In Nova Scotia, Canada, a previous study of colorectal cancer (CRC) cases diagnosed between January 1, 2001, and December 31, 2005, found that patients with stage IIB CRC had similar 5-year overall survival (OS) to those with stage IIIC cancer. This study sought to examine factors contributing to the observed stage IIB outcome, specifically nodal harvest, receipt of chemotherapy, and use of a new coding system to derive stage.
The provincial cancer registry identified all CRC cases diagnosed during the study period and staged this cohort using the Collaborative Stage (CS) Data Collection System. All patients with stage II and III cancer in the cohort were examined. Kaplan-Meier (KM) survival curves compared 5-year OS for patients with stage IIB cancer based on the factors of interest, and compared patients with stage IIB cancer to those with stage IIA and III cancer.
OS for patients with stage IIB cancer (n = 187) was 44.7%, and differed depending on adequacy of nodal harvest (P = .005) and whether pathological or clinical/mixed evidence was used to derive stage (P = .013). Pathologically-staged patients with stage IIB cancer who had adequate nodal harvest had marginally improved OS compared to pathologically-staged patients who had inadequate nodal harvest (P = .07), and improved survival compared to patients with clinical/mixed stage (P = .004). Pathologically-staged patients with stage IIB cancer with adequate nodal harvest demonstrated similar 5-year OS to those with stage IIA and III cancer (P = .52 and P = .25, respectively). Cox proportional hazards models supported these findings.
Colorectal cancer (CRC) is a highly prevalent cancer in Canada, reflecting increased incidence and a 63% 5-year relative survival.1 Long-term survival is related to stage of disease at diagnosis, with lymph node (LN) status critical to determining stage, particularly in differentiating between stage II and III disease.2 Surgical resection is the primary treatment for CRC,3-5 with postoperative adjuvant chemotherapy provided according to clinical and/or pathological stage of disease.6-8 Receipt of adjuvant chemotherapy has clearly demonstrated benefit in patients with stage III CRC, although the benefit is less conclusive for patients with stage II cancer.
Although imperative for estimating prognosis and appropriate treatment planning, determining stage at diagnosis is also important for evaluating outcomes and cancer control measures. In 1990, the US National Institutes of Health recommended use of the American Joint Committee on Cancer (AJCC) tumor-node-metastasis (TNM) Classification and Stage system to move toward a standard staging system in both clinical practice and research trials.6 Concurrently, many epidemiologists and registrars have used the Surveillance, Epidemiology and End Results (SEER) Summary Stage system to monitor variations and trends in stage distribution.9 Recently, another data collection and coding framework, the Collaborative Stage (CS) Data Collection System, has been developed to resolve discrepancies in staging guidelines among the various major staging systems and provide a common coding scheme that meets clinical and epidemiological needs.10
CS is not a staging system, but rather a data collection/coding system to derive stage. CS collection requires variables (eg, tumor size, nodal extension) specific to disease site, and contains disease-specific algorithms to generate a TNM “best” stage group or a SEER Summary Stage. These algorithms incorporate pathological and/or clinical (eg, radiographic) information to derive stage groups. Specifically, when pathological information is unavailable or incomplete, a TNM stage group is derived using either mixed (clinical and pathological) or clinical information alone.
In Nova Scotia (NS), Canada, CS variables were collected for all individuals diagnosed with CRC over a 5-year period, and CS-derived AJCC 6th edition TNM stage groups were generated. Examination of these population-based data found that patients with stage IIB (T4N0M0) cancer had significantly worse 5-year overall survival (OS) than patients with stage IIIA (T1-2N1M0) and IIIB (T3-4N1M0) cancer (P < .0001 and P = .01, respectively) and similar OS to patients with stage IIIC (any TN2M0) cancer (P = .48).11 Although a worse survival among patients with stage IIB, compared with some patients with stage III cancer, has been reported elsewhere in both colon12, 13 and rectal14-16 cancer, the reasons for the poor outcome are relatively understudied. It may reflect clinical practice wherein a lower proportion of patients with stage IIB disease receive beneficial adjuvant chemotherapy, stage migration from variable LN harvest, or a greater than recognized prognostic significance of the depth of tumor invasion (T component of TNM staging).17 In the NS cohort, however, the similar 5-year OS estimates we observed for patients with stage IIB and IIIC disease may also reflect use of a coding system that uses both clinical and pathological data to derive a stage group.18 For example, the use of clinical data on nodal status, when LNs are pathologically unassessed, could potentially misclassify some patients with true stage III as having lower stage disease.
