Introduction

Esophageal cancer is a common cause of cancer death worldwide1, it is also the 5th leading cancer in incidence and 4th in mortality in China2. In 2010, there were 287,632 new cases and 208,473 deaths of esophageal cancer in China2. In despite of the development of comprehensive treatment strategies, 5-year overall survival (OS) rate for esophageal cancer is 15ā€“35% and the prognosis remains dismal3,4. Esophageal cancer is divided into two main pathological subtypes: squamous cell carcinoma and adenocarcinoma. The predominant histological type of esophageal cancer in China is esophageal squamous cell carcinoma (ESCC), which accounts for over 90% of cases5. Traditionally, the prognosis of ESCC was performed according to the 6th and 7th edition American Joint Committee on Cancer tumor-node-metastasis (AJCC TNM) staging system6. However, ESCC patients at the same TNM stage and received similar therapy usually had variable outcomes7, suggesting that the current AJCC staging system that only assesses anatomical factors may be inadequate to make a treatment decision and evaluate the prognosis. Therefore, there is an urgent demand for a new tool that can provide reliable prognostic information in individual patient.

The visual format of nomogram is a simple and advanced prediction model that estimates the survival of individual patient by incorporating multiple clinical variables and their interdependent relationships8. The nomogram has been extensively used for many cancers and it has been proposed as an alternative or even as a new standard9,10,11,12,13,14. Recently, several studies reported that nomogram combined with the biomarkers of systemic inflammation response could provide more accurate prediction than conventional staging systems in a variety of tumors15,16,17,18. The systemic inflammation-based prognostic scores, including the Glasgow Prognostic Score (GPS), modified GPS (mGPS), C-reactive protein albumin (CRP/Alb) ratio, neutrophil lymphocyte ratio (NLR), platelet lymphocyte ratio (PLR) and lymphocyte monocyte ratio (LMR) have emerged as prognostic factors in ESCC19,20. Compared with other numerous prognostic factors, the inflammation-based prognostic scores are simple, inexpensive and widely available from preoperative evaluation of blood test. However, there is few study establishing a prognostic nomogram for ESCC based on these biomarkers. This study aimed to establish a prognostic nomogram for resectable thoracic ESCC based on the clinicopathological parameters and the inflammation-based prognostic scores, to determine whether this model provides more accurate prediction of patient survival compared with the 6th and 7th edition of AJCC TNM classifications.

Results

Clinicopathological characteristics of patients

The clinicopathological characteristics of patients in the primary cohort (nā€‰=ā€‰633) and validation cohort (nā€‰=ā€‰283) are listed in Table 1. The ratio of men to women in the primary and validation cohort was 3.26:1 and 2.94:1, respectively. The median age in the primary and validation cohorts was 60 years (range, 37ā€“83 years) and 61 years (range, 38ā€“84 years), respectively. In the primary cohort, the median OS was 40 months (range, 3 to 146.2 months) and the rate of 3- and 5-year OS was 53.1% and 43.2%, respectively. In the validation cohort, the median OS was 44 months (range, 3 to 82 months) and the rate of 3- and 5-year OS was 54.4% and 44.6%, respectively.

Table 1 Clinicopathological characteristics and inflammation-based prognostic scores of patients with esophageal squamous cell carcinoma.
Full size table

Nomogram development and internal validation

In univariate analysis, histological grade, T stage, modified N stage, PLR, NLR, LMR, GPS, mGPS and CRP/Alb were found to be significant prognostic factors, while age, sex, body mass index (BMI), tumor location, tumor length and examined lymph nodes showed no statistical differences (Table 2). We explored the association among these inflammation-based prognostic scores. It was found that the classifications of GPS, mGPS and CRP/Alb were highly correlated. Thus three separate multivariate models (GPS, mGPS and CRP/Alb) were run to avoid problems with the presence of multicollinearity. Multivariate analyses demonstrated that histological grade, T stage, modified N stage, NLR, GPS, mGPS and CRP/Alb were independent risk factors for OS (Table 2). Backward stepwise selection with the Akaike information criterion (AIC) in Cox proportional hazards regression modeling was used to find a best-fit model among these independent risk factors. Finally, the nomogram that integrated five variables: histological grade, T stage, modified N stage, NLR and CRP/Alb was used to predict 3- and 5-year OS in the primary cohort (Fig. 1). The concordance index (C-index) for OS prediction was 0.72 (95% CI, 0.69ā€“0.75). The calibration plot for the probability of survival at 3 or 5 years after surgery showed a good correlation between the prediction by nomogram and actual observation (Fig. 2A,B).

