Effect of Incorporation of Pretreatment Serum Carcinoembryonic Antigen Levels Into AJCC Staging for Colon Cancer on 5-Year Survival
2Department of Internal Medicine, University of Buffalo, Buffalo, New York
3Department of Biostatistics, Roswell Park Cancer Institute, Buffalo, New York
4Department of Surgery, Baptist Memorial Health Care, Memphis, Tennessee
Objectives To evaluate whether long-term overall and cancer-specific survival is affected by inclusion of C stage into the standard AJCC TNM staging and to study the implications on survival estimates.
Design, Setting, and Participants We performed a retrospective study of all patients diagnosed as having histologically proven colonic adenocarcinoma from January 1, 2004, through December 31, 2005, from the National Cancer Institute’s Surveillance, Epidemiology, and End Results database. We stratified each AJCC stage as C0 (normal) or C1 (elevated) based on the pretreatment serum CEA level. Median follow-up was 71 months.
Main Outcome and Measures Five-year estimates of overall and disease-specific survival and hazard ratios (HRs) for estimates of risk of overall and disease-specific mortality.
Results A total of 16 619 patients were evaluated, and of these, 8878 patients had C0 disease and 7741 had C1 disease. C1 stage was independently associated with a 51% and 59% increased risk of overall (HR, 1.51; 95% CI, 1.44-1.59; P < .001) and disease-specific mortality (HR, 1.59; 95% CI, 1.49-1.69; P < . 001) at a median follow-up of 71 months. Analysis of survival of the AJCC stages subdivided as C0 or C1 revealed a significant worse prognosis of C1 AJCC stages compared with the respective C0 AJCC stages. The magnitude of change in survival was large enough to cause clustering of survival estimates of C1 vs C0 cancers across various AJCC stages. Analysis of patients with stage I, II, and III cancer revealed that node-negative C1 disease was associated with prognosis similar or worse than node-positive C0 disease.
Conclusions and Relevance Inclusion of C stage into the AJCC TNM staging of colon cancer revealed significant differences dependent on C stage in terms of 5-year survival. C-stage inclusion resulted in substantial change in survival estimates, with C1 status portending a prognosis to certain stages similar to or worse than higher AJCC TNM stages with C0 status. We recommend routine pretreatment CEA testing as standard of care in colon cancer and use of C stage for multimodality treatment planning and risk stratification in prospective studies and randomized clinical trials.
Colon cancer is the fourth most common cancer and the second most common cause of cancer-related mortality in the United States.1 Adequate staging allows for prognostication and formulation of the plan of treatment. The TNM system developed by the American Joint Committee on Cancer (AJCC)2 is the most widely adopted method of staging of all cancers, including colon cancer.3 On the basis of the local extent of the primary tumor (T stage), the extent of regional lymph node involvement (N stage), and the presence or absence of distant metastases (M stage), colon cancer had been broadly classified into 4 overall stages: stage I (primary tumor localized within the subserosa and no regional lymph node involvement), stage II (primary tumor extending beyond the subserosa and without regional node involvement), stage III (any involvement of regional lymph nodes without distant metastases), and stage IV (distant metastases). On the basis of survival data obtained from large population-based studies, the AJCC periodically updates the arrangement of these TNM combinations into the overall stages to refine the prognostic utility of the staging system.4
Although the TNM system of classification describes the anatomical extent of malignant neoplasms and projects a stage-derived survival estimate, it likely oversimplifies assessment of the biological potential of the tumor and the overall risk of recurrence and death.5,6 The AJCC has identified the potential limitation of pure anatomical staging in colon cancer.5,7 Nonanatomical prognostic factors related to biological aggressiveness have been increasingly adopted to complement and enhance the value of the AJCC TNM staging of certain cancers, such as histologic grade in sarcoma, mitotic count in melanoma, age in thyroid cancer, and prostate-specific antigen and Gleason score in prostate cancer.2
Using evidence from several studies8– 15 that found pretreatment serum carcinoembryonic antigen (CEA) to be a stage-independent risk factor for poor prognosis in colorectal cancer, the Colorectal Working Group of the AJCC, as early as 2000, proposed the inclusion of serum CEA level (C stage) to complement and modify anatomical TNM staging of colon cancer.7 The group considered adding a C stage classified into substages, namely, not assessable (Cx), not elevated (ie, serum CEA level <5 ng/mL [to convert to micrograms per liter, multiply by 1]) (C0), or elevated (serum level CEA >5 ng/mL) (C1). Overall stages would be expressed as stages I through IV Cx, C0, and C1. Although it has been widely recognized that elevated pretreatment CEA level portends poor prognosis, this proposal has not been adopted into clinical practice or into AJCC staging, primarily because of the lack of validation using long-term outcomes data and the lack of information on the magnitude of effect of C-stage inclusion on the prognosis of each overall AJCC stage (ie, lack of actuarial survival estimates).
