Fax: (215) 955-5681
Occult tumor burden contributes to racial disparities in stage-specific colorectal cancer outcomes
Article first published online: 1 SEP 2011
Copyright © 2011 American Cancer Society
Volume 118, Issue 9, pages 2532–2540, 1 May 2012
How to Cite
Hyslop, T., Weinberg, D. S., Schulz, S., Barkun, A. and Waldman, S. A. (2012), Occult tumor burden contributes to racial disparities in stage-specific colorectal cancer outcomes. Cancer, 118: 2532–2540. doi: 10.1002/cncr.26516
- Issue published online: 20 APR 2012
- Article first published online: 1 SEP 2011
- Manuscript Accepted: 22 JUL 2011
- Manuscript Revised: 21 JUL 2011
- Manuscript Received: 23 JUN 2011
- racial disparities;
- colorectal cancer;
- molecular staging;
- occult tumor burden;
- guanylyl cyclase C;
There are differences in outcomes in blacks compared with whites with lymph node–negative (pN0) colorectal cancer. Recurrence in pN0 patients suggests the presence of occult metastases undetected by conventional approaches. This study explores the association of racial differences in outcomes with occult tumor burden in regional lymph nodes.
Lymph nodes (range, 2-159) from 282 prospectively enrolled pN0 colorectal cancer patients followed for a median of 24 months (range, 2-63 months) were subjected to molecular analysis. Occult tumor burden was estimated by quantifying the expression of GUCY2C, a biomarker for metastatic colorectal cancer cells. Risk categories defined using occult tumor burden was the primary outcome measure. Association of prognostic variables and risk were defined by multivariate polytomous logistic regression.
Occult tumor burden stratified this cohort of 259 whites and 23 blacks into categories with low (60%; recurrence rate [RR] = 2.3%; 95% confidence interval [CI], 0.1%-4.5%), intermediate (31%; RR = 33.3%; 95% CI, 23.7%-44.1%), and high (9%; RR = 68.0%; 95% CI, 46.5%-85.1%; P < .001) risk. Blacks compared with whites exhibited 4-fold greater occult metastases in individual lymph nodes (P < .001). Multivariate analysis revealed that race (P = .02), T stage (P = .02), and number of lymph nodes collected (P = .003) were independent prognostic markers of risk category. Blacks compared with whites were more likely to harbor levels of occult tumor burden, associated with the highest recurrence risk (adjusted odds ratio = 5.08; 95% CI, 1.69-21.39; P = .007).
Racial disparities in stage-specific outcomes in colorectal cancer are associated with differences in occult tumor burden in regional lymph nodes. Cancer 2012. © 2011 American Cancer Society.
There is a widening racial gap in mortality from colorectal cancer, the fourth most common incident cancer and the second leading cause of cancer death in the United States.1-4 For example, whereas disease-specific mortality has decreased 54% for non-Hispanic white men, non-Hispanic black men have experienced an increase of 28%, since 1960.5 Racial differences in mortality reflect tumor clinicopathologic characteristics, including advanced stage of disease at diagnosis associated with poorer outcomes in blacks compared with whites.6-10 In turn, differences in disease stage at diagnosis reflect disparities in socioeconomic status and access to quality health service.5-11 Tumor characteristics, socioeconomic status, and health services access contribute about 50% to excess mortality reflecting race.8, 11, 12 Other factors underlying race-based excess mortality in colorectal cancer remain undefined.5
Beyond clinicopathologic differences at diagnosis, there is an underappreciated racial disparity in stage-specific mortality in colorectal cancer.5, 6, 8 For patients with regionally advanced disease (lymph node–positive; stage III), blacks experience 10% excess mortality compared with whites.5, 6, 8 This difference is amplified in patients with local disease (lymph node–negative [pN0]; stages I and II), where blacks exhibit 40% excess mortality compared with whites.5, 6, 8 These stage-specific disparities appear to be one primary driver of overall differences in mortality in blacks and whites.5, 6, 8 Socioeconomic status is one factor contributing to these racial disparities in stage-specific outcomes.5, 8, 12 Other clinicopathologic processes contributing to these differences have not been defined.5, 6, 8, 11 However, the predominance of this racial gap in the earliest stages (pN0) of disease, which receive minimal postsurgical intervention,2, 3 suggests contributions by factors other than therapeutic application, acceptance, or compliance.8