Table 2 Univariate and multivariate cox regression analyses for overall survival in patients with esophageal squamous cell carcinoma.
Full size table
Figure 1
figure 1

Evaluation of nomogram integrated systemic inflammation scores in the patients with esophageal squamous cell cancer after radical esophagectomy.

To use the nomogram, the value attributed to an individual patient is located on each variable axis and a line is drawn upwards to determine the number of points received for each variable value. The sum of these numbers is located on the total points axis and a line is drawn downward to the survival axis to determine the likelihood of 3- or 5-year survival. CRP/Alb, C-reactive protein/albumin; NLR: neutrophil lymphocyte ratio.

Full size image
Figure 2
figure 2

The calibration curve for predicting patient survival at 3-year (A) and 5-year (B) in the primary cohort.

Time-dependent receiver operating characteristic (ROC) curves by nomogram, 6th AJCC-TNM staging system and 7th AJCC-TNM staging system for 3-year (C) and 5-year (D) OS in the primary cohort. Decision curve analyses by nomogram, 6th AJCC-TNM staging system and 7th AJCC-TNM staging system for 3-year (E) and 5-year (F) OS in the primary cohort.

Full size image

Comparison of predictive accuracy for OS between nomogram and conventional staging systems

As shown in Fig. 3, the 6th and 7th AJCC classifications showed good prognostic stratification for most patients. However, the 7th AJCC classifications were unsatisfactory in stratifying patients between stages IIIB and IIIC, while the 6th AJCC classifications were unsatisfactory in stratifying patients between stages IIA and IIB.

Figure 3
figure 3

Kaplan-Meier curve of the primary cohort stratified for 6th AJCC-TNM staging system (A) and 7th AJCC-TNM staging system (B).

Kaplan-Meier curve of the validation cohort stratified for 6th AJCC-TNM staging system (C) and 7th AJCC-TNM staging system (D).

Full size image

Our nomogram displayed better accuracy for predicting the survival in the primary cohort. The C-index of the nomogram was 0.72, which was significantly higher than that of the 7th AJCC staging system (0.68) and the 6th AJCC staging system (0.66) (Pā€‰<ā€‰0.001). The time-dependent receiver operating characteristics (ROC) curve showed higher sensitivity and specificity for predicting OS at 3- and 5-year of follow-up (Fig. 2C,D). In the decision curve analysis, the nomogram demonstrated high potential of clinical application because it ensured better net benefits throughout the entire range of threshold probabilities for survival after 3 or 5 years compared with the TNM staging systems (Fig. 2E,F). These results suggest that our nomogram has better performance for predicting OS than the AJCC TNM classifications.

Validation of predictive accuracy of the nomogram for OS

Calculation of OS was done using the designed nomogram on each patient in the validation cohort. The calibration curves showed good consistency in the probability of 3- and 5-year survival between the actual observation and the nomogram prediction (Fig. 4A,B). The C-index of the nomogram for predicting OS was 0.71 (95% CI, 0.67 to 0.77) in the validation cohort, which was also significantly higher than the C-index (0.68) of the 6th TNM classification (Pā€‰<ā€‰0.001) and the C-index (0.69) of the 7th TNM classification (Pā€‰<ā€‰0.001). The ROC curve also showed the similar results (Fig. 4C,D). These results suggest that the nomogram is a more accurate and useful tool for the prediction of OS in patients with resectable ESCC.

Figure 4
figure 4

The calibration curve for predicting patient survival at 3-year (A) and 5-year (B) in the validation cohort.

Time-dependent receiver operating characteristic (ROC) curves by nomogram, 6th AJCC-TNM staging system and 7th AJCC-TNM staging system for 3-year (C) and 5-year (D) OS in the validation cohort.

