A practical nomogram for predicting cancer-specific survival in patients with clear-cell renal cell carcinoma

Background: It has limitations in predicting patient survival to use of the traditional American Joint Committee on Cancer (AJCC) staging system alone. Objectives: We aimed to establish and evaluate a comprehensive prognostic nomogram and compare its prognostic value with the AJCC staging system in adults diagnosed with ccRCC. Patients and Methods: We used the SEER database to identify 24477 cases of ccRCC between 2010 and 2015. The patients were randomly divided into two groups. In the development cohort, we used multivariate Cox proportional-hazards analyses to select significant variables, and used R software to establish a nomogram for predicting the 3-year and 5-year survival rates of ccRCC patients. In the development and validation cohorts, we compared our survival model with the AJCC prognosis model to evaluate the performance of the nomogram by calculating the concordance index (C-index), area under the receiver operating characteristic curve (AUC), net reclassification improvement (NRI), and integrated discrimination improvement (IDI), and performing calibration plotting and decision curve analyses (DCAs). Results: Eleven identified independent prognostic factors were used to establish the nomogram. Age at diagnosis, being unmarried, higher grades, larger tumor size, higher AJCC stage, lymph node metastases, bone metastases, liver metastases, lung metastases, radiotherapy, and no surgery were risk factors for the survival of ccRCC. The C-index, AUC, NRI, IDI, and calibration plots demonstrated the good performance of the nomogram compared to the AJCC staging system. Moreover, the 3-year and 5-year DCA curves showed that the nomogram yielded net benefits that were greater than the traditional AJCC staging system. Conclusion: This study is the first to indicate that married status is an important prognostic parameter in ccRCC. Our results also demonstrate that the developed nomogram can predict survival more accurately than the AJCC staging system alone. The prognostic factors were easily obtained.


Introduction
Renal carcinoma accounts for around 3% of all adult malignancies 1 , and represents the tenth most common cancer in females and the sixth most common in males 2 . It caused an estimated 175,098 deaths (1.8% of the total cancer deaths) ever year 3 . Most (80-85%) renal carcinomas are renal cell carcinoma (RCC), and they constitute the third most commonly diagnosed urogenital malignancy 4 . Clear-cell renal cell carcinoma (ccRCC) patients constitute 80-90% of all RCC patients 5 . ccRCC is a potentially aggressive neoplasm reported to have an overall 5-year progression-free survival rate of 70% and a cancer-specific mortality rate of 24% 6 . Establishing an effective prediction model can help clinicians to make clinical decisions.
The American Joint Committee on Cancer (AJCC) staging system 7 is a classification system for describing the extent of disease progression in cancer patients. It is based on the TNM stage that is generally believed the most powerful prognostic indicator for RCC, and it remains the most-used tool to classify RCC patients in clinical practice. However, research has shown that multivariate Cox proportional-hazards regression analyses including pathological and multiple clinical covariates were more accurate than the TNM stage in predicting patient survival 8 . Several pathology-based systems for predicting clinical outcomes, including those measuring gene expression, have been established to predict the prognosis of patients with RCC, such as the UISS (University of California Los Angeles integrated staging system), Mayo Clinic SSIGN (stage, size, grade, and necrosis) score, 4 TNM stage, and TCGA (The Cancer Genome Atlas) 8 . However, these prediction models are based on difficult-to-obtain genetic data, have a low prediction accuracy, or lack systematic evaluations of the models on which they are based. Moreover, these are used to predict prognosis of patients with RCC rather than ccRCC.
We therefore aimed to establish a comprehensive prognostic nomogram and assumed it has better performance than the AJCC classification in patients diagnosed with ccRCC.

Patients
Information about all of the included patients was retrieved from the latest version of the Surveillance, Epidemiology, and End Results (SEER) database. This study was approved by the Ethics Committee of the Ninth Hospital of Xi'an. The inclusion criteria were as follows: 1.
Renal carcinoma patients with an ICD-O-3/WHO 2008 histological type code of 8312/3 We collected the following data for each patient: age, race, sex, marital status, insurance recode, tumor grade, tumor size, tumor site, AJCC stage, Mayo Clinic stage, surgery status, radiation status, chemotherapy status, lymph node metastases, bone metastases, brain metastases, liver metastases, lung metastases, and survival time (in months). The SEER cause-specific death classification was the endpoint event. The application of the inclusion and exclusion criteria resulted in the identification of 24477 patients in the SEER database between 2010 and 2015.