To address the above limitations, we used the National Cancer Institute’s Surveillance, Epidemiology, and End Results (SEER) database to analyze the survival outcomes after inclusion of pretreatment CEA into the AJCC staging as recommended by the Colorectal Working Group. We also aimed to study the association of C stage and N stage with prognosis to assess the magnitude of the effect of an elevated serum CEA level on prognosis compared with regional lymph node involvement.
Data Collection and Patient Selection
The SEER database includes approximately 26% of the US population.16 Given that these data are deidentified and ethics approval is waived, approval was not sought through our institution and informed consent was not obtained from the study participants. Using SEER*Stat, an online access program provided by the SEER program, we extracted data from January 1, 2004, through December 31, 2005, on all patients diagnosed as having microscopically confirmed colon adenocarcinoma with available pretreatment serum CEA level information. These analyses of the SEER data are only possible recently because SEER did not routinely collect CEA level information for cases diagnosed before 2004.
We included data pertaining to the age at diagnosis; sex; race/ethnicity; marital status at diagnosis; location of the colonic primary tumor (ascending colon, hepatic flexure, transverse colon, splenic flexure, descending colon, sigmoid colon, or overlapping site); the International Classification of Diseases for Oncology, Third Edition, histology code; histologic grade; stage; and use of surgical resection. The SEER database coded data on the pretreatment serum CEA level as “test not done,” “positive/elevated,” “negative/normal; within normal limits,” “borderline; undetermined whether positive or negative,” “ordered, but results not in chart,” or “unknown or no information.” A single cutoff value for determining an abnormal elevation of the serum CEA level may be impractical and inappropriate in a large population-based database because of variation in the absolute serum CEA levels with patient-related factors (such as heavy smoking and liver dysfunction) and assay techniques used by the testing laboratory. The SEER program therefore used the physician’s interpretation of CEA level information in the medical record when available or the reference values provided by the laboratory reporting the result. We assigned stage C0 to patients with a serum CEA level coded as “negative/normal; within normal limits” and C1 to those with a serum CEA level coded as “positive/elevated.” We used the following exclusion criteria: nonadenocarcinomatous histologic findings, lifetime occurrence of another primary malignant neoplasm and patients recorded to have borderline CEA levels, or CEA levels not recorded or not completed. Survival estimates were based on data from the SEER database. Our final cohort consisted of 16 619 unique records, which were then used for statistical analyses. Using the codes CS (collaborative stage) extension and regional nodes positive (in year 2004 or later), detailed T-stage and N-stage information was extracted, and all cancers were staged according to the AJCC Cancer Staging Handbook.2 Further details about the data reporting of follow-up and linkage can be obtained by accessing the SEER Data Management System User Manual (seer.cancer.gov).
Patient characteristics were reported as means (SDs) for continuous variables and as relative frequencies for categorical variables. The survival outcomes, overall survival (OS), and disease-specific survival (DSS) (ie, cancer), were analyzed using the overall AJCC stage and AJCC and C-stage combinations using standard Kaplan-Meier methods. Estimates of 5-year survival rates were obtained for each stage group, with the corresponding 95% CIs. The log-rank test was used to compare survival among the stage groups in a pairwise fashion. The Tukey honestly significant difference was used to adjust for multiple comparisons. Cox regression models were used to obtain hazard ratios (HRs) for comparing risks of overall and disease-specific mortality among the stage groups. In one multivariate regression analysis, we used a novel covariate (NC), which is a combination of each tumor’s N and C stage to study the interaction of these 2 covariates on prognosis. All analyses were conducted using SAS statistical software, version 9.3 (SAS Institute Inc), at P < .05.
