Prognostic Factors in Conventional Chondrosarcoma Patients: a SEER Database Analysis

Background: Conventional chondrosarcoma, a rare type of bone tumor, was resistant to chemotherapy and radiation therapy, so more characteristics were needed. In clinic, the use of small series and single-institution studies have limited the investigation of chondrosarcoma. The Surveillance, Epidemiologic, and End Results is the most complete and comprehensive database worldwide. Based on this database, the study aimed to collect clinicopathological features and detect the prognosis of patients with conventional chondrosarcoma. Methods: Clinicopathologic and survival data of 782 patients from 2010 to 2016 and 353 patients from 2007 to 2009 were downloaded and analyzed. Overall survival was analyzed using the Kaplan-Meier method and veried by univariable Cox regression, and independent prognostic factors were assessed using the multi-variable Cox regression hazards model. Then, nomogram was established and the one-, three-, and ve-year survival rates could be calculated with the nomogram. Competitive risk models were conducted to identify prognostic risk factors related to competitive endpoint events in patients with conventional chondrosarcoma. Results: In total, 361 extremities, 360 axial bones, and 61 cranial samples were collected from the 2010 to 2016 cohort. The median survival time of patients with conventional chondrosarcoma was 35 months, and the independent prognostic factors were sex, grade, surgery, AJCC_M age, and tumor size. Thereafter, a nomogram was established based on those independent prognostic factors. The competitive risk model revealed no competitive risk for the cancer specic endpoint event. Those data from 2007 to 2009 were used to validate the results from 2010 to 2016 with general consistency. This retrospective study determined the prognostic factors in patients with conventional chondrosarcoma using the Cox regression hazards model. A nomogram was established to help oncologists assess this rare malignant tumor with low heterogeneity. Conclusions: In the study, independent prognostic risk factors for conventional chondrosarcoma were identied, and a nomogram predicting three- and ve- year overall survival rates were established, which may help physicians to predict the prognosis of patients with chondrosarcoma.


Introduction
Chondrosarcoma, which has an estimated incidence of 1:200000 [1] , is a type of malignant bone tumor characterized by the production of a cartilaginous matrix. Chondrosarcoma accounts for approximately 30% of all primary bone malignancies [2] , second only to osteosarcoma. Conventional chondrosarcoma constitutes approximately 85% of all chondrosarcoma cases, and the remaining subtypes include dedifferentiated [3] , mesenchymal [4] , myxoid [5] , and clear cell chondrosarcoma [6] . The discrepant clinical outcomes of chondrosarcoma depend on histology, grade, and sex. Chondrosarcoma is resistant to chemotherapy and radiotherapy because of its extracellular matrix, low percentage of dividing cells, and poor vascularity [1,7] . The identi cation of IDH1/2 [8] , COL2A1 [9] , and HIF2α [10] in chondrosarcoma may be related to targeted agents, although the future role and effectiveness of which are dismal. Surgical excision remains the mainstay of treatment.
The Surveillance, Epidemiology, and End Results (SEER) database comprises approximately 30% cancer survival data of the US population and is used to describe and analyze prognostic factors of malignant tumors worldwide [11] . The SEER database is important for rare diseases such as chondrosarcoma. Large multi-agency datasets allow the representation of disease statistics and population levels impossible to replicate outside of national databases. Some studies have collected raw data from the SEER database dating back to the 1970s despite the improvements in diagnosis and treatment with time, which may result in bias. In this study, we collected data on conventional chondrosarcoma from the SEER database from 2010 to 2016 as the test group and from 2007 to 2009 as the validation group. Thereafter, we identi ed prognostic factors using univariable and multivariable regression analyses and established a nomogram to predict the three-and ve-year overall survival of chondrosarcoma.