Statistical analysis
All variables are presented as median (25th-75th percentile) values because continuous variables such as age and survival time did not conform to a normal distribution. The cox regression model analysis determined the hazard ratios (HRs) and 95% confidence intervals (CIs).
Patients were randomly divided into a validation cohort (30% of patients) and a development cohort (70% of patients). The log-rank test was used to verify any differences between these two cohorts. In the development cohort, significant variables selected by multivariate Cox regression analysis were used as predictors for the nomogram, which was established using R software. Interactions between variables were assessed. The nomogram was internally and externally validated in the development and validation cohorts, respectively.
To compare the discrimination performance of our nomogram with AJCC modeling, we calculated the concordance index (C-index) as described by McKeigue et al. 13 and used the areas under the two receiver operating characteristic (ROC) curves (AUCs) as described by DeLong et al. 14 . We also evaluated the improvement in the predictive discrimination of our nomogram by calculating the relative integrated discrimination 6 improvement (IDI) and the net reclassification improvement (NRI), as described by Pencina et al. 15 . Calibration plots were generated to evaluate the predictive accuracy by comparing the nomogram-predicted and actually observed 3-year and 5-year survival probabilities, as described by Vuk et al. and Cohen et al. 16,17 . We also estimated the clinical usefulness and net benefit of our nomogram using decision curve analysis (DCA), as described by Vickers et al. 18 .
All P values were two-sided, with P≤0.05 considered statistically significant. The data were obtained using SEER* Stat version 8.3.5, and the statistical analyses were performed using SPSS version 21.0 and R software.

Prognostic nomogram for 3-year and 5-year survival probabilities
Age at diagnosis, marital status, tumor grade, tumor size, AJCC stage, surgery status, radiation status, lymph node metastases, bone metastases, liver metastases, and lung metastases were significant predictors for ccRCC in the development cohort (Table 2).
These variables were used to develop the predictive nomogram (Fig. 1).

Validation of the prognostic nomogram
We used the C-index, AUC, NRI, and IDI to assess the discrimination performance of the nomogram. The C-index was higher for the nomogram than for the AJCC staging system

Discussion
In addition to histological grade, the tumor size, Mayo Clinic stage at presentation, vascular invasion, and tumor necrosis are prognostic factors that are routinely utilized to predict the ultimate patient survival 19,20 . We found that age at diagnosis, being unmarried, and metastases in the lymph nodes, bone, liver, and lung were risk factors for survival. In particular, this is the first study to include a married status in a survival prediction model of ccRCC. The risk of death increased with Mayo Clinic stage, AJCC stage, and tumor size.
As is well known, surgery remains the most important and probably the only curative approach in ccRCC 21 . Our study found that surgery can improve the prognosis of ccRCC, whereas radiation therapy is a risk factor for survival. This might be due to radiotherapy long being considered a valueless approach for managing primary disease, and so mainly being prescribed to treat distant metastases, especially brain and painful bone metastases, with a palliative intent 2222 . Moreover, patients with radiotherapy were in a more advanced state or had metastases comparing with patients without radiotherapy.
Therefore, the prognosis of radiotherapy patients was worse than that of patients who had not received radiotherapy.
Nomograms have been used in most cancer types in recent years [24][25][26][27][28][29][30][31] , including for ccRCC [32][33][34] . However, there has been a lack of overall evaluations of the developed nomograms, or the variables used for prediction have not been readily available. The clinical applicability and ease of use are highly attractive features of the comprehensive prognostic nomogram we constructed in this study, and we have compared its prognostic value with that of the AJCC classification. Our nomogram model contains risk factors that are easy to obtain from historical records.
To further determine whether the prognostic model performed better than the traditional AJCC staging system, we evaluated the performance of our survival model using several basic features of model validation: C-index, AUC, NRI, IDI, calibration plots, and DCA. The ROC curve or C statistic is typically used to assess the discrimination performance 15 . The IDI and categorical NRI were also used to assess discrimination in terms of the additional diagnostic value of our model compared to the AJCC model. All of these indicators showed that our model has better discrimination performance than the AJCC staging system. The calibration plots resembled a 45-degree line, indicating that the nomogram predictions were well calibrated (Figure 3). DCA is used to evaluate clinical usefulness, and it shows the minimal net benefit of modified scores that incorporate an index. Some studies have demonstrated the benefits of DCA and recommended its use 35,36 . The present results for the 3-year and 5-year DCA curves showed that our model yielded net benefits that were greater than those for the traditional AJCC staging system in both the development and validation cohorts (Figure 4).
The above-described findings indicate that using our new nomogram can ameliorate the gap that exists relative to predictions based on the AJCC staging system alone. This supports that our nomogram is a useful tool for optimizing treatment in the clinical setting of ccRCC.

Limitations
This study was subject to some limitations. The patients were mainly white, and so the results might not be applicable to other racial groups. Our data set and follow-up data came from the SEER database, which is retrospective and so has inevitable inherent bias.
There was also selection bias in the selection and exclusion of patients, because we only selected the patients with complete information. In addition, many factors were not included, such as the statuses of VEGF, HIF-1α, HIF-2α, p53, and Ki-67 37,38 , which have been shown to influence the prognosis of ccRCC. Another limitation of this study is the relatively small sample, and so more data are needed to provide more accurate performance assessments of the model. Finally, the predicted values calculated from the nomogram are for reference use by clinicians only, and the nomogram should be externally validated in another population in the future.

Conclusions
This study is the first to indicate that married status is an important prognostic parameter in ccRCC. Our results also demonstrate that the developed nomogram can predict survival more accurately than the AJCC staging system alone. The prognostic factors were easily AJCC.