Comparison of Patients With C0 and C1 Disease
Of the entire study cohort (16 619 records), 8878 patients had C0 disease and 7741 patients had C1 disease. Comparison of patient and tumor characteristics between patients with C0 and C1 disease is given in Table 1. Patients with C1 disease more commonly presented at later stages compared with patients with C0 disease. There were 24.0% of C0 patients vs 41.5% of C1 patients presenting with stage IV disease.
C Stage as a Prognostic Factor in Colon Cancer
Multivariate analyses were used to identify factors independently associated with overall and disease-specific mortality at a median follow-up of 71 months (Table 2). C1 stage was a stage-independent poor prognostic factor associated with an increase in overall mortality (HR, 1.51; 95% CI, 1.44-1.59; P < .001) and disease-specific mortality (HR, 1.59; 95% CI, 1.49-1.69; P < .001).
We analyzed the implications of the inclusion of C stage into the AJCC staging by calculating the 5-year OS and DSS outcomes for various AJCC and C-stage combinations, such as stages I C0, I C1, IIA C0, IIA C1, and so on (Table 3). We calculated the hazards for overall and disease-specific mortality separately, along with the respective 95% CIs, by using stage I C0 as the reference value to allow comparison of the outcomes among the various stages. We made 2 major observations. First, within each AJCC stage, patients with C1 disease had a worse prognosis compared with patients with C0 disease. For example, prognosis of patients with stage I C1 disease was worse than patients with stage I C0 disease and so on. Second, the magnitude of the difference in survival between patients with C0 and C1 disease within each stage was large enough that the prognosis of patients with C1 disease of lower AJCC stages was similar to or worse than that of patients with C0 disease of higher AJCC stages. This finding can be gleaned by observing the overlap of the 95% CIs. For example, overall mortality of patients with stage IIA C0 disease (HR, 1.41; 95% CI, 1.25-1.59) is lower than that of patients with stage I C1 disease (HR, 2.01; 95% CI, 1.68-2.40), and overall mortality of patients with stage IIIA C0 disease (HR, 0.89; 95% CI, 0.68-1.15) is lower than that of patients with stage IA C1 or IIA C1 disease (HR, 2.27; 95% CI, 2.01-2.57). A similar pattern can be observed in DSS outcomes as well. This phenomenon is better illustrated in the Kaplan-Meier curves (Figure), where the curves of patients with C0 disease of certain AJCC stages appear to be grouped or clustered, whereas the curves of patients with C1 disease of the respective AJCC stages appear to be clustered separately.
Patients with stage I C0, IIA C0, and IIIA C0 disease had similar prognosis and appear grouped or clustered together, whereas patients with stage I C1, IIA C1, and IIIA C0 disease had a similar phenomenon. Overall survival of patients with stage IIIA C0 disease is better than for the patients with stage I C1 disease (P < .001) and stage IIA C1 disease (P < .001). B, Patients with stage IIB C0, IIC C0, and IIIB C0 disease had similar prognosis and appear grouped or clustered together, whereas patients with stage IIB C1, IIC C1, and IIIB C1 disease had a similar phenomenon. Overall survival of patients with stage IIIB C0 disease is better than for patients with stage IIB C1 disease (P < .001), whereas it is similar for patients with stage IIC C0 disease (P = .16). Refer to Methods section for C stage definition.
Interaction of C Stage and N Stage With Colon Cancer Prognosis
Next, we analyzed the association of various combinations of N and C stages with prognosis among all patients with nondistant metastatic (stages I, II, and III) colon cancer to understand the interaction of nodal status and CEA status (Table 4). We calculated the hazards for overall and disease-specific mortality separately, along with the respective 95% CIs, for each N-stage and C-stage combination by using the N0 and C0 combination as the reference value to allow comparison of the outcomes among various combinations. We again observed that patients with C1 disease of each nodal stage had a worse prognosis than patients with C0 disease of the respective nodal stage, and the prognosis of patients with C1 disease with lower nodal stages was worse or similar to that of patients with C0 disease of higher nodal stages.