Regardless of the underlying mechanisms, diagnostic methods that categorize prognostic risk could identify patients vulnerable to excess mortality who might benefit from clinical management to diminish this racial gap. The most informative prognostic marker of survival and predictive marker of response to therapy in colorectal cancer is the presence of tumor cells in regional lymph nodes.1-3, 13, 14 Despite their importance, techniques to assess nodal metastases remain imperfect, and about 25% of patients with histology-negative lymph nodes die of recurrent disease, reflecting the presence of occult metastases that escape detection by conventional methods.1, 3, 13-18 There is an emerging paradigm in which specific tumor markers are coupled with nucleic acid amplification techniques, such as the reverse transcriptase-quantitative polymerase chain reaction (RT-qPCR), to identify clinically meaningful occult metastases.13-18
The intestinal tumor suppressor GUCY2C (guanylyl cyclase C) is the receptor for the paracrine hormones guanylin and uroguanylin, gene products universally lost early in intestinal neoplasia.19, 20 Loss of hormone expression silences GUCY2C signaling, which contributes to transformation by promoting proliferation, crypt hypertrophy, metabolic remodeling, and genomic instability.20 Highly selective expression by intestinal epithelial cells normally and universal overexpression by intestinal tumor cells suggested that GUCY2C might be a specific molecular marker for metastatic colorectal cancer.17, 18, 21-24
Recently, the prognostic utility of occult lymph node metastases detected by GUCY2C RT-qPCR in pN0 colorectal cancer patients was validated in a prospective, multicenter, blinded clinical trial.18, 25 The categorical presence (yes/no) of occult metastases was a powerful independent predictor of time to recurrence and disease-free survival in pN0 patients.18 This utility of GUCY2C for detecting occult metastases in lymph nodes that predict clinical outcomes in pN0 colorectal cancer has been independently validated.26-28 This paradigm was refined to quantify occult tumor burden (how much) across the regional lymph node network, which classified patients with near-zero risk, those with elevated (33%) risk, and those with high (70%) risk of unfavorable outcomes.25 The current study examines the relation between occult tumor burden and disparities in outcomes in blacks and whites with pN0 colorectal cancer. It considers the relative level of GUCY2C mRNA expression in lymph nodes from black and white patients from the prospective trial.18 Moreover, these analyses consider the contribution of race to prognostic stratification of risk by occult tumor burden.25
MATERIALS AND METHODS
This prospective observational trial at 9 centers in the US and Canada explored the prognostic utility of GUCY2C RT-qPCR in lymph nodes of pN0 colorectal cancer patients.18, 25 Investigators and clinical personnel were blinded to results of molecular analyses, and laboratory personnel and analysts were blinded to patient and clinical information. The original study had at least 80% power to detect a hazard ratio of 1.6 or larger (P < .05, 2-sided) based on a categorical (yes/no) assessment of GUCY2C status. The study protocol was approved by the Institutional Review Board of each participating hospital. The 299 pN0 patients who met eligibility criteria provided 7310 lymph nodes (range, 2-159; median, 21 lymph nodes/patient) for histopathologic examination, of which 3093 nodes (range, 1-87; median, 8 lymph nodes/patient) were obtained by fresh dissection and eligible for analysis by RT-qPCR.18, 25 Disease status, obtained in routine follow-up by treating physicians, was provided for all patients through December 31, 2009.