The purpose of this study was to examine factors that may contribute to the observed stage IIB outcome in the NS population, specifically adequate nodal harvest, receipt of chemotherapy, and use of the CS coding system to derive TNM stage.
MATERIALS AND METHODS
The Nova Scotia Cancer Registry (NSCR) adopted the CS system,18 in alignment with national standards set for the Canadian Cancer Registry for collection of registry-based stage data in Canada. The NSCR identified all CRC cases diagnosed in NS between January 1, 2001, and December 31, 2005, and undertook a comprehensive chart review to stage this cohort. This provincial registry contains case records for all NS residents diagnosed with cancer; nonresident cancer records were excluded from our cohort. Experienced coders reviewed the medical charts for each case and abstracted the elements required to determine a stage group at diagnosis. Validation of the CS data was performed by an external expert reviewer. Histological classification was determined by International Classification of Diseases (ICD) for Oncology19 codes: ICD-O3 C18 (excluding C18.1), C19, and C20. Cases diagnosed in individuals <20 years of age were excluded. If patients had more than 1 CRC diagnosis in the time period, only 1 case per patient was retained. The result was a patient-level cohort of 3501 individuals. Ethical approval was obtained from the Capital District Health Authority's Research Ethics Board.
For this study, all patients with stage IIB cancer in the population-based cohort were reviewed as well as all patients with stage IIA (T3 N0 M0) and III cancer for comparative purposes. The NSCR provided demographic, CS, and death data. CS data included information on nodal harvest. The patients with CS-derived TNM stage IIB disease were further divided into pathological stage (stage was derived from pathological evidence using pTpN and M, where M is either positive pM or a known cM), clinical stage (stage was derived entirely from clinical evidence using cTcNcM), and mixed stage (stage was derived from all other combinations of pathological and clinical evidence; eg, pTcNcM). The CS “Eval” fields recorded the pathologic or clinical source of evidence used. If a rectal cancer patient received neoadjuvant therapy, the clinical information on extension, nodal status, and metastases took precedence unless the pathologic evidence was more extensive following treatment. When the pathological evidence was more extensive, the resultant stage was considered pathologic.
Chemotherapy data were captured from NSCR data (which contains data on chemotherapy receipt at the 2 tertiary cancer centers in the province), the provincial physician billing database, and data obtained through comprehensive chart review. The chart review was performed on all patients for whom there was no indication of chemotherapy receipt in the administrative data files to determine whether the patient actually received chemotherapy. Further details on this review are reported elsewhere.20 The combination of data sources provided complete population-based chemotherapy data for patients with stage IIB CRC in NS. Data on comorbidities were obtained by linking the cohort to hospital discharge abstracts data, and a comorbidity score was computed based on the list of comorbid conditions developed by Elixhauser et al,21 excluding all cancer-related comorbidities.
Chi-square tests were performed to examine differences in demographic, treatment, and staging factors across stages. Given the potentially high interaction between the variables of interest (nodal harvest, chemotherapy receipt, and source of evidence) and their impact on the outcome, our a priori analytic approach was to examine multiple strata through separate analyses rather than combine these variables in a single regression model. Kaplan-Meier (KM) survival curves and corresponding log-rank tests were used to compare OS for all stage II and III substages, and then for patients with stage IIB cancer with the factors of interest. The start time for all curves was the date of diagnosis. For patients with stage IIB cancer, KM survival curves were first generated by the following subgroups:
2Receipt versus nonreceipt of adjuvant chemotherapy; and
3Pathological TNM-derived stage (CSp) versus clinical or mixed TNM-derived stages (CSc/m).
