Five-year disease-specific survival according to the American Joint Committee on Cancer (AJCC) TNM staging system (A) and a new prognostic staging system (B). The tick marks on the curves indicate deaths secondary to other causes that were censored.
eTable. Backward Variables in Model
Customize your JAMA Network experience by selecting one or more topics from the list below.
Huang C, Chen L, Chang S, Wu H, Ting W, Yang C. Accuracy of a Staging System for Prognosis of 5-Year Survival of Patients With Nasopharyngeal Carcinoma Who Underwent Chemoradiotherapy. JAMA Otolaryngol Head Neck Surg. 2017;143(11):1086–1091. doi:10.1001/jamaoto.2017.1562
Does a new prognostic staging system that combines tumor and clinical characteristics improve the accuracy of prognosis of survival of patients with nasopharyngeal carcinoma?
In this cohort study involving 207 patients with nasopharyngeal carcinoma after concurrent chemoradiotherapy, a new prognostic staging system demonstrates better monotonicity and better discriminatory ability for 5-year disease-specific survival than the routinely used American Joint Committee on Cancer and the International Union Against Cancer TNM staging system.
A new staging system could help to identify high-risk patients with nasopharyngeal carcinoma for more intense treatment and care.
Concurrent chemoradiotherapy delivers a high level of tumor control and survival benefits for patients with nasopharyngeal carcinoma (NPC). However, many uncertainties still exist regarding the outcomes of chemoradiotherapy, making a more precise survival prognostic system necessary.
To introduce a new staging system that combines tumor and clinical characteristics to improve the accuracy of prognosis for patients with NPC.
Design, Setting, and Participants
This cohort study enrolled 207 patients with newly diagnosed NPC who underwent concurrent chemoradiotherapy between January 1, 2007, and December 31, 2014, at Chi-Mei Medical Center in Tainan, Taiwan. Data on these patients were collected from the cancer registry database of the Chi-Mei Medical Center. Patients who had a history of cancer or were unable to complete a full course of radiotherapy were excluded. Follow-up was completed on September 30, 2016, and the data analysis was performed from January 1, 2017, to February 28, 2017.
Main Outcomes and Measures
The risk factors associated with 5-year disease-specific survival were incorporated into the American Joint Committee on Cancer (AJCC) and the International Union Against Cancer TNM staging system to construct a new prognostic staging system. The χ2 test for linear trend, the Akaike information criterion, and the C statistic were used to evaluate the monotonicity and discriminatory ability of the new prognostic staging system and the AJCC TNM staging system.
Of the 207 patients enrolled in the study, 157 (75.8%) were men, and the mean (SD) age was 48 (11) years. Multivariate analysis identified advanced clinical T stage (adjusted hazard ratio [aHR], 3.20; 95% CI, 1.58-6.48), poor performance status (aHR, 2.62; 95% CI, 1.30-5.28), and cumulative cisplatin dose lower than 100 mg/m2 (aHR, 2.28; 95% CI, 1.10-4.74) as independent prognostic factors. The β coefficients from the Cox proportional hazards regression model were used to develop an integer-based, weighted point system; advanced clinical T stage, poor performance, and cumulative cisplatin dose lower than 100 mg/m2 were each assigned a score of 1. The sum of these risk scores was stratified into new stage I (score of 0), new stage II (score of 1), new stage III (score of 2), and new stage IV (score of 3). Compared with the AJCC TNM staging system, the new prognostic staging category had better monotonicity with a higher χ2 value (17.8 vs 25.6) for linear trend, better discriminatory ability with a smaller Akaike information criterion (367 vs 360), and a greater C statistic (0.702 vs 0.740) for 5-year disease-specific survival.
Conclusions and Relevance
The new prognostic staging system has a better accuracy of prognosis of survival than the routinely used AJCC TNM staging system and thus is more useful in identifying high-risk patients for more intense treatment and care.
Nasopharyngeal carcinoma (NPC) is rare in Western countries but is endemic in Southeast Asia and North Africa.1 In Taiwan, the annual incidence of NPC is 6.17 per 100 000 people and increases each year.2 Concurrent chemoradiotherapy (CCRT) is the first line of therapy for locally advanced disease and has demonstrated a high level of tumor control and survival benefits.3 However, the TNM staging system by the American Joint Committee on Cancer (AJCC) and the International Union Against Cancer4 is not precise enough for estimating survival outcomes because it uses only information about the tumor itself. The outcomes of patients with the same stage are still heterogeneous. Therefore, searching for additional risk factors and modifying the present AJCC TNM staging system may help improve its accuracy of prognosis of survival.