Full size image

Discussion

In recent years, nomograms have been constructed in many malignancies and some of these nomograms have been found to be more reliable prediction than the traditional staging system9,10,11,12,13,14. Despite many advantages, there is few study on the prognostic nomogram design for resectable ESCC patients. However, that nomogram did not match well to our patient cohort. In this study, the clinicopathological variables including histological grade, T stage, modified N stage, NLR and CRP/Alb ratio were integrated in a prognostic nomogram. This nomogram predicted OS with an accuracy of C-index 0.72, which showed significantly better prediction of OS than the 6th or 7th TNM staging system in the primary cohort. The ROC curve also showed higher sensitivity and specificity for predicting 3- and 5-year OS compared with the 6th or 7th TNM staging system. These results were subsequently validated by an independent external data set. The calibration plots from both the primary and validation cohorts revealed good correlation between the predicted survival probability and the actual survival rate. The decision curve analysis showed more potential of clinical application of the prediction models compared with TNM staging system. Moreover, the information of CRP/Alb ratio and NLR can be obtained from the peripheral blood tests that were routinely conducted during preoperative examinations. Therefore our nomogram is a reliable tool to predict survival in resectable ESCC and is helpful to make individualized treatment decision.

Compared with Suā€™s nomogram21, our prediction model only included three clinicopathological factors: histological grade, T stage and modified N stage. Other clinicopathological factors such as tumor length and number of examined lymph nodes are not independent risk factors due to low associated hazard ratio (almost close to 1) in our study. It is worth mentioning that modified N stage instead of N stage was integrated in our nomogram. Our study found the count of examined lymph nodes was associated with the survival of the node-negative ESCC patients. Patientsā€™ survival is positively correlated with the increasing number of negative lymph nodes for cancer examination. It is well accepted that small number of resected lymph nodes may miss positive lymph nodes and lead to the incorrect diagnosis22,23, but excessive lymphadenectomy will increase the risk of complications, such as anastomotic leakage, recurrent laryngeal nerve damage and respiratory complications24. In addition, extensive lymphadenectomy would lead to poor immune function and slow the postoperative recovery25. So the lymph node-negative ESCC patients (N0 stage patients) were divided into two groups based on examined 5 lymph nodes in our study.

Our nomogram also included the inflammation-based prognostic scores. Although the inflammation-based prognostic scores are not included in traditional staging systems, their roles in increasing predictive performance have been observed recently. The relationship between inflammation and tumor was first reported in 186326. Over the past decades, accumulating evidence has indicated that inflammation contributes to tumor growth, progression and metastasis27. Recently, the systemic inflammatory response biomarkers such as acute-phase proteins and circulating immune cells have been found to be independent markers of prognosis in a variety of cancers28,29,30,31, including ESCC19,20. Albumin and CRP are accepted markers of acute-phase proteins. The most common prognostic scores based on serum CRP and albumin concentrations are GPS and mGPS. Besides GPS and mGPS, CRP/Alb ratio can be used as an independent prognostic factor in cancer20,32. The GPS, mGPS and CRP/Alb were all independent prognostic biomarkers with high correlation in our study. However, when classified by the mGPS and GPS in our primary cohort, 86.4% and 94.3% of patients were classified in the group of score 0. In the validation cohort, more than 90% patients with a score of 0 were classified by the mGPS and GPS. Therefore, GPS and mGPS apply only to a small group of patients and have little clinical significance. Furthermore, backward stepwise selection chose the CRP/Alb ratio instead of GPS and mGPS to build the best-fit prediction model. So the CRP/Alb is superior to GPS and mGPS in our study. The PLR, NLR and LMR are the common prognostic scores based on circulating immune cells. In our study, the NLR, PLR and LMR were significant prognostic factors in univariate analysis, in while only the NLR was an independent prognostic factor for OS in multivariate analysis. Finally, the CRP/Alb ratio and NLR were included in our nomogram. Recently, Liu et al. built a nomogram based on various inflammatory biomarkers for respectable ESCC33. However, the nomogram of their study contained PLR, LMR and GPS, but not CRP/Alb radio and NLR. We found better prognostic effect of CRP/Alb radio than GPS for respectable ESCC, which was also confirmed by many recent articles20,32,34,35. For this reason, we use CRP/Alb radio instead of GPS. In addition, Liu et al. did not put the NLR included into the nomogram, because the multivariate analysis indicated the NLR not an independent prognostic factor. But significant prognostic effect was observed in the univariate analysis. This may be caused by the deficiency of correlation analysis among NLR, PLR and LMR before multivariate analysis. Moreover, Liu et al. constructed a nomogram without the necessary process of Performance and Application8,36. In this study we conducted such a supplement to make the results more scientific and reliable.