Given the findings of this initial assessment, KM curves were generated to compare OS among the following stage IIB subgroups:
4CSp + adequate LN harvest versus CSp + inadequate LN harvest versus CSc/m; and
5CSp + adequate LN harvest versus all patients with stage IIA versus all patients with stage III cancer.
Cox proportional hazards models were performed for each comparison, controlling for age, sex, and comorbidity. All analyses were repeated for colon cancer patients only; analyses for rectal cancer only were not feasible due to the small number of patients. All causes of death were included, and patients were followed until December 31, 2009. Analyses were performed using SAS, version 8 (SAS Institute, Durham, NC). The P values reported are 2-sided, with P < .05 considered statistically significant.
Over the 5-year period, 187 adult patients were diagnosed with stage IIB CRC in NS (122 and 65 with colon and rectal cancer, respectively). Table 1 presents descriptive data for these patients. Furthermore, 19 of 65 (29.2%) rectal cancer patients received neoadjuvant therapy. All were staged based on clinical or mixed evidence.
Table 1. Descriptive Data for All Study Patients
Stage IIB (n = 187)
Stage IIA (n = 882)
Stage III (n = 916)
Colon (n = 122)
Rectum (n = 65)
Evidence for stage
Figure 1 presents stage-specific KM survival curves for all patients with stage II and III cancer. The 5-year OS for patients with stage IIB cancer was 44.7%. There was no difference in OS between patients with stage IIB colon and stage IIB rectal cancer (P = .32). Figures 2, 3, and 4 present KM survival curves for patients with stage IIB disease according to nodal harvest, chemotherapy, and CS source of evidence, respectively. Five-year OS significantly differed between patients with adequate versus inadequate nodal harvest (P = .005), and between patients with CSp versus CSc/m (P = .013). All differences remained after controlling for age, sex, and comorbidity. No significant differences in OS were identified among patients with stage IIB disease who did and did not receive adjuvant chemotherapy (P = .176).
Patients with stage IIB cancer with CSp and adequate LN harvest had marginally improved survival compared to patients with CSp and inadequate LN harvest (P = .07), and a significantly improved survival compared to patients with CSc/m (P = .004; Fig. 5). When compared to patients with stage IIA and III cancer, patients with stage IIB cancer with CSp and adequate LN harvest demonstrated similar 5-year survival (P = .52 and P = .25, respectively; Fig. 6). Table 2 provides the 5-year OS estimates and hazard ratios (HR) for patients with stage IIB cancer by all of the strata examined. Table 3 provides the same estimates and HR for stage IIB colon cancer patients. All findings presented above were also observed in the subcohort of colon cancer patients.
Table 2. Overall Survival (OS) Estimates and Hazard Ratios (HR) for Patients With Stage IIB (T4N0M0) Colorectal Cancer by Nodal Harvest, Chemotherapy Receipt, and Pathological Versus Clinical/Mixed TNM-Derived Stages (N = 187)
Kaplan-Meier, Log-Rank Tests
Multivariate Cox Regression
5-y OS (%)
P value reflects comparison between the 2 subgroups (≥12 lymph nodes, <12 lymph nodes) within the pathologic group.
Table 3. Overall Survival (OS) Estimates and Hazard Ratios (HR) for Patients With Stage IIB (T4N0M0) Colon Cancer by Nodal Harvest, Chemotherapy Receipt, and Pathological Versus Clinical/Mixed TNM-Derived Stages (N = 122)
Kaplan-Meier, Log-Rank Tests
Multivariate Cox Regression
5-y OS (%)
P value reflects comparison between the 2 subgroups (≥12 lymph nodes, <12 lymph nodes) within the pathologic group.