Previous studies have reported the many factors, such as age, World Health Organization (WHO) histological type,5 serum lactate dehydrogenase level, tumor volume, and pretreatment Epstein-Barr virus DNA load, associated with local control and survival.6-9 Furthermore, during CCRT, patients may experience toxicities and may not be able to comply with the full course of chemotherapy as prescribed. Some reports have indicated that the cumulative cisplatin dose (CCD) delivered during CCRT has prognostic significance for patients with NPC.10,11 In addition, Guo et al12 have demonstrated that cisplatin at a dose of 100 mg/m2 may be adequate to achieve a survival benefit when combined with intensity-modulated radiation therapy.
Given this background, we aimed to retrospectively analyze the outcomes of patients with NPC who underwent CCRT in the Chi-Mei Medical Center in Tainan, Taiwan, and to evaluate the prognostic significance of CCD and other clinical risk factors. In addition, we sought to revise the routinely used AJCC TNM staging system by constructing a new staging category that incorporates important factors to better estimate the 5-year disease-specific survival (DSS) for patients with NPC.
This study was reviewed and approved by the institutional review board of the Chi-Mei Medical Center. The requirement for patient informed consent was waived by this institutional review board because all identifying information was removed from the data set before analysis.
Patients with newly diagnosed NPC (n = 207) who underwent CCRT between January 1, 2007, and December 31, 2014, were included in this study. Data on these patients were retrieved from the cancer registry database of the Chi-Mei Medical Center. Patients who had a history of cancer or were unable to complete a full course of radiotherapy were excluded. Follow-up was completed on September 30, 2016, and data analysis was performed from January 1, 2017, to February 28, 2017.
Among the data collected were the date of diagnosis; age; sex; clinical TNM stage; chemotherapy regimen, dose, and timing; performance status; WHO histological type; and cause of death. The CCD was calculated as patients received intravenous cisplatin during a 6- to 7-week course of external radiotherapy, and the last radiotherapy fraction ended the treatment. Patients who received cisplatin on a weekly (30-40 mg/m2) or triweekly (100 mg/m2) basis were all included in the analysis. Jagdis et al13 reported that weekly regimens were as effective as the triweekly regimens and that the CCD achieved did not differ significantly between these 2 dose fractionation schemes. All patients were restaged according to the seventh edition of the AJCC staging system.14 The primary end point was the 5-year DSS rate. Deaths from cancer were defined as valid events, and deaths secondary to other causes were censored.
All registered variables were analyzed using a univariate Cox proportional hazards regression model for estimating hazard ratios with 95% CIs. Factors with statistical significance in univariate analysis—advanced clinical T stage, poor performance, and CCD lower than 100 mg/m2—were entered into a multivariate Cox proportional hazards regression model. All significant factors were then used to estimate the coefficients for a new prognostic staging category and to create a scoring system by following the methods in the Framingham study.15 The β coefficients from the Cox proportional hazards regression model were used to develop an integer-based, weighted point system for stratifying the 5-year DSS rate. The referent value for each variable was 0, and the coefficients for the other variables were adjusted proportionally and rounded to the nearest integer. Individual risk scores were calculated by summing their factor scores; advanced clinical T stage, poor performance, and CCD lower than 100 mg/m2 were each assigned a score of 1. The categories and total scores for the new prognostic staging system were as follows: new stage I (score of 0), stage II (score of 1), stage III (score of 2), and stage IV (score of 3).
All statistical analyses and graphics were executed using IBM SPSS Statistics for Windows, version 20.0 (IBM Corp), and a 2-sided P < .05 was determined to be statistically significant. Continuous variables were analyzed with 1-way analysis of variance, and categorical variables were compared by the Pearson χ2 or Fisher exact test. The Kaplan-Meier method was used to analyze the cumulative 5-year DSS rates according to both the AJCC TNM staging system and the new prognostic staging system. The primary event was disease-specific mortality, and survival curves were calculated from the date of diagnosis. The proportional hazard assumption was valid for all variables. After adjusting for tumor and clinical factors, we compared the 5-year DSS rates for these 2 risk stratification models—AJCC TNM staging system and new prognostic staging system—using Cox proportional hazards regression models. The Akaike information criterion (AIC)16 and the C statistic17 were applied to evaluate the relative discriminatory ability of the AJCC TNM staging system and the new prognostic staging system. A χ2 test for linear trend was used to assess monotonicity, with a higher value indicating better monotonic trend.18 In addition, the AIC was used to evaluate the predictive ability of a statistical model that allows the model to be directly compared on the same data set. A lower AIC indicates a more explanatory model because of the lower risk of overfitting. The C statistic served as a measure of discrimination. The accuracy of DSS prognosis was quantified (0.5, equal chance; 0.7-0.8, acceptable; 0.8-0.9, excellent; and 0.9-1.0, outstanding).
