Abstract
Objective. To determine the efficacy of clinical characteristics in the prediction of prognosis in patients with ovarian cancer. Methods. Clinical data were collected from 3 datasets from TCGA database, including 1680 cases of ovarian serous cystadenocarcinoma, and were analyzed. Patients with ovarian cancer admitted to our hospital in 2016 were retrieved and followed up for prognosis analysis. Results. From the datasets, for years old at the time of diagnosis, histologic grade and mutation count were good predictors for disease-free survival, while for years old at the time of diagnosis, histologic grade, race, fraction genome altered, and mutation count were good predictors for overall survival. In the patients () retrieved from our hospital, the longest dimension of lesion (cm) and body weight at admission were good predictors for overall survival. Conclusions. Those clinical factors, together with the two predictive equations, could be used to comprehensively predict the long-term prognosis of patients with ovarian cancer.
1. Introduction
Ovarian cancer is the third most common as well as the fifth cause of deaths of gynecologic cancers. The American Cancer Society estimates that in 2022, about 19,880 women will be newly diagnosed of ovarian cancer and about 12,810 women will die from it [1].
Patients with ovarian cancer have different clinical characteristics and prognosis. Based on data collected between 2010 and 2016, only about 20% of ovarian cancers were diagnosed at an early stage, which had a 5-year survival rate of 94%. In contrast, the 5-year relative survival rate of all SEER (Surveillance, Epidemiology, and End Results) stages combined invasive epithelial ovarian cancer patients was only 48%, with localized ones being 93% and distant ones being 31% [2]. Therefore, more efforts are needed to accurately predict the prognosis in later stage ovarian cancer in order to find clues to improve the prognosis.
In this study, we investigated into the clinical characteristics that could be used to effectively predict the prognosis of patients with ovarian cancer.
2. Materials and Methods
2.1. Patient Sources
Patient clinical data were obtained from the TCGA database (https://cancergenome.nih.gov), including the Firehose Legacy dataset (), the Nature 2011 dataset (), and the PanCancer Atlas dataset (). All cases were included for analysis when the corresponding parameter was available. Patients diagnosed with primary ovarian cancer admitted to our hospital from January 1, 2016, to December 31, 2016, were retrieved and followed up for prognosis analysis. The retrospective portion of this study was approved by our hospital’s ethical committee, and informed consents were obtained from the enrolled patients or their family member (if the patients died) during follow-up contact.
2.2. Data Extraction
Two authors independently extracted data and confirmed the accuracy of data. Clinical characteristics, such as age at diagnosis, disease-free survival, overall survival, clinical stages, histologic grades, race, fraction genome altered, Karnofsky performance score, longest dimension of lesion, lymphovascular invasion indicator, primary tumor site, neoplasm status, and mutation count were extracted from the above database when available. Due to the inconsistency in data collection among datasets, the exact number of patients in each parameter might be different.
2.3. Statistical Analyses
Statistical analyses were carried out by a third author. Measurement data were shown as (SD). The Kaplan-Meier survival curve was used to analyze the associations between clinical characteristics and prognosis, including disease-free survival and overall survival. Receiver operating characteristic (ROC) curve was used to illustrate the predictive value of clinical characteristics on 5-year survival. Predictive equation for 5-year survival based on clinical factors was obtained using multinomial logistic regression. All statistical analyses were carried out using SPSS 24.0 (SPSS Inc., Chicago, USA). A value < 0.05 (two-sided) was considered statistically significant.
3. Results
3.1. Clinical Characteristics of Enrolled Patients
There were 1692 patients retrieved from the three TCGA database (Table 1). The mean age at diagnosis was 59.6 years old. Disease-free survival and overall survival were shown using . Clinical stages, histologic stages, race, fraction genome altered, longest dimension of lesion, primary tumor site, and mutation count were further analyzed in corresponding subgroups. There were 38 cases with available data during our follow-up contact, with an average age of years old.
3.2. Value of Clinical Predictive Factors for Disease-Free Survival
Age at years old (), clinical stages ( for overall and subgroups), histologic stage ( for overall and and for stage III), longest dimension of lesion >3 cm (), neoplasm status (), and mutation count ( when >30 and when >50) were significantly associated with disease-free survival (Table 2). The Kaplan-Meier survival curves and ROC curves of corresponding factors are shown in Figures 1 and 2, respectively. According to the area under the curve, neoplasm status showed the best value (0.878) in prediction of long-term disease-free survival.