This study analyzed the long-term prognosis of patients with histologically proven colonic adenocarcinoma after the incorporation of information of their pretreatment serum CEA level into the TNM staging system. The SEER database study was limited to those whose conditions were diagnosed after 2003 because SEER did not collect CEA level information until 2004. With the most recent update of the SEER database (2010 submission), long-term follow-up data were available for analysis for patients whose conditions were diagnosed in 2004 and 2005, thus making it possible to conduct such a large-scale study for long-term survival outcomes.
A previous study17 evaluated the effect of C stage on anatomical TNM staging for colon cancer, including its effect on prognosis in terms of 2-year overall survival rate and the hazards for overall mortality, using the SEER database. This study revealed that (1) C1 stage was a stage-independent poor prognostic factor associated with a 60% increased risk of overall mortality; (2) prognosis of C1 cancers of lower AJCC stages were similar or worse than the prognosis of C0 cancers of corresponding higher AJCC stages (ie, inclusion of C stage into the TNM staging resulted in a magnitude of effect large enough to cause stage migration); and (3) node-negative C1 cancers were associated with a poorer prognosis than C0 cancers with limited nodal burden (N1 and N2a). Nevertheless, this study was limited because only short-term outcomes were available for analysis (median follow-up of 27 months) and prognosis was evaluated only in terms of hazards of overall mortality and overall survival, without assessing disease-specific (ie, cancer-specific) outcomes.
A prerequisite to proceed with the inclusion of C stage into the TNM system is confirmation of C stage as a prognostic factor independent of T, N, and M stages. Through multivariate regression analyses, C1 stage emerged as an independent risk factor for poor prognosis (while controlling for T, N, and M stages), thereby satisfying this prerequisite for proceeding with survival analysis of individual AJCC and C-stage groups. We noted a consistent pattern with OS and DSS wherein survival of patients with C0 and C1 disease of each AJCC stage was significantly different, barring few exceptions. Furthermore, the magnitude of this separation between C0 and C1 stages within each overall AJCC stage was large enough to result in clustering of prognosis of different AJCC stages based on their C stage. This finding indicates the important predictive value of including C stage into the conventional TNM staging. In essence, the 5-year OS and DSS values for each overall traditional AJCC stage (ie, based on TNM staging) before the incorporation of C stage fall somewhere between 2 vastly prognostically different subgroups based their C-stage status, which become evident after the incorporation of C stage into the overall staging system.
The results of this study may have potential implications on the contemporary paradigm of management of colon cancer, which depends on the TNM-based AJCC stage.18,19 Currently, patients with stage I/II (node-negative) disease are generally treated with surgery alone, and patients with stage III (node-positive) disease are treated with surgery and adjuvant systemic chemotherapy. Thus, the decision to recommend adjuvant therapy is often driven mainly by node-positive status. However, analysis of every combination of N stage and C stage revealed that C1 stage portended poor prognosis to a similar magnitude as early nodal disease, and within each nodal substage, C1 disease portended a worse prognosis compared with patients with C0 disease and the correspondingly higher nodal stages. Patients with node-negative early-stage colon cancer with elevated preoperative CEA levels may have a poor prognosis similar to those with node-positive disease. In the current paradigm, such node-negative patients would not be offered adjuvant chemotherapy because their poor survival based on C stage had not been estimated. Our study suggests that based on their poor prognosis, these patients with node-negative C1 disease may be candidates for adjuvant chemotherapy.