Several potential mechanisms can probably be used to explain the prognostic values of the inflammatory biomarkers in cancer: Firstly, C-reactive protein (CRP) and neutrophile granulocytes were triggered by cancer-related inflammatory factors, such as interleukin-6 (IL-6), tumor necrosis factor (TNF) and myeloid growth factors37,38. These inflammatory mediators facilitate the growth, invasion, metastasis and angiogenesis of tumor, disrupt host immune response and induce the resistance to cytotoxic drugs26,39,40. Secondly, elevated neutrophils can secrete plenty of nitric oxide, arginase and reactive oxygen species (ROS), leading to T cell activation disorders41. Meanwhile, increased circulating neutrophils have been reported to produce vascular endothelial growth factor (VEGF), causing tumor angiogenesis42,43. Thirdly, circulating monocytes determine the number of macrophages in tumor tissue and the density of tumor-associated macrophages has been proven to correlate with angiogenesis, tumor invasion and poor prognosis39,44.

Although our nomogram demonstrated good predictive accuracy for survival, there are still several limitations in this study. First, the nomogram was established based on the data from an individual institution in China. Second, our study was a retrospective study and there may exist selection bias during retrospective data collection. Third, there was heterogeneity in the reported thresholds that were used to define an elevated the inflammation-based prognostic scores in the literature. Therefore, our results need to be further verified in a prospective, large-scale collaborative study.

In conclusion, our proposed nomogram integrated the systemic inflammation scores can accurately predict the prognosis of patients with ESCC after radical esophagectomy. We believe that our nomograms would facilitate making the therapeutic decision and individualized patient counseling.

Materials and Methods

Patients

The study included 916 resectable ESCC patients who underwent radical esophagectomy in the Third Affiliated Hospital of Soochow University (Changzhou, China) from January 2002 to December 2012. Transthoracic esophagectomy with mediastinal and abdominal two-field lymphadenectomies was carried out in the present study. The inclusion criteria are as follows: radical thoracic ESCC, R0 resection, no combined malignancy, no distant metastasis, no preoperative or postoperative radiotherapy and/or chemotherapy. Among all enrolled patients, 283 patients from January 2008 to December 2009 were enrolled in the external validation cohort of this study, while the other patients were included in the primary cohort. The study protocol was performed in accordance with the guidelines outlined in the Declaration of Helsinki and was approved by the Ethics Committee of Third Affiliated Hospital of Soochow University. Written informed consent was obtained from all participants.

According to clinical findings or statistical methods, clinical data were collected and categorized as follows: age (ā‰¤60, >60), sex (male, female), BMI (<18.5, 18.5ā€“24.5, >24.5), tumor location (upper, middle and lower), histological grade (well, moderately and poorly or not differentiated), tumor length (ā‰¤4ā€‰cm, 4ā€“8ā€‰cm, >8ā€‰cm), T stage according to 7th edition of AJCC TNM staging (T1a-lamina propria or muscularis mucosae, T1b-submucosa, T2-superficial and deep muscular layer, T3-adventitia, T4a-pleura, pericardium, diaphragm, or adjacent, T4b-other unresectable adjacent structures), number of examined lymph nodes (ā‰¤5, 6ā€“15, >15). To maximize the performance of the nomogram, we chose modified N stage instead of N stage. The modified N stage was defined as N stage in the 7th AJCC stage system except N0 stage. According to examined lymph nodes, the N0 stage was divided into two categories in the modified N stage. Thus the modified N stage was divided into N0 (examined lymph nodesā€‰>ā€‰5), N0 (examined lymph nodesā€‰ā‰¤ā€‰5), N1 (positive lymph nodes 1-2), N2 (positive lymph nodes 3ā€“6) and N3 (positive lymph nodesā€‰ā‰„ā€‰7).

