SEER database were 25, 26, 28, 35, 37, 50, 52, 55, 57, 70, 71, 72, 73 and 74.
Patients were randomly assigned to the training and validation cohorts in a 7:3
ratio. The primary endpoints were CSS. Categorical variables are expressed as
frequencies and proportions. The clinicopathological characteristics of the
training and validation cohorts were compared using the chi-square test. Through
multivariate analysis of the COX proportional hazards model, related prognostic
factors were identified, and nomograms related to CSS were constructed in
combination with the final independent risk factors. The nomogram was internally
validated, and the Harrell Concordance Index (C-index) of 0.5–1.0 was used to
evaluate the discriminative ability of the nomogram. A calibration curve (1000
bootstrap resamples) was generated to test the agreement between predicted and
actual 3-year and 5-year CSS. The receiver operating characteristic curve (ROC
curve) was used to determine the correctness of the model. Decision curve
analysis (DCA), as a new method, was used to evaluate the potential clinical
value of nomograms. In addition, the entire cohort was risk-stratified, and
Kaplan-Meier analysis was used to explore differences in survival between risk
subgroups. All statistical analyses were performed using SPSS (version 25.0,
SPSS, Chicago, IL, USA) and R software (version 3.6.2;
http://www.r-project.org/). A p value of
A total of 596 eligible patients with OCCC were included in the present study.
Demographic and clinical characteristics are presented in Table 1. The majority
of patients were in the early stage of the tumor (FIGO stage I; 61.07%). A great
number of patients were Caucasian (71.9%), with the onset age of 55–59 years
old (19.9%). Most of the tumors were unilateral, accounting for 87.58%. The
data showed that married women accounted for 56.38% of patients, the tumor size
ranged from 5 to 800 mm; the tumors
Variable | Total | Training group | Validation group | p-value | |
n (%) | n (%) | n (%) | |||
596 | 420 | 176 | |||
Stage | 0.106 | ||||
IA/B | 197 (33.05) | 151 (35.95) | 46 (26.14) | ||
IC | 167 (28.02) | 118 (28.09) | 49 (27.84) | ||
II | 66 (11.07) | 43 (10.24) | 23 (13.07) | ||
III | 128 (21.48) | 85 (20.24) | 43 (24.43) | ||
IV | 38 (6.38) | 23 (5.48) | 15 (8.52) | ||
Race | 0.661 | ||||
Black | 22 (3.69) | 15 (3.57) | 7 (3.98) | ||
White | 429 (71.98) | 288 (68.57) | 141 (80.11) | ||
Other | 145 (24.33) | 117 (27.86) | 28 (15.91) | ||
Age | 0.538 | ||||
158 (26.51) | 117 (27.86) | 41 (23.30) | |||
438 (73.49) | 303 (72.14) | 135 (76.70) | |||
Laterality | 0.445 | ||||
Bilateral | 74 (12.42) | 48 (11.43) | 26 (14.77) | ||
Right | 235 (39.43) | 199 (47.38) | 88 (50.00) | ||
Left | 287 (48.15) | 173 (41.19) | 62 (35.23) | ||
LODDS | 0.288 | ||||
1 | 399 (66.95) | 285 (67.86) | 114 (64.77) | ||
2 | 126 (21.14) | 85 (20.24) | 41 (23.30) | ||
3 | 71 (11.91) | 50 (11.90) | 21 (11.93) | ||
Surgery | 0.842 | ||||
Other | 171 (28.69) | 110 (26.19) | 61 (34.66) | ||
RS | 382 (64.09) | 276 (65.71) | 106 (60.23) | ||
FSS | 43 (7.22) | 34 (8.10) | 9 (5.11) | ||
Chemotherapy | 0.946 | ||||
No/Unknown | 84 (14.09) | 64 (15.24) | 20 (11.36) | ||
Yes | 512 (85.91) | 356 (84.76) | 156 (88.64) | ||
CA125 | 0.092 | ||||
Negative | 162 (27.18) | 122 (29.05) | 40 (22.73) | ||
Positive | 434 (72.82) | 298 (70.95) | 136 (77.27) | ||
Grade | 0.297 | ||||
1 | 9 (1.51) | 6 (1.43) | 3 (1.70) | ||
2 | 59 (9.90) | 42 (10.00) | 17 (9.66) | ||
3 | 320 (53.69) | 226 (53.81) | 94 (53.41) | ||
4 | 208 (34.90) | 146 (34.76) | 62 (35.23) | ||
Size | 0.453 | ||||
173 (29.03) | 122 (29.05) | 51 (28.98) | |||
423 (70.97) | 298 (70.95) | 125 (71.02) | |||
Martial | 0.594 | ||||
Single | 152 (25.50) | 110 (26.19) | 42 (23.86) | ||
Married | 336 (56.38) | 234 (55.71) | 102 (57.95) | ||
Other | 108 (18.12) | 76 (18.10) | 32 (18.18) |
LODDS, log odds of positive lymph nodes; FSS, Fertility-Sparing Surgery; CA125, carbohydrate antigen 125; RS, radical surgery.