Patients and Tissues
Between March 2002 and June 2007, we enrolled 299 stage 0-II pN0 colorectal cancer patients who provided informed consent in writing prior to surgery at 1 of 7 academic medical centers and 2 community hospitals in the US and Canada. Patients were ineligible if they had a previous history of cancer, metachronous extra-intestinal cancer, or perioperative mortality associated with primary resection. For all eligible patients, preoperative and perioperative examinations revealed no evidence of metastatic disease. Lymph nodes, and when available tumor specimens (51%), were dissected from colon and rectum resections and frozen at −80°C within 1 hour to minimize warm ischemia. Half of each resected lymph node was fixed with formalin and embedded in paraffin for histopathologic examination. Lymph node specimens were subjected to molecular analysis if 1) tumor samples, where available, expressed GUCY2C mRNA above background levels in disease-free lymph nodes (>30 copies), and 2) at least 1 lymph node was provided that yielded RNA of sufficient integrity.18, 24, 25 Analysis of the 3093 lymph nodes available from the 299 pN0 patients revealed 236 nodes (7.6%) yielding RNA of insufficient integrity, which were omitted from further analysis.18, 25 For 2 patients, all lymph nodes harvested were of insufficient integrity, resulting in their exclusion.18, 25 Moreover, GUCY2C expression in tumors was below detectable levels in 6 patients who also were excluded.18, 25 Of the 291 remaining eligible patients, 23 self-identified as black, 259 as white (Table 1), and 9 were of another race or their race could not be identified. These analyses focus on the 282 white and black patients because other categories were too small for conclusive analysis.
|Overall||Low Risk (n=170)||Moderate Risk (n=88)||High Risk (n=24)||Pb|
|Age, y, at diagnosis||.83|
|Surgery + chemotherapy||64||59.4||28.1||12.5|
|Lymph nodes harvested||.002c|
RNA was extracted from tissues by a modification of the acid guanidinium thiocyanate-phenol-chloroform extraction method.22, 23 Briefly, individual tissues were pulverized in 1.0 mL Tri-Reagent (Molecular Research Center, Cincinnati, OH) with 12 to 14 sterile 2.5-mm zirconium beads in a bead mill (Biospec, Bartlesville, OK) for 1 to 2 minutes. Phase separation was performed with 0.1 mL bichloropropane, and the aqueous phase re-extracted with 0.5 mL chloroform. RNA was precipitated with 50% isopropanol and washed with 70% ethanol. Air-dried RNA was dissolved in water, concentration determined by spectrophotometry, and stored at −80°C.
GUCY2C mRNA was quantified by RT-qPCR employing an established analytically validated assay.24 The EZ RT-PCR kit (Applied Biosystems, Foster City, CA) was employed to amplify GUCY2C mRNA from total RNA in a 50-μL reaction. Optical strip-tubes were used for all reactions, which were conducted in an ABI 7000 Sequence Detection System (Applied Biosystems). In addition to the kit components (50 mM Bicine [pH 8.2], 115 mM KOAc, 10 μM EDTA, 60 nM ROX, 8% glycerol, 3 mM Mg(OAc)2, 300 μM each dATP, dCTP, and dGTP, 600 μM dUTP, 0.5 U uracil N-glycosylase, and 5 U rTth DNA polymerase), the reaction master mix contained 900 nM each of forward (ATTCTAGTGGATCTTTTCAATGACCA) and reverse primers (CGTCAGAACAAGGACATTTTTCAT), 200 nM Taqman probe (FAM-TACTTGGAGGACAATGTCACAGCCCCTG-TAMRA), and 1 μg RNA template. The housekeeping gene β-actin was amplified, employing similar conditions except that forward (CCACACTGTGCCCATCTACG) and reverse (AGGATCTTCATGAGGTAGTCAGTCAG) primers were 300 nM each, whereas the Taqman probe (FAM-ATGCCC-X(TAMRA)-CCCCCATGCCATCCTGCGTp) was 200 nM. The thermocycler program employed for RT included: 50° × 2 minutes, 60° × 30 minutes, 95° × 5 minutes; and for PCR included: 45 cycles of 94° × 20 seconds, 62° × 1 minute. Reactions were performed at least in duplicate and results averaged.
Procedures for reporting statistical methods, including validation procedures for ensuring the accuracy of estimates of hazard ratios and P-values, were specifically guided by the REMARK Guidelines.25, 29 Statistical methods for estimating GUCY2C and β-actin mRNA expression by logistic regression analysis have been described.18, 30 A linear mixed effects model of GUCY2C relative expression across all lymph nodes from eligible patients included random effect of patient, fixed effects of center, and race. This linear model was used to determine differences in GUCY2C expression in patient lymph nodes based on race, after adjusting for potential center-to-center differences and for repeated measures within patient.