All peripheral blood was collected and tested for neutrophils, lymphocytes, platelet, monocyte counts, serum C-reactive (CRP) and albumin levels just before operation. The inflammation-based prognostic scores in this study were defined and calculated as follows: (1) GPS, patients with both CRP >10ā€‰mg/L and albumin <35ā€‰g/L were allocated a score of 2; patients with CRP > 10ā€‰mg/L or albumin <35ā€‰g/L were allocated a score of 1; and patients with both CRP <10ā€‰mg/L and albumin>35ā€‰g/L were allocated a score of 0. (2) mGPS, patients with CRPā€‰<ā€‰10ā€‰mg/L were allocated a score of 0; patients with CRPā€‰>ā€‰10ā€‰mg/L or albumin >35 g/L were allocated a score of 1; patients with both CRP > 10ā€‰mg/Land albumin< 35ā€‰g/L were allocated a score of 2. Optimal cutoff values including NLR (NLRā‰¤1.7, NLR>1.7), PLR (PLRā€‰ā‰¤ā€‰120, PLRā€‰>ā€‰120), LMR (LMRā€‰ā‰¤ā€‰3.57, LMRā€‰>ā€‰3.57) and CRP/Alb (CRP/Albā€‰ā‰¤ā€‰0.06, 0.06ā€‰<ā€‰CRP/Albā€‰ā‰¤ā€‰0.12, CRP/Albā€‰>ā€‰0.12) were determined by using X-tile software (http://www.tissuearray.org/rimmlab)45.

Follow-Up

Follow-up was conducted as previously described46. All patients were followed up every three months in the first 2 years, every six months until 5 years and then once annually. All patients underwent clinical, laboratory, imaging and endoscopy examinations for assessing recurrence or metastasis.

The latest follow-up was conducted at the end of December 2014. All patients were followed up by phone calls and regular letters. The observation time in this study was the interval from the date of surgical resection to death or latest follow-up. Survived patients were censored on the day of the last follow-up. OS was determined by the period from the time of surgery to the last follow- up or date of patient death. The median follow-up was 39 months (range, 3 to 146.2 months).

Statistical Analysis

Statistical analysis was carried out using SPSS 17.0 for windows (SPSS, Chicago, IL) and R software version 3.2.0 (http://www.r-project.org/) with Hmisc, rms and survival ROC packages. Survival curves were made using the Kaplan-Meier method and compared using the log-rank test. All variables that achieved significance at Pā€‰<ā€‰0.05 in univariate analyses were enrolled in multivariate Coxā€™s proportional hazards model. The nomogram was formulated based on the results of multivariate analysis. A final model selection was performed using a backward stepdown selection process with the AIC47. To evaluate the nomogram performance, we assessed both the discrimination and calibration of these models. The analysis of time-dependent ROC curve and C-index were used to compare the discrimination power for OS between different models. Confidence intervals (CIs) were obtained by creating 1000 bootstrap samples from the entire dataset and replicating the estimation process. The larger the C-index, the more accurate was the prognostic prediction48. Decision curve analysis was used to evaluate the clinical application of prediction models by quantifying the net benefits49. During the external validation of the nomogram, the total points of each patient in the validation cohort were calculated according to the generated nomogram, then Cox regression in this cohort was performed using the total points as a factor and the C-index and calibration curve were finally derived based on the regression analysis. Nomogram construction and validation were performed with nomogram guide8,36. A P value less than 0.05 was considered to be statistically significant unless otherwise specified.

Additional Information

How to cite this article: Shao, Y. et al. Prognostic nomogram integrated systemic inflammation score for patients with esophageal squamouscell carcinoma undergoing radical esophagectomy. Sci. Rep. 5, 18811; doi: 10.1038/srep18811 (2015).