Development and Validation of Nomograms Predicting Overall and Cancer-specific Survival of Spinal and Pelvic Tumor Patients with Distant Metastasis

Background: Primary spinal bone tumors with distant metastasis are a sign of advanced stage and are usually accompanied by poor prognosis. This study is to identify the risk factors and establish prognostic nomograms to predict 1- and 3-year overall survival (OS) and cancer-specific survival (CSS) rates for spinal and pelvic bone tumor patients with distant metastasis. Patients and methods: Spinal and pelvic bone tumor patients with distant metastasis between 1998 and 2016 were selected for this study from the Surveillance, Epidemiology, and End Results (SEER) database. Nomograms to predict 1- and 3-year OS and CCS rates were constructed based on independent risk factors identified by univariate and multivariate Cox analyses. Concordance indexes (C-indexes), receiver operating characteristic (ROC) curves, calibration plots, and decision curve analysis (DCA) were used to assess the nomograms. Results: All patients (n=343) were randomly divided into a training cohort (n=243) and validation cohort (n=100). No significant differences were found in the demographic data of all patients in the training and validation cohorts. Ultimately, only four independent risk factors (patient age, histology, grade and surgery) were identified as significantly associated with OS and CCS. The C-indices were 0.722 (95% CI, 0.685 to 0.759) and 0.686 (95% CI, 0.61 to 0.760) for the internal validation and external validation of the OS nomogram, respectively. Similarly, the C-indices based on the CCS nomogram were 0.717 (95% CI, 0.678 to 0.757) and 0.695 (95% CI, 0.619 to 0.771) for the internal validation and external validation, respectively. The calibration curves revealed that the predicted survival and actual survival were in concordance. DCA showed the clinical utility and benefits of the nomograms. Conclusion: The nomograms we constructed based on the SEER database can accurately predict individual patient survival.

The purpose of the current study was to establish prognostic nomograms based on the SEER database to predict 1-and 3-year overall survival (OS) and cancer-specific survival (CSS) rates for spinal and pelvic bone tumor patients with distant metastasis.

Patients and data source
Patient data were obtained and filtered using SEER*Stat software (version 8.3.6; NCI, Bethesda, MD, USA) from SEER web site (https://seer.cancer.gov/data/). Ethics approval was not required for the study because the data were publicly available and without identifying information. (10) The inclusion criteria for data screening were as follows: 1.Patients were diagnosed with a histologically confirmed primary osseous tumor (Ewing sarcoma, chordoma, chondrosarcoma, or osteosarcoma) and distant metastasis.
2.The primary site was limited to the osseous spine and pelvis.
The exclusion criteria for data screening were as follows: 1.Unknown survival months after diagnosis and unknown cause of death.
2.Unknown general information (age, race, sex, year of diagnosis and marital status).

Clinicopathological features
The patients were described based on the following clinicopathological features of interest: patient age (<23 years, 23-68 years, or >68 years), tumor size (<9 cm, 9-11 cm, or >11 cm), sex (female or male), race (white, black, or other), year of diagnosis (1998-2007 or 2008-2016), marital status (married or unmarried), grade (grade I/II, grade III/IV, or unknown), lung metastasis (yes or no), histology (Ewing sarcoma, chordoma, chondrosarcoma, or osteosarcoma), surgery (yes or no), chemotherapy (yes or no/unknown), radiotherapy (yes or no/unknown), primary site (vertebral column or pelvic bones, sacrum, coccyx or associated joints), survival time (months), and survival status (alive, cancer-specific death or all-cause death). The optimal cutoff values of age and tumor size were determined by X-tile software (version 3.6.1, Yale University, New Haven, USA) ( Figure 1). (11) The primary endpoints of our study included OS and CSS. OS was defined as the duration from the date of diagnosis to the date of the last follow-up or date of all-cause death. CCS was defined as the duration from the date of diagnosis to the date of the last follow-up or date of cancer-specific death.

Statistical analysis and nomogram construction
All patients were randomly divided into a training cohort and a validation cohort by using R software version 3.6.2 (www.r-project.org). The chi-square test was applied to compare the clinical categorical variables between the cohorts. Univariate Cox proportional hazards regression analysis was used to evaluate prognostic factors related to OS and CCS in the training cohort.
The factors with P values < 0.2 selected in univariable analysis were further incorporated into the multivariable Cox proportional hazards regression analyses.(7) Multivariate Cox proportional hazards regression analysis was performed to determine the independent risk factors (P<0.05) and removal factors (P >0.05) from the associated models. The hazard ratios and corresponding 95% CIs of the variables were also calculated. All statistical analyses were performed using SPSS 22.0 (IBM Corp, Armonk, NY, USA) and R software version 3.6.2 (www.r-project.org). However, based on multivariate analysis, only patient age, grade, histology and surgery were identified as significant independent risk factors associated with both OS and CCS. Therefore, the prognostic nomograms for 1-and 3-year OS and 1-and 3-year CSS were constructed based on these four independent risk factors.

Nomogram validation
The OS and CSS nomograms were validated both internally and externally by the concordance indexes (C-indexes), receiver operating characteristic (ROC) curves, calibration curves, and decision curve analysis (DCA).
The C-index was used to measure the performance and predicted results of the nomograms. The Cindex is also a useful evaluation value similar to calculating the area under the ROC curve (AUC). (12) ROC curves were used to assess the sensitivity and specificity of the nomograms. The calibration curves were used to determine whether the predicted survival and actual survival were in concordance based on an optimal model constituted by a 45-degree line. DCA was used to further investigate the clinical utility and benefits of the nomograms. patients was the vertebral column. Based on the optimal cutoff value determined by X-tile software for age and tumor size, 153 (44.61%) patients were between 23 and 68 years old, and 69 (20.12%) patients had a tumor size between 9 and 11 cm. Ewing sarcoma was the most common histopathological type, and grades I/II and III/IV of tumor differentiation accounted for 11.08% and 34.11% of the cases, respectively. Only 119 (34.69%) patients underwent surgery, and 92 (26.82%) patients had metastasis to the lung. It should be noted that 233 patients died, of which 220 patients died from cancer. The clinicopathological features in the training and validation cohorts are shown in Table 1. No significant differences were found in the demographic data of all patients in the training and validation cohorts.  Table 3). The 1-and 3-year OS and CCS nomograms were constructed based on the four independent risk factors (patient age, histology, grade and surgery) that were identified in the multivariate analysis (Fig. 2).   were all higher than those of each independent risk factor, which demonstrated that the nomograms had better discriminative ability (Fig. 3, Fig. 4).
As shown in the figure, the calibration plots of the OS and CCS nomograms showed excellent agreement between the actual survival and nomogram prediction for 1-and 3-year survival (Fig. 3,   Fig. 4). Moreover, the DCA curves showed ideal net clinical benefits of the 1-and 3-year OS and CCS nomograms in the training and validation cohorts (Fig. 5). Most studies have demonstrated that tumor size is also a risk factor for the overall and cancer-specific survival prognosis of patients with spinal tumors. (4,5,15,16)       year cancer-specific survival. Notes: The horizontal axis is the threshold value and the vertical axis is the net benefit rate. The larger net benefit implies more benefit of the nomogram.