A nomogram for predicting cancer-specific survival in patients with osteosarcoma as secondary malignancy

The prognostic factors for survival among patients with secondary osteosarcoma remain unclear. The aim of this study was to develop a practical nomogram for predicting cancer-specific survival (CSS) in patients with osteosarcoma as a secondary malignancy. The surveillance, epidemiology, and end results database was used for the identification of osteosarcoma cases. The total sample comprised 5860 cases of primary osteosarcoma and 268 cases of secondary osteosarcoma during the period from 1973 to 2015. The CSS and overall survival (OS) of primary and secondary osteosarcomas were analyzed. The predictors of CSS for secondary osteosarcoma were identified and integrated to build a nomogram. Validation of the nomogram was performed using concordance index (C-index) and calibration plots. The results indicated that patients with secondary osteosarcoma had poorer CSS and OS than patients with primary osteosarcoma. The nomogram model exhibited high discriminative accuracy in the training cohort (C-index = 0.826), which was confirmed in the internal validation cohort (C-index = 0.791). In addition, the calibration plots confirmed good concordance for prediction of CSS at 3, 5, and 10 years. In conclusion, we developed a practical nomogram that provided individual predictions of CSS for patients with secondary osteosarcoma. This nomogram may help clinicians with prognostic evaluations and with the development of individualized therapies for this aggressive disease.

Nomograms have been successfully used as prognostic tools for predicting the probability of disease outcomes with a simple visualization figure that integrates the relevant variables in complex mathematical models 18,19 . Nomograms can improve the discriminatory accuracy of outcome predictions; these have therefore been widely used to quantify the risk of various malignancies 20,21 . However, no nomogram has been developed for patients with secondary osteosarcoma to date. The present study developed an elaborate nomogram for assessing individualized prognoses for secondary osteosarcoma in terms of 3-year, 5-year, and 10-year cancer-specific survival (CSS) using data from the Surveillance, Epidemiology, and End Results (SEER) database 22 .

Methods
Patients and selection criteria. We queried nine population-based cancer registries in the SEER program to obtain records for patients seen during the period from 1973 to 2015 (November 2017 submission) using SEER*Stat software (version 8.3.5) 23 . The SEER program is a population-based cancer registry system with data collected from 18 registries in 14 states across the U.S., representing nearly 30% of the U.S. population. The selection of osteosarcoma cases was done using the Histologic International Classification of Diseases (ICD)-O-3 codes 9180/3-9186/3, 9192/3-9194/3, and 9120/3. No written informed consent was obtained for this study because the data were de-identified and publicly available.
Patients were divided and classified by their sequence numbers: patients with primary osteosarcoma without any prior malignancy were assigned sequence number = 1, and patients with subsequent osteosarcoma following prior malignancies were assigned sequence numbers ≥ 2. Osteosarcomas that occurred following the primary malignancy were considered as "secondary osteosarcoma" in our study.
The exclusion criteria were missing or incomplete data including survival status and time, age, sex, race, and prior malignancies and diagnosis of osteosarcoma at the time of autopsy or on the death certificate. The demographic and clinico-pathological data of all eligible cases were collected and analyzed. Endpoint definition. Cancer-specific death was taken as the primary endpoint of the study. The cause of death was defined as death from osteosarcoma, according to the SEER database. The primary endpoint in this study was defined as the interval between the diagnosis of osteosarcoma and the occurrence of cancer-specific death. The secondary endpoint was overall survival (OS), which was defined as the interval between the diagnosis of osteosarcoma and death from any cause or last follow-up.
Statistical analyses. Statistical analyses were performed using R software version 3.5.1 (R Foundation for Statistical Computing, Vienna, Austria) and SPSS version 20.0 (IBM Corporation, Armonk, NY). The t-test was used to examine differences between mean values. The χ 2 or Fisher's exact test was used to compare proportions. Survival was assessed using the Kaplan-Meier method and compared using the log-rank test. Cox proportional hazard regression analyses were performed to identify independent prognostic factors of survival (univariate and multivariate). Significant variables (P < 0.1) in univariate analyses were included in multivariate regression analyses. Variables that were significant in multivariable analyses were incorporated to formulate the nomogram.
Adequate discrimination and calibration were performed to test and validate the prognostic accuracy of the nomogram model 24 . Discrimination was quantified using Harrell's concordance index (C-index), in which an absolute value close to 1 indicates that a nomogram model has strong predictive ability. The nomogram was further subjected to bootstrapping validation (1000 bootstrap replicates) to calculate the relatively corrected C-index. Calibration plots were developed to evaluate predictive accuracy and, further, to assess the concordance between predicted and observed ongoing survival probabilities. A two-sided P < 0.05 was taken to indicate statistical significance.