The above 11 variables were included in the Cox regression model for
multivariate analysis, and 5 variables, i.e., FIGO stage (p
Clinical features | Total | p-value | ||
Stage | ||||
IA/B | 143 | Reference | ||
IC | 120 | 0.5797 | ||
II | 49 | 0.0002 | ||
III | 84 | |||
IV | 24 | |||
Race | ||||
Black | 15 | Reference | ||
White | 304 | 0.0131 | ||
Other | 101 | 0.0058 | ||
Age | ||||
112 | Reference | |||
308 | ||||
Laterality | ||||
Bilateral | 51 | Reference | ||
Right | 173 | 0.0062 | ||
Left | 196 | 0.0254 | ||
LODDS | ||||
I | 285 | Reference | ||
II | 85 | 0.0901 | ||
III | 50 | 0.0312 |
According to the above variable screening results, 3-year and 5-year CSS
prediction models were constructed, respectively. The proportion of each variable
in the model is shown in Fig. 2. Complex Cox regression analysis was transformed
into visualizations with nomograms. Each variable was plotted at a scale on the
same plane using tick line segments in the nomogram to represent the contribution
of each variable in the predictive model to the outcome event. At the same time,
the 3-year and 5-year survival rates of OCCC patients were clearly obtained from
the nomogram. The nomogram showed that scores increased with increasing tumor
FIGO stage, with the highest scores among all races being achieved by African
American people. OCCC patients aged
Predictive model of CSS. CSS, cancer-specific survival; LODDS, the log odds of positive lymph nodes; W, white; B, black; O, other; L, left; R, right.
The C index is in the range of 0 to 1; the closer the value is to 1, the better the differentiation of patients on the nomogram. For the CSS prediction model in this study, the C-index of CSS predicted by the model in the training set data was 0.786, and the C-index of the CSS predicted by the model in the validation set data was 0.742, which indicated that our model had high accuracy.
The area under the curve (AUC) ranged from 0.5 to 1.0; the closer the value was to 1, the greater the degree of patient differentiation on the nomogram. The AUC values of the 3-year and 5-year ROC curves of the training set were 0.818 and 0.824, respectively (as shown in Fig. 3A,B), and the AUC values of the 3-year and 5-year ROC curves of the validation set were 0.816 and 0.808, respectively (as shown in Fig. 3C,D). The predictive model accuracy was very good.
3-year and 5-year ROC curves of the training and the validation sets. (A) Three-year ROC curve of the training group. (B) Five-year ROC curve of the training group. (C) The validation set 3-year ROC curve. (D) The validation set 5-year ROC curve. ROC, receiver operating characteristic; AUC, the area under the curve; TP, ture positive; FP, false positive.
For the CSS prediction model, the calibration curves of the training set and the validation set are shown in Fig. 4A,B, respectively. The difference between the 3-year and 5-year CSS and the actual CSS was small, and the model accuracy rate was acceptable.
Calibration curves of the training and the validation sets. (A) Training set calibration curve. (B) Validation set calibration curve.
For the CSS prediction model, the DCA curves of the 3-year and 5-year prediction
models compared with the conventional FIGO staging are shown in Fig. 5A,B,
respectively. The 3-year and 5-year CSS prediction models have obvious clinical
benefits compared with the FIGO staging model, where the clinical benefit of the
annual CSS prediction model was particularly prominent when the predicted
probability was
3-year and 5-year CSS DCA curves. (A) Three-year CSS DCA curve analysis. (B) Five-year CSS DCA curve analysis. CSS, cancer-specific survival; DCA, decision curve analysis.
The Kaplan-Meier survival curves of each related factor affecting the prognosis
of ovarian clear cell carcinoma are shown in Fig. 6A–E. This model could
effectively identify patients with postoperative high-risk ovarian clear cell
carcinoma. As the surgical method, postoperative chemotherapy, preoperative CA125
level, and tumor size did not result as independent risk factors in the COX
survival analysis, they were not included in the prognostic model, but we could
be clearly seen on the survival curve that debulking surgery other than FSS and
RS, postoperative chemotherapy, preoperative CA125 positive, tumor size
Kaplan-Meier survival curves of each related factor. (A) CSS high- and low-risk associated survival curves. (B) Survival curves related to CSS surgical methods. (C) Survival curve related to chemotherapy after CSS operation. (D) Survival curve related to carbohydrate antigen 125 (CA125) level before CSS surgery. (E) Tumor size-related survival curve before CSS surgery.