Materials And Methods
This retrospective study utilized the SEER database to acquire cases in accordance with the Declaration of Helsinki (as revised in 2013). All data were collected from 18 local cancer registries, representing the total US population. SEER*Stat [12] (version 8.3.5; National Cancer Institute, Bethesda, MD, USA) was used to acquire clinical data. The following information was collected in a uni ed form for each case: patient ID, sex, year of diagnosis, age at diagnosis, race, histology, stage, grade, type of surgery, cause of death, and COD (cause of death) to site recode. Samples of chondrosarcoma from 2010 to 2016 were collected as the test group and those from 2007 to 2009 were collected as the validation group.
ICD-O-3 [11,13]  found, whereas a small amount of information related to radiotherapy was found and collected. However, the radiotherapy plan was not given; thus, the variable radiotherapy was not added to the information table.
Patient identi cation was eliminated from the SEER database, therefore clinical studies using the SEER database do not require approval from the ethics board. All demographic data and tumor characteristics were identi ed using the χ 2 test for correlations between categorical variables. Kaplan-Meier and Cox proportional hazard analyses were conducted to evaluate the prognostic value of gender, grade, primary site, AJCC_T(American Joint Committee on Cancer_T), AJCC_N, AJCC_M, tumor size, and surgery group for conventional chondrosarcoma.
Nomograms for one-, three-, and ve-year overall survival (not cancer-speci c survival) were established according to the previously mentioned results of multivariable Cox analysis. Then, the cmprsk [11,14] package in R was used to detect and visualize competing risk events. There may be more than one endpoint event for chondrosarcoma, and all other causes of death unrelated to the tumor are referred to as competing risk events. Patients died due to the tumor, the side effects of treatment, or cardiovascular disease. All statistical tests were performed at a level of signi cance of p < 0.05.

Results
In this study, data were collected from the SEER database from which 782 conventional chondrosarcoma patients from 2010 to2016 categorized as the test group and 353 patients from 2007 to 2009 classi ed as the validation group. All the demographic data from both the test and validation groups are listed in Table 1. There was no signi cant difference in the distribution of data between the two groups. The mean age of the test group was 53.05 ± 17.88, and the mean age of the validation group was 50.31 ± 17.94. Sex was a signi cant factor, with male patients having worse prognosis than female patients (p < 0.05) ( Table 1). Regarding histological grade, patients with undifferentiated grades had the shortest survival (HR = 21.88, p < 0.001). In addition, an increase in tumor stage (based on the AJCC system), age, and tumor size resulted in shortened survival for patients. Patients undergoing surgery had prolonged survival. Statistical tests were consistent between the test and validation groups, except for sex and surgery.

Kaplan-Meier analysis and Cox regression model establishment
Kaplan-Meier and log-rank tests were used to compare the differences between variables ( Table 1). Factors that had a better outcome included: tumors with a lower histological grade, smaller tumor size, female sex, and a history of surgery (Fig. 1, Figure S1). Patients with an undifferentiated pathologic grade had the worst survival rate. In this study, a history of radiotherapy was included in the "type of resource" term, but we failed to segregate the variable due to the small sample size of patients undergoing radiotherapy. No differences were observed with respect to primary site.
Based on the a-forementioned results of the univariable analysis, a multivariable Cox regression model was established to nd the independent factors associated with long-term survival. A forest map was then established (Fig. 2). Sex, tumor size, surgery history, distant metastasis, grade, and age were independent prognostic factors for conventional chondrosarcoma patients. All the above results were consistent in the validation group. In the validation test, age and sex were heterogeneous factors ( Figure  S2). Therefore, we searched the "PubMed" database for con rmation, and we found a study that detected that females under 50 years old had a better outcome than those older than 50.

Nomogram prediction establishment
The signi cant prognostic factors generated from the multivariable Cox regression model were matched to establish the nomogram for one-, three-, and ve-year overall survival (Fig. 3). In the nomogram, every variable had a scale value, and the sum of the scores predicted the survival rate. Thereafter, internal validation was performed using the bootstrapping method (parameter: m = 40, b = 1000), and the c-index was 0.85 (Fig. 4).

Competing risk model analysis
In this retrospective study, there were causes of death other than conventional chondrosarcoma. All the above analyses were related to overall survival, rather than cancer-speci c survival; therefore, we conducted a competing risk analysis to differentiate the endpoint events (cancer-speci c and non-cancerrelated death, and death from other causes). In total, 52 patients died from other causes of death including infection, heart disease, and chronic obstructive pulmonary disease. The model (Fig. 5,   TableS1) indicated that the other causes of death had minor effects on the survival of patients with conventional chondrosarcoma.

