Risk stratification for central conventional chondrosarcoma of bone: A novel system predicting risk of metastasis and death in the Cancer Registry of Norway cohort

Interobserver variability in histological grading of central conventional chondrosarcoma (CCCS) limits the quality of patient information and research progression. We aim to quantify known and new prognostic variables and propose a risk stratification model.

Many cohorts with central conventional chondrosarcoma (CCCS) alone are limited to specific anatomical locations or with multivariate analyses for limited variables in institutional cohorts. 6,7,18 Histological grade significantly predicts survival at univariate analysis in two national CCCS cohorts, 14,19 but has not been proven in multivariate models.
Neither does it convey the established importance of anatomical location and soft tissue extension which drives treatment of CCCS. 17 A new system of risk stratification for CS of bone is needed. 20 It should ideally depict differing risk groups with regard to both local and systemic control in as few categories as possible 21 and must be specifically validated for each subtype. The methodology should be simple to ensure low inter-observer variability, thereby establishing common definitions to allow for the pooling of cases vital to rare disease research.
We have previously shown the prognostic importance of a soft tissue component in CCCS of bone. 14 We, therefore, chose to look at whether the soft tissue components' size by standardized measurement was associated with prognosis.
Our primary aim was to use prognostic analysis from a complete national cohort of CCCS of bone 14 to quantify the influence of anatomical location, presence, and size of a soft tissue component, as well as malignancy grade.
The secondary aim was to propose a risk stratification model for CCCS based on the above and to present results from our cohort by this stratification.

| MATERIALS AND METHOD
The Cancer Registry of Norway (CRN) cohort of CS of bone from 1990 to 2013 has been previously described, and the same definitions were applied to this study 14

(Supporting Information Material).
It is a prospective register, but we have retrospectively quality controlled all data. There were 197 eligible cases of CCCS in the cohort for all anatomical locations excluding head and neck. Radiology was reviewed for 134 cases. Radiology was missing in 63 cases due to the hospitals deleting the images 10 years after diagnosis. Of the 63 cases, 48 had histologically proven soft tissue components.
These 48 were therefore excluded since we had no means of measuring the soft tissue component, while the other 15 cases were intramedullary cases and were included in the study. This resulted in 149 cases eligible for analysis ( Figure 1).
Thirty-eight of the 149 cases were selected for histological review by set criteria to complement missing or unclear data. All 149 cases are histologically confirmed and have complete data sets and documented follow-up. All information in the register for all cases was quality controlled by the main author from the clinical notes.

Borderline malignant lesions have intentionally been excluded
under the selection of the original cohort (11 cases). 14 The size of the soft tissue component was measured using radiology software to the nearest mm, perpendicular to the stipulated outer cortex in the plane that best allowed the greatest measurement by coauthor IT, a senior sarcoma radiologist. Examples are illustrated in Figure 2. Edema was not measured. In the case of circumferential tumor, measurement was made where the largest.
Extremity location was defined by the glenohumeral and hip joints, meaning that the pelvis and scapula are denoted as part of the axial skeletal location group.
Grading has been practiced in accordance with WHO criteria 22 in a four-grade system. CCCS is graded 1 to 3, while dedifferentiated CS as grade 4 was not included in this study.
The term "Atypical Cartilaginous Tumor (ACT)" has not been adopted in Norway during the study period and as such was not introduced into the manuscript.
The Strengthening of Reporting of Observational Studies in Epidemiology (STROBE) checklist was used as far as the methodology allowed. 23

| Clinical setting
Investigation, treatment, and follow-up of chondrosarcoma in Norway during the period was highly centralized. Ninety-one percent were referred to as untouched cases, 3% had a biopsy outside a referral center, and 6% had contaminated surgeries. In the early1990s treatment was driven by grade on open biopsy, but this F I G U R E 1 Flowchart illustrating methodology for inclusion/exclusion of cohort [Color figure can be viewed at wileyonlinelibrary.com] changed in the mid-90s with the acceptance of skeletal location and soft tissue components role in depicting biology together with the awareness of challenges involved in defining accurate histological grade on biopsy. Since then, curettage has been performed on extremity intramedullary lesions with limited signs of radiological aggressiveness, while all others have been resected. Similarly, needle biopsy has primarily been performed to confirm a radiological chondrosarcoma diagnosis before more extensive surgery (ie resection) and increasingly not at all, in the latter part of the study period. Reporting to the CRN was mandatory by law throughout the period.

| Bias
Possible bias has been addressed by the use of multiple sources.
Review of both radiology and pathology where available, or by set criteria, including all cases at risk of wrongful diagnosis or with missing data. The review was partly in a group setting. The cohort is population-based, which should control for random patterns in local or regional referral practices.
There exists a possibility for selection bias since cases with a soft tissue component and missing radiology were excluded (48/197 cases = 25%). Thus, the influence of the soft tissue component on our findings may be underestimated in magnitude in prognostic analysis when compared to intramedullary disease. We have reason to believe that our national cohort is complete. 24 A recent national Dutch cohort 19 report an overall CS incidence nearly twice that of Norway and clearly highest in the world. 25 Their documented incidence increase is driven by ACT's while ours is for grade 2 disease and as such it seems likely that they define ACT at a lower level of aggressiveness in the Netherlands as opposed to Norway. Since ACT's are by definition tumor with limited or no metastatic potential this should not effect our results.