Results
Study cohorts. The total sample was comprised of 6128 patients, out of which 5860 patients were diagnosed with primary osteosarcoma (sequence number = 1) and 268 patients were diagnosed with secondary osteosarcoma (sequence number ≥ 2). Osteosarcoma was a second malignancy in 231 cases, third malignancy in 34 cases, fourth malignancy in 2 cases, and sixth malignancy in 1 case. Comparisons of baseline demographic and clinicopathological characteristics between patients with primary and secondary osteosarcomas are presented in Table 1. Patients with secondary osteosarcoma were older than those with primary osteosarcoma at the time of diagnosis (55.1 vs. 29.8 years, respectively; P < 0.001), 194 (72.4%) secondary osteosarcoma patients were older than 40 years at diagnosis. The ratio of females to males was higher in the secondary osteosarcoma group than in the primary osteosarcoma group (53.4% vs. 44.8%, respectively; P = 0.007). Furthermore, the primary site was less likely to be an extremity in cases of secondary osteosarcoma, the pelvis was the most commonly affected site (77 out of 268, 28.7%). Non-pagetic osteosarcoma was more common in patients with primary osteosarcoma, while pagetic osteosarcoma was more common in patients with secondary osteosarcoma, which demonstrates the significant differences in histological subtype between groups. In cases of secondary osteosarcoma, the first primary malignancies included 157 carcinomas (58.6%), 42 sarcomas (15.7%), 41 lymphomas/leukemias (15.3%), 14 retinoblastomas (5.2%), and 14 other cancers (5.2%). Among these 268 patients, 54.9% (147 cases) had received radiotherapy for prior malignancies; secondary osteosarcomas occurred within the prior radiation field in 104 patients (38.8%) and outside the radiation field in 43 patients. The median latency interval between the first primary malignancy and the diagnosis of secondary osteosarcoma was 98.5 months (2-501 months). www.nature.com/scientificreports/ The mean follow-up times were 90.1 and 39.3 months in the primary and secondary osteosarcoma cohorts, respectively. As primary osteosarcoma mostly occurred in children and adolescents, the prognosis of this was good. This could be the main reason for substantial variation of the follow-up period.
Survival in primary and secondary osteosarcoma. Median  Prognostic factors associated with CSS in patients with secondary osteosarcoma. The prognostic factors for CSS for secondary osteosarcoma are shown in Table 2. In univariable analyses, younger age at diagnosis, Caucasian ethnicity, unmarried marital status, chemotherapy for prior malignancies, later year of diagnosis, first primary malignancy other than carcinoma, extraskeletal tumor location, an extremity primary site, non-pagetic osteosarcoma histology, localized disease at presentation, surgical resection, chemotherapy and no radiation therapy for osteosarcoma were significantly associated with improved CSS. These 13 factors were submitted to multivariable analysis. The results showed that age, race, year of diagnosis, a skeletal/extraskeletal tumor location, stage and surgical resection retained significance in the multivariate analysis. www.nature.com/scientificreports/ independent prognostic factors associated with oS in patients with secondary osteosarcoma. Univariable analysis suggested that younger age at diagnosis, unmarried marital status, later year of diagnosis, first primary malignancies other than carcinomas, chemotherapy for prior malignancies, osteosarcomas occurring outside the prior radiation field, non-Pagetic osteosarcoma histology, localized disease at presentation, surgical resection, chemotherapy and no radiation therapy for secondary osteosarcoma were favorable predictors of OS. Similar to CSS, age, year of diagnosis, stage, and surgical resection for osteosarcoma were independent prognostic factors associated with OS in multivariable analyses, but with the addition of osteosarcoma occurring within/outside the prior radiation field. To be noted, surgical resection was an independent favorable factor for both CSS and OS in the present cohort (Fig. 1). Unlike CSS, however, race and skeletal/extraskeletal tumor location did not have any bearing on OS among patients with secondary osteosarcoma (Table 2).  www.nature.com/scientificreports/ location, stage, and surgical resection. The nomogram showed that the largest contributions to prognosis were the location (skeletal or extraskeletal tumor) and age at diagnosis, followed by stage and year of diagnosis. Each variable was assigned a score according to the demographic and clinical features of individual patient (Table 3). By adding up these scores according to a patient's condition, the total score was computed by summing the individual scores. Then, the total score was located on the total point line, and a straight line could be drawn to estimate the patient's probability of 3-year, 5-year, and 10-year CSS from the nomogram (Fig. 2). The C-index for the CSS prediction nomogram was 0.826 (95% CI: 0.787-0.865) for the training cohort and was confirmed to be 0.791 through bootstrapping validation, which suggested that the model had good discriminative ability. The calibration plots for CSS probability at 3-year, 5-year, and 10-year showed that the concordance between predicted and observed survival was optimal (Fig. 3).