The demographic and clinical characteristics of this study are shown in Table 1. A total of 207 patients treated for newly diagnosed NPC with curative CCRT were enrolled. Most of these patients were men (157 [75.8%]), and the mean (SD) age was 48 (11) years. All patients received either intensity-modulated radiation therapy or volumetric modulated arc therapy. Most patients (166 [80.2%]) had WHO category type 2 tumors, and 35 patients (16.9%) received induction chemotherapy and related regimens, including cisplatin and fluorouracil. The number of DSS events in this study population was 37, and the median (interquartile range) follow-up time was 39.9 (1-60) months.
In addition, univariate Cox proportional hazards regression analysis for DSS prognosis showed that advanced clinical T stage (cT3-T4), poor performance status (Eastern Cooperative Oncology Group [ECOG] grades 1-4), WHO tumor category types 1 and 2, and lower CCD (<100 mg/m2) were significant factors with hazard ratios summarized in Table 1. Multivariate analysis for DSS prognosis identified 3 of these factors: poor performance status (adjusted hazard ratio [aHR], 2.62; 95% CI, 1.30-5.28), advanced clinical T stage (aHR, 3.20; 95% CI, 1.58-6.48), and CCD lower than 100 mg/m2 (aHR, 2.28; 95% CI, 1.10-4.74) (Table 1 and the eTable in the Supplement). The new prognostic staging system used an integer-based, weighted point system for stratification (cT3-4: score of 1; ECOG grades 1-4: score of 1; and CCD lower than 100 mg/m2: score of 1). On the basis of the cumulative score, we stratified the risk of the 5-year DSS into 4 levels: new stage I (score of 0), stage II (score of 1), stage III (score of 2), and stage IV (score of 3).
The Figure shows the Kaplan-Meier plots of DSS rates according to the AJCC TNM staging system (Figure, A) and the new prognostic staging system (Figure, B). Under the AJCC TNM staging system, the 5-year DSS rates were 97.7% at stage II, 85.6% at stage III, and 56.0% at stage IV. Under the new prognostic staging system, the 5-year DSS rates were 86.4% at stage II, 72.2% at stage III, and 40.3% at stage IV (Table 2).
A comparison of the 2 staging systems (Table 3) showed that the new prognostic staging system had better monotonicity with a higher χ2 value for linear trend than the AJCC TNM staging system (25.6 vs 17.8), better discriminatory ability for 5-year DSS with a smaller AIC than the AJCC system (360 vs 367), and greater C statistic (0.740 vs 0.702). These values indicated that the new prognostic staging system, which incorporates clinical features such as performance status and CCD into the AJCC TNM staging system, provided better 5-year DSS prognosis accuracy for patients with NPC who underwent CCRT.
In this study, we confirmed that several factors, including poor performance status, advanced clinical T stage, and CCD lower than 100 mg/m2, were independent risk factors associated with poor 5-year DSS among patients with NPC who underwent CCRT. The routinely used AJCC TNM staging system might not be sufficient to estimate survival outcomes for patients with NPC, and integration of CCD and performance status to create another prognostic model significantly improved the accuracy of the 5-year DSS prognosis. This finding might help identify high-risk patients with NPC and further improve their cancer care.
This study had several strengths. First, to our knowledge, no other risk-stratification model or prognostic scoring system incorporates CCD into the AJCC TNM staging system to estimate survival among patients with NPC. Second, all the patients analyzed in this study were treated at a single institution. Thus, the treatment and follow-up were consistent with the treatment guidelines of the hospital. Third, the AJCC TNM staging system for NPC does not consider clinical factors such as age, performance status, or chemotherapy compliance. The new prognostic staging system introduced in this study, compared with the current AJCC TNM staging system, improved the precision of prognosis because it included not only clinical but also treatment information.
During CCRT, the optimal cumulative dose of cisplatin or the frequency of treatment cycles (eg, weekly or triweekly) has not been clearly determined. A few studies have focused on the association of CCD with prognosis and have proposed different CCD cutoff values ranging from 200 to 300 mg/m2.19,20 Peng et al20 reported that a CCD of 240 mg/m2 or higher was associated with significantly improved disease-free survival among patients with NPC receiving cisplatin as the only regimen. Guo et al12 retrospectively analyzed the data from 491 patients with advanced-stage NPC who received definitive CCRT using cisplatin. Their study found that overall survival was lower for patients who received a lower CCD (≤100 mg/m2) than for patients who received a higher CCD, and the rate was similar between patients treated with a medium CCD (101-200 mg/m2) and those treated with a high CCD (>200 mg/m2). As in our study, their multivariate analysis for DSS revealed that a CCD of 100 mg/m2 or lower was significantly associated with poorer outcome (aHR, 2.28; 95% CI, 1.10-4.74). Therefore, the cutoff value of less than 100 mg/m2 was included in our new prognostic staging system.