| Statistics
Stata 14 software was used for statistical analysis. Significance was set at P < .05 and was tested using two-tailed tests.
Descriptive statistics are presented as mean values with range for continuous variables and frequencies and relative frequencies for categorical variables. The Kaplan-Meier estimator was used to establish rates of local recurrence (LR), rate of metastasis (Met), and disease-specific survival (DSS) at 2, 5, and 10 years of follow-up. The log-rank test was used to test for differences in survival curves.
Patients were followed from the date of diagnosis to the date of the event of interest, death, or end of clinical follow-up, whichever came first. Death was treated as a censoring mechanism in analyses of LR and Met, but as an event in DSS. We have used the CRN definition

| Ethics
Retrieval and handling of data is in accordance with the Helsinki declaration. The project is based at the CRN and data retrieval is based on the quality control charter of the cancer registry act of 1967/2014. The regional ethics board (REK) has been consulted and accepts this foundation.

| Follow-up
Median follow-up based on observed follow-up from diagnosis to death or end follow-up at 10 years was 8.29 years (range, 0.08-10 years). In all, 61 patients completed 10 years of follow-up while 32 died from any cause. The combined axial location with ≥1 cm sized soft tissue component variable remained strongly statistically significant for rate of metastasis (P < .001) when corrected for age at diagnosis, size of tumor, sex, and histological grade (Table 1B), while histological grade became insignificant.

| Prognostic analysis
The same combined variable remained a significant independent predictor of DSS (P = .03), while histological grade became insignificant at multivariate analysis (Table 1C). In addition, DSS was independently influenced by age (P = .004) and metastasis at diagnosis (P < .001).

| Cohort by risk stratification
The demographic data and tumor characteristics by risk group are presented in Table 2. There was an even distribution by sex with the exception of a higher proportion of female cases in the low-risk type I    and high-risk cohorts (76% and 67%), despite wide differences in the rates of metastasis between these two groups (0% vs 33% at 10 years).

| DISCUSSION
Prognostic analysis concerning the size of the soft tissue component has not been described previously. This is also the first report from a national cohort which attempts to quantify and organize prognostic analysis for CCCS related to the rate of metastasis, which is likely a more direct outcome measure as compared to survival. This results in a simple and novel tool for presenting systemic outcomes for patients with CCCS without the use of histological grade allowing for improved information of patients at the time of diagnosis before surgery and the pooling of results for the research. This cohort provides further evidence towards the safety of clinical CS management not driven by biopsy 28,29 | since the overall results are good when compared to other CCCS cohorts. 6,7 Our most important finding is the organization of the cohort into a low-vs high-risk group. This dichotomous division creates a large low-risk group and a small high-risk group. Interobserver variability has not been studied for our proposed stratification, but is likely less, since the methodology has fewer elements and is less subjective by nature.
If our findings are confirmed, the clinical implications are numerous. First, the CS community will have a common language essential for cooperation between centers. As opposed to osteosarcoma or Ewing patients; CS patients are seldom included in prospective cohorts and numbers reported are often either small or hampered by poor data quality. There is a clear need for multicenter prospective register reporting with quality data at a subtype level, but for this to occur there must first be a common language of accepted and reproducible definitions.
The low/high-risk finding also creates a division which can guide with soft tissue components. Since size previously has been considered a prognostic criteria, larger intramedullary CS was resected without a clear cut-off. This can naturally influence our findings.
We present a methodology based on a single measurement of the soft tissue components size by standardized means, somewhat similar to that for Peripheral CS measurement of a cartilage cap. 31 Our initial investigation sought to look at a ratio between the size of the intramedullary part and soft tissue part of the tumor based on the notion that a small intramedullary tumor with large soft tissue component appears more aggressive than a large intramedullary tumor with small soft tissue component. This was complex to interpret, and as such we chose to look at single centimeter measurements (1 cm, 2 cm, 3 cm) of the soft tissue component by standardized means. In this cohort, 1 cm appeared to be a cut off in predicting risk of metastasis. This has the appeal of simplicity, but needs to be confirmed in other cohorts.
The low-risk II group has increased rates of local recurrence and poorer survival compared to low-risk I and similar rates of overall survival as the high-risk cohort, despite a huge difference in their metastatic rates. The low-risk II group had higher age, larger tumors, and larger soft tissue components than low-risk I. Possibly, low-risk II might be a more morbid subgroup than low-risk I with later presentation, more demanding surgery, and poorer non-oncological survival, but this needs to be investigated further.

| 1121
This finding underlines that a censor for metastatic events is the most important direct measure of systemic biology for CCCS.
Measuring outcomes solely by survival will introduce other confounders and insecurities, particularly in a slow to intermediate growing disease like CS with an overall low rate of metastasis.
The accuracy of a diagnosis of lung metastasis in this setting has not been studied to our knowledge, but there is little reason why there should be a discrepancy between the groups.
Our cohort shows increased number of men in the high-risk subcohort and women in the low-risk subcohort and as such and an increased influence of sex on risk of local recurrence. This may be random in a small cohort and rare disease, but could also represent different tumor biology or medical seeking behavior between the sexes. Our proposed system predicts LR at univariate ( Figure 4A), but not at multivariate level (Table 1A) and as such should be re-examined in larger cohorts. Furthermore, large cohorts have failed to contribute significant new findings in predicting CS biology, 11,12,19 even when reexamined with machine learning, 38 likely due to limited variables and data quality. 39 We believe the quality of our data to be unique both with regard to its completeness, range of variables and quality control of every variable with a complete data set for all included at a subtype level. The CRN has a documented 97% completeness for bone sarcoma before our further quality control. 24 This is a historical prospective analysis of a small national

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are in part available from the corresponding author. Restrictions apply to the availability of these data, which were used under license for the current study from the National Cancer Registry, and so are not publicly available.
Data are however available from the corresponding author upon