This study examined whether nodal harvest and/or receipt of chemotherapy contributed to the poor stage IIB survival in a population-based cohort of CRC patients, and/or whether the observed survival disadvantage reflects use of a coding system that includes clinical/mixed evidence into staging classifications. Differences in survival existed between stage IIB patients who had adequate LN harvest versus those who did not (P = .005), and between patients who were staged solely using pathological evidence and those staged using clinical/mixed evidence (P = .013). Further stratified analysis found that when patients were subdivided into pathologically-staged adequate LN harvest and pathologically-staged inadequate LN harvest, there was no statistically significant difference in survival between groups, but those with pathologically-staged adequate LN harvest had superior survival to patients with clinical/mixed stage. These findings suggest that the inclusion of clinical/mixed evidence into the stage classification algorithms used by CS and, perhaps to a lesser extent, the adequacy of nodal harvest contribute to the poor outcome we observed in the patient population classified as having stage IIB disease. The results of an additional Cox regression model with age, sex, comorbidity, and all the variables of interest (nodal harvest, chemotherapy receipt, and source of evidence) included, performed after conducting our a priori analysis, were consistent with our planned analysis with source of evidence appearing to be the stronger factor after adjusting for all other factors (nodal harvest: HR = 0.613, P = .072; chemotherapy receipt: HR = 0.656, P = .085; source of evidence: HR = 0.535, P = .003).
That patients who are staged using clinical or mixed evidence have lower survival estimates is not surprising. Indeed, although the AJCC TNM staging system includes rules for pretreatment clinical staging (designated by cTNM), it advises against including individuals without histopathological confirmation in survival analyses due to their known poorer survival.2, 22 Using the CS system, however, when a stage can be derived using the information available to the coder (which, when pathological information is unavailable or incomplete, may include solely clinical or mixed information), this stage is included in cancer registry data and used in surveillance analyses, even though some of these patients likely have a poor probability of long-term survival. An example of such a situation would be a patient with computed tomography scan evidence of a rectal cancer with bladder base involvement without lymphadenopathy. If surgical resection was not performed in such a patient, CS would generate T4N0M0 classifications (ie, stage IIB), although clearly there is a significant likelihood that nodal metastases exist.
CS coding, as with any coding system, affects how patients are classified into stage groups, which are then evaluated for survival. In our cohort, 48 patients (25.7%) with stage IIB disease were staged using clinical information alone, whereas 18 patients (9.6%) were staged using a mixture of clinical and pathological information (these proportions are much lower in patients with stage IIA and III cancer; Table 1). Several factors contribute to these high proportions in patients with stage IIB cancer, one of which is that all 19 rectal cancer patients who received neoadjuvant therapy were staged using clinical/mixed evidence. This is reflected in the high proportion of rectal cancer patients who had undergone resection (53.3%) with stage derived from clinical/mixed evidence, despite the availability of a tissue sample. Of these patients, >80% had their stage grouping derived from clinical/mixed information, according to CS rules, because they received neoadjuvant treatment or because no LNs were removed/examined during their resection.
Another contributing factor is the high proportion (30.8%) of rectal cancer patients who did not receive curative-intent resection. Although we cannot ascertain the reason(s) for nonresection from administrative data, a plausible reason is that many of these patients actually had known or strongly suspected metastatic disease. However, the M classification was coded nonmetastatic (M0), using clinical information, perhaps due to incomplete/contradictory documentation to indicate distant metastases or because the clinician used ambiguous terminology (eg, “suggests,” “worrisome,” “encasing”) to denote metastases (CS coding identifies which ambiguous terms coders should consider/not consider involvement18). Subsequently, patients may be down-staged using CS rules, which state that “if there is doubt concerning the T, N, or M classification to which a particular case should be assigned, then the lower (less advanced) category should be assigned.”18 Thus, the use of clinical (or mixed) evidence to derive stage could potentially misclassify some patients with advanced stage as having lower stage disease, contributing to the outcome we observed in our patients with stage IIB cancer. Additional, but uncommon, reasons for not undergoing resection include chemoradiotherapy given as the primary treatment, complete response to neoadjuvant therapy that obviates a decision for resection, and patients declining definitive surgical treatment for their diagnosis.