Discussion
The incidence of chondrosarcoma is increasing: van Praag et al reported that the incidence of chondrosarcoma was 8.78 per million people between 2005 and 2013 in Netherland [15] , which is three times higher than that in the 1990s, and similar data exists for Norway [16] . The most common primary sites of chondrosarcoma are the bones of the limbs and associated joints, pelvic bones, ribs, clavicles, and the vertebral column [15,17] . To date, histological grade may still be the most important factor related to the prognosis of chondrosarcoma. Patients with grade II/III/IV disease have a signi cantly worse prognosis than those with grade I. There are four main pathological forms of chondrosarcoma: conventional, myxoid, juxtacortical, and dedifferentiated [18][19][20] . In our study, 20 patients (2.6%) were diagnosed with a dedifferentiated morphology, which was reported to have a high distant metastasis rate. Nie et al [21] demonstrated that the prognosis of chondrosarcoma was improved when comparing outcomes by decade based on the SEER database. However, chemotherapy and radiotherapy are not the mainstay therapies for chondrosarcoma. The primary goal is to better understand the characteristics of conventional chondrosarcomas since Amer et al [13,22] reviewed and described non-conventional chondrosarcomas. In the present study, we identi ed prognostic factors of chondrosarcoma based on data from the SEER database from 2010 to 2016, which was veri ed using data from 2007 to 2009. Then, we established a nomogram to predict the overall survival of chondrosarcoma patients.
We found that females had a better prognosis in the univariable and multivariable analyses. The effect of sex on the prognosis of conventional chondrosarcomas is controversial. Several studies [9,20,23,24] have reported sex as an independent risk factor based on the SEER database; however, sex did not in uence survival in other studies [25][26][27] . Notably, Laitinen et al [19] found that females of hormone active age with chondrosarcoma had an improved overall survival, and survival was not carried forward in elderly patients. This was consistent with previous studies demonstrating that the estrogen receptor β (ERβ) can be found in bone and endometrial sarcoma. Patients expressing hormone receptors had favorable outcomes [11,28,29] . Further studies are necessary to elucidate the mechanism of sex hormones and chondrosarcoma.
Multivariable analysis showed that age, grade, AJCC_M, tumor size, sex, and surgery were independent risk factors for overall survival of patients with conventional chondrosarcoma, which was in line with previous studies. Surgery remains the optimal treatment for conventional chondrosarcoma and is associated with better outcomes than other treatments. However, the type of surgery for low-grade chondrosarcomas remains controversial. Donati [30] suggested intralesional curettage in grade 1 pelvic chondrosarcoma to reduce operative morbidity, and Verdegaal [31] (2012) reported that the use of phenol as an adjuvant after intralesional curettage of low-grade chondrosarcoma of a long bone had a good prognosis. However, Bus [25] found that wide resection margins were associated with better outcomes compared to intralesional curettage in conventional chondrosarcoma of the pelvis. Hodel found that a wide resection was associated with a better prognosis than an unplanned intralesional resection, and the presence of an anatomical barrier was more important than the metric distance of the surgical margin.
Stevenson [19] reported that local recurrence in patients treated with intralesional curettage was high, which had a signi cant effect on survival, and suggested a 4-mm margin in all grades of chondrosarcoma because inadequate margins were associated with a high local recurrence rate. Therefore, we recommend that wide resection should be performed for all grades of chondrosarcoma.
We Several limitations should be considered in the present study. First, we did not include variables such as speci c tumor location, radiotherapy, and chemotherapy, which were associated with prognosis. Second, the SEER database does not provide information on local recurrence or marginal status. Third, the category of histological grade from the SEER database was different from the World Health Organization classi cation, which may be inaccurate. Finally, this was a retrospective study, which may have introduced bias.
In conclusion, we identi ed risk factors for overall survival of conventional chondrosarcoma based on the SEER database in the modern era, and established a nomogram to predict three-and ve-year overall survival rates. New advances in the treatment of chondrosarcoma are expected, especially for patients with unresectable and metastatic diseases.

Declarations
Ethics approval and consent to participate All data were downloaded from online database, thus no ethical approval and patient consent are required.

Availability of data and materials
All the data used in this study can be acquired from the online SEER database.

Competing interests
The author(s) declared no potential con icts of interest with respect to the research, authorship, and/or publication of this article.
Authors' contributions Z. Gao was responsible for the design. G. Chen and Y. Bao provided the administrative support. P. Fu and T. Sun were for the collection and assembly of data. P. Fu and Z. Gao were responsible for manuscript writing. All the authors approved the nal manuscript.   Internal validation of (A) three-year and (B) ve-year overall survival with calibration curves.

Figure 5
Competitive factors that may affect chondrosarcoma-speci c death with the x-axis representing the survival of months and the y-axis representing cumulative incidence.