(a)
(b)
(c)
(d)
(e)
(f)
(a)
(b)
(c)
3.3. Value of Clinical Predictive Factors for Overall Survival
Age at diagnosis ( for overall and subgroups), clinical stages ( for IIC and for IIIA and above), histologic grade ( for overall and for grade III), race (), fraction genome altered ( for the 0.4 and above group, for the 0.5 and above group, and for the 0.6 and above group), neoplasm status (), and mutation count ( for all subgroups) were significantly associated with overall survival (Table 3). The Kaplan-Meier survival curves and ROC curves of corresponding factors are shown in Figures 3 and 4, respectively. According to the area under the curve, clinical stage above III showed the best value (>0.64) in prediction of long-term overall survival. In the patients () retrieved from our hospital, the longest dimension of lesion (cm, ) and body weight at admission () were good predictors for overall survival (Table 4).
(a)
(b)
(c)
(d)
(e)
(f)
(g)
(a)
(b)
(c)
3.4. Predictive Equations for Disease-Free and Overall Survival
In order to obtain a more practical way to predict the prognosis and to test if all factors based on Kaplan-Meier survival curves and ROC curves are good predictors for prognosis, a predictive equation for disease-free survival based on clinical factors was obtained using multinomial logistic regression: where is the probability of death within 5 years, Longest Dimension of if >3 cm and =1 if ≤3 cm, and Neoplasm if cancer lesion remained and =1 if cancer lesion was removed completely.
A predictive equation for overall survival based on clinical factors was obtained using multinomial logistic regression: where is the probability of death within 5 years, Neoplasm if cancer lesion remained and =0 if cancer lesion was removed completely, Diagnosis if >50 years old and =1 if ≤50 years old and Mutation if counted >10 and =1 if counted ≤10.
4. Discussion
In the present study, there were 17 patients diagnosed at clinical stage I, 57 patients diagnosed at clinical stage II, 841 patients diagnosed at clinical stage III, and 168 patients diagnosed at clinical stage IV. The majority of late stage cases showed the importance of identifying accurate predictive factors for prognosis and the possibility of improving the life expectancy and quality based on those important factors.
There have been reports of various biological prognostic biomarkers for ovarian cancer [3–7]. Interestingly, Yang et al. showed that some clinical variables were good predictors [8]. Their findings were based on TCGA OvCa cohort (), and they found that age (>60 years old), nodule of residual disease, tumor status, and clinical stage could significantly predict the prognosis. Our findings, based on 1692 cases from the updated TCGA OvCa cohort, showed that patients >75 years old had a significantly shorter disease-free survival, while patients >50 years old had a significantly shorter overall survival, which showed more challenges for the prognosis of patients diagnosed at a younger age. We also showed that histologic grade and mutation count were good predictors for disease-free survival, while histologic grade, race, fraction genome altered, and mutation count were good predictors for overall survival. The above parameters coincide with some recent reports [9].
Our study also showed that the total number of mutations, when considered as a whole, contributed positively to the long-term survival of ovarian cancer patients, which is consistent with previous studies including only BRCA1 or BRCA2 mutant cases [10]. The underlying mechanisms include different pathways of DNA repair, and more studies are needed when considering all patients not restricted to BRCA1 or BRCA2 mutant cases.
In the data extracted from patients admitted to our hospital, some parameters which were good predictors for prognosis in the published datasets seems to be invalid. This may be due to the smaller number of cases enrolled, the difference in race between the datasets and our own data, and the unavailability of certain parameters in our data.
With the development of big data techniques, data mining from available database has received more and more attention [11, 12]. Therefore, an updated analysis with more available datasets is beneficial in discovery of more valuable predictive factors. Due to the limitation in study design and retrospective manner of data retrieval, details in treatment methods [13, 14], psychological factors, and social-economical factors were missing from the available databases, such as anxiety or depression [15], income, nutrient conditions, and living habits, which could also contribute to the prognosis .
In summary, we showed from published datasets that for patients >75 years old at the time of diagnosis, histologic grade and mutation count were good predictors for disease-free survival, while for patients >50 years old at the time of diagnosis, histologic grade, race, fraction genome altered, and mutation count were good predictors for overall survival. On the other hand, the longest dimension of lesion and body weight at admission were good predictors for overall survival in our own retrieved data. Those clinical factors, together with the two predictive equations, could be used to comprehensively predict the long-term prognosis of patients with ovarian cancer.
Data Availability
The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.
Conflicts of Interest
The authors declare that they have no conflicts of interest.