In addition to using clinical or mixed evidence to derive a “best” stage, the specific CS rules for computing stage group18 have been shown to differ from those of the AJCC TNM and SEER Summary Stage coding manuals.9, 23 As a result, it is perhaps unreasonable to expect identical stage distribution and, subsequently, survival across systems. In a comparison of CS-derived versus directly-coded SEER Summary Stage, researchers found that, for many cancer sites, differences in stage distribution in 40 population-based US registries between 2003 and 2004 is likely attributable, at least in part, to implementation of the CS system.9 As surveillance organizations transition to CS data collection and coding and begin to report outcomes using CS-derived stage groups, it is important to recognize that any observed changes in stage distribution and/or survival may be a result of using a different coding and stage classification system. Accordingly, researchers/surveillance organizations should continue to assess and publish on CS methodology to improve our understanding of this new coding system and its potential impact on cancer stage groupings. Given our findings, we also recommend that researchers using CS-derived stage groups control for pathological versus clinical/mixed stage when using regression models.
Inadequate nodal harvest may have also contributed to poorer survival in our patients with stage IIB disease. We detected differences in 5-year OS when patients were separated by adequacy of nodal harvest (P = .005). Improved survival among CRC patients with greater nodal harvests has been demonstrated repeatedly,24-28 including improved survival in stage II disease when ≥12 LNs are evaluated.29-31 When we examined only patients who were staged using pathological evidence, the difference in survival between those with ≥12 LNs examined and those with <12 LNs examined was marginally statistically significant (P = .07), and likely did not reach the a priori significance level due to small numbers in each group (n = 40 and n = 81, respectively). Yet, upon examining the KM curves, there is clearly an important difference in survival between the 2 groups. Conversely, receipt of chemotherapy did not appear to be a determinant of long-term survival in patients with stage IIB disease. However, it is possible that any potential survival differences due to chemotherapy receipt were undetected due to the small number of patients with stage IIB cancer.
When patients with stage IIB cancer with pathologic stage and adequate LN harvest were compared to those with stage IIA and III cancer, there were no significant survival differences among the groups. This contrasts with findings from the entire stage IIB cohort (regardless of nodal harvest status or source of staging evidence; Fig. 1), which demonstrated significantly poorer survival than in patients who had stage IIA and III cancer. This suggests that when stage is based on pathologic assessment and when that pathologic assessment results in an adequate number of LNs examined (increasing the probability of “true” LN-negative disease), the survival curve for patients with stage IIB cancer moves closer to what would be expected in a cancer staging system, which ought to demonstrate prognostic changes consistent with advancing stage. A recent study that assessed the seventh edition AJCC TNM revised TN categorizations for colon cancer demonstrated that relative survival for T4N0 lesions improves as nodal harvest increases.12 The work of Gunderson and colleagues also suggests the biological interaction between depth of invasion (T) and nodal status (N) is more complex than previously recognized12, 14; this complexity is observed in the seventh edition AJCC TNM revisions for CRC staging.
The strength of this study is the assessment of staging and survival data for a 5-year population-based cohort of CRC patients. Another strength is the use of population-based chemotherapy data to assess whether receipt of chemotherapy contributed to our survival findings. In both Canada and the United States, there are gaps in the availability and completeness of chemotherapy data in administrative databases20, 32-34; in this study, linked administrative and chart review data were used to complete chemotherapy capture for all patients with stage IIB cancer. Despite our concerted attempt to complete chemotherapy data, however, a small number of patients in one health region (n = 10; 5.3% of patients with stage IIB disease) may have received chemotherapy in an adjoining province, which would not have been captured in our chemotherapy data sources.