Among 596 OCCC patients included in the present study, the age of clinical
diagnosis, which was mainly 55–59 years old, was relatively lower than in
patients with other epithelial ovarian cancers. In addition, the onset was often
in the early stage of the disease (namely stage I and II), and the tumor volume
of
Through previous literature reading, we selected 11 relevant variables for
current research. Following COX survival analysis, the final modeling variables
were FIGO stage, race, tumor location, age, and LODDS. As the most definitive
diagnosis and treatment basis for EOC, FIGO tumor staging is still feasible in
OCCC [13] for prognosis prediction [14]. FIGO staging also has an important role
in our CSS prediction model. Because the onset of OCCC is mostly in the early
stage, there are many studies on patients with stage I. Herein, we subdivided
stage I into IA/B and IC stages, after which it was found that patients with IC
stage had high CSS scores. FSS is feasible
and whether postoperative chemotherapy is required remain research topics of
interest with a certain degree of interpretation and inclination. In terms of
incidence research, OCCC has a high specificity in Asian populations. In terms of
mortality prediction, although some studies have reported that race is not an
important indicator of progression-free survival (PFS) and OS of the disease
[15], our model clearly showed that African Americans had higher 3-year and
5-year cancer-specific mortality rates, which may be related to many social,
economic, and even cultural reasons, and may also be related to the population
included in the SEER database. In our follow-up studies, we plan to continue to
focus on the effect of race on mortality risk. Multivariate analysis showed no
significant difference between the tumor located on the left and right sides;
however, compared with bilateral ovarian tumors, the 3-year and 5-year survival
rates of OCCC confined to one ovary were significantly improved. A larger tumor
burden was associated with a greater tumor burden, which we believe is closely
related to the FIGO stage of the disease; thus, future studies could subdivide
stages IA and IB. Previous studies have suggested that age is a high-risk factor
for a prognosis for EOC, and the study has suggested that age
OCCC accounts for 5–25% of ovarian cancers, with obvious ethnic and regional differences in incidence [26]. During the development of OCCC, ARID1A and PIK3CA genes frequently mutate, unlike the common mutations in serous carcinoma BRCA1/2. Therefore, standard high-grade serous ovarian cancer treatment is not fully applicable to OCCC [26]. Studying epidemiology, clinical characteristics, diagnosis, and treatment of OCCC in order to obtain a more specific diagnosis and treatment plan is of urgent importance. Also, due to the ethnic differences in its pathogenesis, we need to pay more attention to exploring disease-targeted therapy.
First of all, this study is a retrospective study, and only death and non-death outcomes were assessed, which has some inevitable bias. Tumor reduction satisfaction rates vary widely across cancer centers, and the lack of detailed information on preoperative evaluation and postoperative complications prevents more rigorous comparisons of surgical procedures and their impact. Secondly, chemotherapy data in the SEER database only recorded whether chemotherapy was performed, but there was no detailed protocol and cycle, and it was not clear whether preoperative neoadjuvant chemotherapy was performed.
The incidence of OCCC is low, and the overall prognosis is poor. The prediction model based on tumor FIGO stage, race, tumor location, age, and LODDS performed well in validation and could ideally divide postoperative patients into high-risk and low-risk groups, thus achieving certain clinical reference values.
The data we used were obtained from the publicly available SEER database (https://seer.cancer.gov/).
All authors contributed to the concept and design of the study. Material preparation, data collection and analysis were performed by MH, LL, YLiu and YLi. The first draft of the manuscript was written by MH and all authors commented on previous versions of the manuscript. All authors read and approved the final manuscript. All authors have participated sufficiently in the work and agreed to be accountable for all aspects of the work.
All procedures performed in these studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards. This study was deemed exempt by the Ethics Committee of Nanjing First Hospital, China, since the data we used were obtained from the publicly available SEER database. Written informed consents were exempt since the data we used were obtained from the publicly available SEER database.
We would like to express our gratitude to all those who helped us during the writing of this manuscript. Thanks to all the peer reviewers for their opinions and suggestions.
This research received no external funding.
The authors declare no conflict of interest.
Publisher’s Note: IMR Press stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.