The primary clinical end point was molecular risk category (low, intermediate, high), based on time to recurrence and recursive partitioning analysis.25 Recursive partitioning was implemented in the R routine RPART.25, 31 This algorithm tests, across all possible variables and levels, for the variable that optimally identifies discrete groups within the study population. The process repeats recursively until a stopping criterion, predefined here as the software default of any subgroup with <20 participants, is achieved.31 Cross-validation (10-fold) during model fitting provided model stability and accuracy and avoided overfitting. This algorithm was applied using quantitative measures of occult tumor burden as variables for risk stratification. Metrics of occult tumor burden by GUCY2C qRT-PCR included median copy number, maximum copy number, median relative (normalized to β-actin) expression, maximum relative expression, total copy number, and total relative expression across lymph nodes, and the total number of GUCY2C-positive lymph nodes.18, 25, 30 Time to recurrence served as the outcome in these analyses. Categories of low, medium, and high risk for time to recurrence were defined by amalgamation.32
Univariate analysis of association of molecular risk category with demographic and prognostic factors was completed using the chi-square test of association. Multivariate analyses using polytomous logistic regression33 employed risk level and established prognostic variables including T stage, grade, lymphovascular invasion, therapy, anatomic location, number of lymph nodes collected for histopathology, and race.1, 3, 18, 25 Initial multivariate models included all established prognostic measures regardless of significance, and a manual backward stepwise approach was used to establish the final model of association with occult tumor burden risk level. Variables with the least association with outcome were removed one at a time until all remaining variables were significant by a type 3 test of association at P < .05. Predicted conditional probabilities and 95% 2-sided confidence intervals were estimated from the final multivariate model. These probabilities are reported to demonstrate the contribution of each variable to the final model of molecular risk strata. Exact adjusted odds ratios were calculated and reported for factors with small cell sizes in multivariate models when appropriate.
Confidence intervals for raw survival rates were computed by the exact method of Clopper-Pearson.34 All tests were 2-sided, and P < .05 was considered statistically significant. All analyses were performed with R v 2.11.2, SAS v9.2 (SAS Institute, Cary, NC).
The 282 black and white pN0 patients had a mean age of 68 years (range, 26-90 years) at diagnosis, and 55% were male (Table 1). Clinicopathologic features, including depth of tumor penetration (T1/2, T3, T4) and tumor anatomic location (right, left, rectal), were similar to national experience.1, 3, 4, 18, 25 Patients with colon cancer represented 85.9%, whereas those with rectal tumors comprised 14.1%. Blacks comprised 7.9% of the total population enrolled, nearly identical to the national average for disease-specific racial distribution.6, 7, 11 There were no significant differences in clinicopathologic characteristics between black and white patients (Table 2).
|Overall||Black (n=23)||White (n=259)||Pb|
|Age, y, at diagnosis||.55|
|Surgery + chemotherapy||64||26.1||22.3|
|Lymph nodes harvested||.69|
Occult Tumor Burden and Risk Stratification
Clinical outcomes in pN0 colorectal cancer patients were analyzed by recursive partitioning using metrics of occult tumor burden estimated by GUCY2C RT-qPCR.25 Based on time to recurrence, GUCY2C RT-qPCR stratified pN0 patients into categories in which 170 (60%) patients exhibited low (MolLow), 88 (31%) exhibited intermediate (MolInt), and 24 (9%) exhibited high (MolHigh) (P < .001) risk of disease recurrence (Fig. 1). All but 4 of the MolLow patients remained free of disease during follow-up (recurrence rate [RR] = 2.3%; 95% confidence interval [CI], 0.1%-4.5%); 29 MolInt patients developed recurrent disease (RR = 33.3%; 95% CI, 23.7%-44.1%); and 16 MolHigh patients developed recurrent disease (RR = 68.0%; 95% CI, 46.5%-85.1%; P < .001; Fig. 1).
Univariate analysis by chi-square test revealed the expected relation between advanced T stage, occult tumor burden, and risk category (P = .008; Table 1). Similarly, molecular staging depended on collecting 13 or more lymph nodes from each patient (P = .002; Table 1), recapitulating established enhancements in histopathologic staging by increased nodal harvests.1, 3, 35, 36
Occult Tumor Burden and Race
Individual lymph nodes from blacks compared with whites harbored 4-fold greater quantities of metastatic tumor cells (P < .001; Fig. 2) identified by GUCY2C RT-qPCR.18, 24 Moreover, blacks harbored a greater burden of occult metastatic tumor across their lymph node network associated with the highest prognostic risk compared with white patients (P = .002; Fig. 2; Table 1). Multivariate analyses revealed that blacks exhibited occult tumor burden associated with the greatest prognostic risk regardless of T stage or number of lymph nodes collected (Fig. 3).