Discussion
The incidence of primary osteosarcoma has always been considered higher in males than in females 25 , while the current study revealed that in cases of secondary osteosarcoma, the majority of the patients were females. The proportions of patients with different races were consistent for primary and secondary osteosarcomas. This study showed that amongst patients with secondary osteosarcomas, 72.4% were older than 40 years at the time of diagnosis, the result was similar to previous reports 16,26 . Primary osteosarcoma mostly occurred in the long bones of the extremities near the metaphyseal growth plates 25 . However, we observed that secondary osteosarcomas were more likely to be located at non-extremity sites. In the current study, the authors found that the most common primary malignancies were carcinomas, followed by sarcomas and lymphomas/leukemias. These results were inconsistent with previous studies 27,28 . Distant metastases were present at diagnosis in 21.6% of secondary osteosarcoma patients. Radiation is a well-documented etiological factor of osteosarcoma, with the median interval between radiation and the occurrence of osteosarcoma reported to be 12-16 years 25 . This study observed a shorter post-radiation latency because the median latency interval between the diagnosis of first primary malignancies and osteosarcoma was 98.5 months, as the exact date of prior radiotherapy was not available. In this cohort, the pelvis was the most commonly affected site, 38.8% of secondary osteosarcomas occurred within the prior radiation field. www.nature.com/scientificreports/ Most previous SEER studies on osteosarcoma either treated primary and secondary osteosarcoma together or were limited to osteosarcoma of specific histological subtypes [29][30][31][32] . There was only one study focusing on secondary osteosarcoma from SEER data, published nearly 20 years ago, which included only 133 patients and indicated that secondary osteosarcoma had poorer OS than primary osteosarcoma. However, that study did not evaluate CSS nor analyze the impact of any treatment on survival 14 . A later study reported that radiationinduced secondary osteosarcoma proved to have similar outcomes to primary osteosarcoma 33 . However, a recent study suggested that the prognosis of secondary osteosarcoma may be more favorable than that of primary osteosarcoma 34 . The survival and prognostic factors of secondary osteosarcoma remain unclear. So, identifying accurate prognostic factors has clinical importance for guiding personalized cancer therapy. The present study provides detailed survival data, and it could be the largest cohort study on secondary osteosarcoma reported to date. Furthermore, an optimal graphical validated nomogram was developed for predicting CSS. The nomogram model exhibited high discriminative accuracy in the training cohort (C-index = 0.826), which was further confirmed in the internal validation cohort (C-index = 0.791). This study suggests the excellent performance of this nomogram for estimating the prognosis of secondary osteosarcoma, as the calibration plots confirmed good concordance for the prediction of CSS at 3-, 5-, and 10-years. To the best of our knowledge, this is the first prognostic nomogram developed for secondary osteosarcoma.
This study revealed that patients with secondary osteosarcoma had poorer CSS and OS than patients with primary osteosarcoma. We identified age, race, year of diagnosis, skeletal/extraskeletal tumor location, stage, and surgical resection as independent factors for CSS. For OS, the independent prognostic factors included age, year of diagnosis, stage, surgical resection, and osteosarcoma occurring within/outside the prior radiation field. Notably, patients with secondary osteosarcomas occurring within the irradiated field had inferior OS compared  www.nature.com/scientificreports/ to patients with secondary osteosarcomas occurring outside the irradiated field. CSS was similar between groups. This suggests that the difference in OS was caused by factors other than secondary osteosarcoma itself. Previous studies have reported that surgical resection significantly improves disease-free survival and OS in patients with secondary osteosarcoma 28,33 . We also observed significant differences in CSS and OS between secondary osteosarcoma cases with or without surgical resection, which demonstrated that surgical resection was an independent factor significantly improving CSS and OS in the present cohort. For osteosarcomas occurring within the prior radiation field, one important issue that must be addressed is that radiation therapy can prolong postoperative complications because the condition of the operative field is entirely altered after radiotherapy 33 . For these patients, surgical options should be prudently adopted 35 . In the present study, data on the postoperative complications were not available due to the limitations of the SEER database; however, the favorable CSS and OS findings strongly justify the surgical resection of secondary osteosarcoma.
Intensive chemotherapy has considerably improved the prognosis of patients with primary osteosarcoma 9 . For secondary osteosarcoma, Shaheen et al. reported that patients treated aggressively with a combination of chemotherapy and surgical resection had better outcomes than patients treated with surgical resection alone 33 . The present study did not demonstrate significant benefits of chemotherapy on CSS or OS in multivariable analyses. However, the heterogeneous regimens and intensity of chemotherapy over more than 40 years may have limited the statistical power of this study. Prior myelosuppressive chemotherapy and/or radiotherapy may limit the tolerance of patients with secondary osteosarcoma who undergo subsequent intensive chemotherapy, and we strongly recommend the prophylactic use of myeloid growth factors after chemotherapy 36 .
This study had several limitations. First, due to the retrospective study design, selection bias was unavoidable. Second, the SEER dataset lacks data on doses of radiotherapy or chemotherapy regimens, and we were therefore unable to evaluate the impacts of these factors on the development and survival of secondary osteosarcoma. Third, due to the rarity of this disease, we were not able to validate the constructed nomogram using other cohorts.

Conclusion
We developed a practical nomogram that provided individual predictions of CSS for patients with secondary osteosarcoma using five clinicopathological factors and one treatment-related factor. Bootstrapping validation of the model confirmed its good performance. This nomogram may help clinicians with prognostic evaluations and with the development of individualized therapy for this aggressive disease. Future prospective studies are required to further determine the impacts of different treatment modalities on the survival of patients with secondary osteosarcoma.

Data availability
Data for this manuscript are available after formal request to the corresponding authors. www.nature.com/scientificreports/ Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creat iveco mmons .org/licen ses/by/4.0/.