All patients (whether treated or not treated with induction or adjuvant chemotherapy) were included in the analysis. Induction chemotherapy offers advantages of early eradication of micrometastases, which can improve distant control in patients at high risk; however, the efficacy of induction chemotherapy remains controversial.21,22 The possible reason for this observation may be the lack of a truly effective induction chemotherapy regimen, such as the one we used (ie, cisplatin and fluorouracil). The adjuvant cisplatin and fluorouracil therapy was first tested in the Intergroup 0099 trial.23 Subsequently, the usefulness of concurrent chemotherapy followed by chemotherapy has also been reported and confirmed in a number of series.24,25 Chen et al26 (who conducted a phase 3 trial of the addition of adjuvant cisplatin and ﬂuorouracil chemotherapy to CCRT) reported no significant improvement in the 2-year failure-free survival (hazard ratio, 0.74; 95% CI, 0.49-1.10). Future studies are needed to explore the optimal dose of adjuvant chemotherapy, especially for patients who receive a lower CCD.
The advanced clinical T category (cT3-cT4) was another independent prognostic factor for the 5-year DSS rate in multivariate analysis. However, the clinical N category was not. No significant difference in 5-year DSS was found between patients with cN0 to cN1 tumors and patients with cN2 to cN3 tumors (aHR, 0.88; 95% CI, 0.46-1.67). The possible explanation could be that intensity-modulated radiation therapy or volumetric modulated arc therapy techniques, which provide excellent coverage for the neck lymph nodes, were used in our hospital for patients with head and neck cancer.27 On the other hand, dose coverage around the primary nasopharyngeal tumor was limited by the proximity to surrounding critical organs, especially locally advanced tumors that were large and had invaded close to vital structures. This factor possibly made the clinical T category more important for prognosis than the clinical N category in our analysis.
Several studies have demonstrated that performance status provides prognostic information on overall survival among patients with head and neck cancer.28,29 In our analysis, poor performance status was also an important part of this new prognostic staging system given the significant difference in 5-year DSS between the group of patients with ECOG grade 0 and the group of patients with ECOG grades 1 to 4 (aHR, 2.62; 95% CI, 1.30-5.28). It was intuitively believed that poorer performance status might correlate with insufficient treatment intensity and thus poorer outcome, but the 5-year DSS was still affected after adjusting for other clinical factors such as age, tumor stage, and CCD. This finding means that poor performance status might indicate poor nutrition status and possibly compromise immune function and thereby might affect the outcome of treatment.
Our study had some limitations. First, the data were analyzed retrospectively, and the patient sample was relatively small because they were from only 1 institution. Further studies with a larger sample size are necessary. Second, this new prognostic staging system is applicable to only patients with locally advanced NPC who undergo CCRT. Patients who cannot complete a radiation course or are treated with chemotherapy alone are not suitable. Third, many studies have demonstrated that Epstein-Barr virus DNA load plays an important role in outcome, but it was not included in this study. No DNA load data were available in our cancer registry database, and no established standard assay existed for the Epstein-Barr virus DNA load. Assays performed in different clinical laboratories could yield variability in copy number. Finally, this study did not include patients’ comorbid diseases, which substantially or directly affect treatment compliance and, subsequently, outcomes.30 Our cancer registry database provided no information on comorbidities, and no standard measurement exists for assessing comorbidities in patients with NPC.
This study revealed that a CCD lower than 100 mg/m2, poor performance, and advanced clinical T stage were independent prognostic factors for outcomes of patients with NPC who underwent CCRT. We proposed a new prognostic staging system that combines these clinical and tumor factors. Compared with the current, routinely used AJCC TNM staging system, our new prognostic staging system has better accuracy and is more useful for clinical decision making and care.
Corresponding Author: Ching-Chieh Yang, MD, MS, Department of Radiation Oncology, Chi-Mei Medical Center, B2, No. 901, Zhonghua Road, Yongkang District, Tainan City 710, Taiwan, Republic of China (firstname.lastname@example.org).
Accepted for Publication: July 2, 2017.
Published Online: September 21, 2017. doi:10.1001/jamaoto.2017.1562
Author Contributions: Dr Yang had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Chen, Chang, Ting, Yang.
Acquisition, analysis, or interpretation of data: Huang, Wu, Yang.
Drafting of the manuscript: Huang, Chen, Wu, Ting, Yang.
Critical revision of the manuscript for important intellectual content: Chang, Yang.
Statistical analysis: Huang, Yang.
Administrative, technical, or material support: Chen, Wu, Ting.
Study supervision: Chang, Yang.
Conflict of Interest Disclosures: All authors have completed and submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest and none were reported.
Additional Contributions: The staff of the Cancer Center of the Chi-Mei Medical Center collected the data. They were not compensated for their contribution.