Finally, this study presents important population-based CS data, which is the coding system now used throughout many parts of North America for cancer registry purposes. The CS system provides advantages for deriving stage over the main staging systems: in particular, the CS manual includes more explicit/detailed coding instructions,9 and CS coding generates fewer unknown stages9, 23, 35 because it uses a broader range of information to generate a “best” stage group. Thus, although a potential limitation of CS algorithms is use of weaker staging evidence to classify some patients, the alternative would be using stage groups classified by pathologic evidence alone and therefore would not be population-based. Nonetheless, our findings, as well as those of others,9, 23 suggest that researchers continue to evaluate and report on CS methodology so we can better understand the strengths and limitations of this new coding system as well as any potential epidemiological implications (eg, when comparing surveillance data over time and between jurisdictions that use CS methodology versus those that do not). These evaluations should include comparisons of population-based CS-derived and directly-coded AJCC TNM (or SEER Summary Stage) outcomes. One means of accomplishing this is to stage data using the old and new, or revised, systems (ie, “double-code”) as part of cancer registry implementation processes.
There are several potential limitations of this study. The first includes miscoding and errors applying the relatively new CS system. However, validation of CS data was performed by an external reviewer with extensive experience in cancer staging and cancer registry management. The second involves the small number of patients with stage IIB disease in our cohort, especially rectal cancer patients, which may have limited our nodal harvest and chemotherapy analyses. Although the population-based cohort consists of all individuals diagnosed with stage IIB CRC in one province over a 5-year period, the total number of patients was only 187. A third limitation is that our cohort was staged using version 1 of the CS manual. It is possible that version 2 has modified some of its coding rules to align more closely with that of the AJCC TNM system. Even if such changes have occurred, however, CS continues to use clinical information in the absence of pathological findings to derive TNM stage. Finally, version 1 derives TNM stage groups based on the AJCC sixth edition TNM staging guidelines. However, the updated seventh edition AJCC TNM manual subdivides T4 lesions into more granular categories based on survival estimates (with T4a and T4b representing stage IIB and IIC disease, respectively). Because our stage IIB survival is similar to observed survival estimates for patients with stage IIC disease in other studies that used the seventh edition manual,12, 14 it is possible that a high proportion of our patients with stage IIB cancer in fact have T4b tumors. Given our data capabilities, we were unable to determine the proportion of T4b tumors in our cohort.
In conclusion, we found that use of the CS system in classification of patients appears to have contributed, at least in part, to the observed worse survival for patients with stage IIB versus stage III cancer. Moreover, adequacy of nodal harvest may have influenced survival in this cohort, although its impact was less when stage was derived using only pathologic evidence. Given our findings, we recommend further assessment to better understand CS methodology and its potential impact on stage grouping. This is true for the introduction of any new or revised coding or stage classification system, and may be particularly important because revisions to staging systems are likely as new prognostic information (eg, molecular markers) becomes incorporated into coding rules and algorithms.
We gratefully acknowledge the work of Nova Scotia Cancer Registry personnel in collaborating with us to conduct the population-based chart review to stage all patients with colorectal cancer. We also acknowledge Cynthia Kendell for her helpful review of the manuscript.
This study was supported by a Team Grant (grant AQC-83513) from the Canadian Institutes of Health Research (CIHR), and local funding partners (Cancer Care Nova Scotia, Nova Scotia Department of Health, Capital District Health Authority, Dalhousie Medical Research Foundation, and Dalhousie University's Faculty of Medicine).
CONFLICT OF INTEREST DISCLOSURES
Maureen MacIntyre currently sits on the project management committee that oversees management and development of the Collaborative Stage Data Collection System. She receives no payment for service to this committee. Furthermore, she did not sit on the committee when this study was conducted and written. The remaining authors made no disclosure.