The authors recognize that the current guideline recommendations by the American Society of Clinical Oncology and National Comprehensive Cancer Network do not include elevated CEA level as an indication or high-risk factor for adjuvant therapy in stage II colon cancer because of the lack of sufficient evidence to support its use.18,19 However, these recommendations have been based on several large adjuvant therapy trials and pooled analyses,20– 24 none of which have used preoperative CEA levels (C-stage) in risk stratification of patients. Given the evidence provided by this study, it is likely that the cohorts of patients with stage II disease randomized in these adjuvant therapy trials are a prognostically heterogeneous (by their C stage) group. Because of heterogeneity in their inherent C-stage–derived prognosis, adjuvant therapy may not have been seen to be beneficial when analyzed on an intent-to-treat basis. A retrospective study by Ogata et al25 concluded that an elevated preoperative CEA level may be an effective indicator for the need for adjuvant therapy in potentially curative stage II colon cancer. A study by Lou et al26 found that an elevated preoperative serum CEA level was the only predictive factor (odds ratio, 15.34; P = .004) for distant metastasis in pathologically T1 colorectal cancer after curative resection. These studies corroborate the interpretations made from the findings of our study.
Although C1 stage may be associated with poor outcomes (prognostic factor), it may not necessarily define or identify those who will benefit from current conventional chemotherapy (predictive factor). However, solely based on poor prognosis conferred by C1 stage, prospective testing on the use of multimodality therapy may be warranted on these patients, and patients in future trials should be risk stratified by the C stage before randomization. It may also be worthwhile exploring the benefit of avoiding or reducing chemotherapy or even close surveillance of patients with early nodal disease (N1a) with normal preoperative CEA levels.
A major limitation of our study is the lack of availability of adjuvant therapy information in the SEER database, and hence survival analysis and prediction of response to therapy based on adjuvant therapy information were not possible, raising the possibility of inaccuracy of risk estimation attributable to C1 stage. A study by Yeh et al27 that used a prospective database of 8861 consecutive patients with colorectal cancer with information on adjuvant therapy derived at-risk estimates for elevated CEA level (odds ratio, 1.61) that were similar to our current study (HR, 1.59) and a previous study17 (HR, 1.60), thereby providing corroborative support to our estimates. It may also be reasonably presumed that a large proportion of patients with stage III disease received adjuvant chemotherapy according to national guidelines, whereas most patients with stage II and stage I disease did not.28 Another limitation of this study is the limited information on the actual numerical value of CEA levels, the reliance on local interpretation of normal and elevated values, and the lack of a uniform cutoff value, although we believe that most US laboratories use similar and standardized reference values. Amri et al29 found that the use of the numerical value of CEA as a continuous variable may not help ascertain long-term prognosis on an individual case-by-case basis. Patients with missing serum CEA levels are unlikely to be random. However, in a prior study,17 the authors found that although missing CEA level information might be associated with overall worse prognosis, it is unlikely to affect the conclusion of the analysis comparing patients with normal and elevated CEA levels because C stage emerged as an independent prognostic factor affecting prognosis. Other limitations include lack of availability of information on staging modalities in the SEER database and the relatively smaller number of patients in some subgroups, especially the stage IIB and IIIA groups. Although we were unable to study the inclusion of C stage using the same reference values proposed by the AJCC Colorectal Working Group, this study would serve as enough proof of principle for the proposal using long-term survival data from nationwide registry information.
We conclude that the proposal of the Colorectal Working Group of the AJCC to include pretreatment CEA level into TNM staging, based on the modified cutoffs for CEA levels defined by SEER, is acceptable, feasible, and well supported by this population-based study. C-stage inclusion yields important AJCC stage–specific prognostic data, which may have important implications in multimodality therapy and future trials. We recommend routine pretreatment CEA testing for all patients with colon cancer and reporting outcomes by C stage in upcoming revisions of the AJCC Cancer Staging Handbook. Further research is warranted to evaluate the value of C stage in identifying those patients who may benefit from adjuvant systemic therapy.
Accepted for Publication: March 8, 2015.
Published Online: June 17, 2015. doi:10.1001/jamasurg.2015.0871.
Study concept and design: Thirunavukarasu, Talati, Munjal, Edge, Francescutti.
Acquisition, analysis, or interpretation of data: Thirunavukarasu, Talati, Attwood, Francescutti.
Drafting of the manuscript: All authors.
Critical revision of the manuscript for important intellectual content: Thirunavukarasu, Talati, Francescutti.
Statistical analysis: Thirunavukarasu, Talati, Munjal, Attwood.
Administrative, technical, or material support: Talati, Edge, Francescutti.
Study supervision: Edge, Francescutti.
Conflict of Interest Disclosures: None reported.