Occult Tumor Burden Is an Independent Prognostic Variable Associated With Racial Disparities in Outcomes
Multivariate analyses employing polytomous logistic regression (Table 3) confirmed that race (P = .02), T stage (P = .02), and number of lymph nodes collected (P = .003) are independently associated with occult tumor burden and stratification into risk categories (Table 3). Patients with T3/T4 tumors were more likely to be categorized as high risk versus low risk (adjusted odds ratio 6.00 [1.69-21.39]; P = .006) compared with patients with T1/T2 tumors. Similarly, patients providing 13 or more lymph nodes were more likely to be categorized as high risk (adjusted odds ratio 8.10 [1.31-∞]; P = .02) compared with patients with fewer lymph nodes collected. Blacks were more likely to be categorized as high risk on the basis of occult tumor burden compared with whites (adjusted odds ratio 5.08 [1.69-21.39] P = .007).
|Black (moderate vs low)||23||1.03||(0.36-2.94)||.95a|
|Black (high vs low)||5.08||(1.55-16.65)||.007a|
|T3/T4 (moderate vs low)||165||1.25||(0.74-2.12)||.41a|
|T3/T4 (high vs low)||6.00||(1.69-21.39)||.006a|
|Lymph nodes harvestedc||.003b|
|≥13 (moderate vs low)||223||0.43||(0.28-1.01)||.06a|
|≥13 (high vs low)||8.10||(1.31-∞)||.02a|
There is a well-established racial disparity in disease mortality in blacks compared with whites with colorectal cancer.5-7, 11 One primary contributor to this disparity is stage-specific differences in cancer survival.5-8, 11, 12 Mechanisms underlying differences in stage-specific outcomes in blacks and whites continue to be debated.5, 6, 8, 11 However, there have been no methods, to date, that identify vulnerable patients at increased prognostic risk that eliminate these racial disparities. Here, we demonstrate that blacks compared with whites exhibit higher levels of occult metastatic tumor cells in regional lymph nodes. These metastases are associated with a greater proportion of blacks compared with whites harboring higher levels of occult tumor burden across their regional lymph node networks. In turn, this occult tumor burden is associated with racial disparities in stage-specific prognostic risk. Indeed, occult tumor burden was an independent marker of excess prognostic risk in blacks. These analyses suggest that racial disparities in mortality, in part, reflect differences in clinically undetected tumor metastasis in blacks compared with whites, revealed by occult tumor burden analysis in regional lymph nodes. This study suggests that quantifying occult tumor burden in regional lymph nodes can identify patients, regardless of race, that are at greatest risk for developing recurrent disease.
Disproportionate occult metastatic disease in blacks compared with whites could reflect race-based differences in staging accuracy.5, 8 In this model, blacks have a greater risk of being understaged. For example, blacks with regional metastases (stage III) might be categorized as having local disease (stage I or II) more frequently than whites. However, systematic race-based staging misclassification should minimize racial disparities in stage at diagnosis.8 In the context of established differences in stage at diagnosis, a contribution of understaging to race-based differences in occult metastatic disease seems less likely.5-7, 11 In the current study, patients of both races were treated at the same hospitals by the same health care providers. They produced equivalent numbers of lymph nodes for staging, which were histologically evaluated by the same pathologists at each institution. These considerations also reduce the likelihood of race-based staging misclassification based on differences in health care services in the current study.5, 8
Differences in occult tumor burden also could reflect race-based disparities in treatment availability, application, or compliance.10, 37, 38 In this paradigm, whites, but not blacks, would receive neoadjuvant therapy prior to surgery that reduces or eliminates metastatic disease in regional lymph nodes inevident by standard histopathology but discernable by molecular analyses. The current study exclusively involved pN0 patients. Colon cancer patients, comprising 86% of the trial cohort, did not receive neoadjuvant treatment, which is required to eliminate nodal metastases at initial staging. Stage II rectal cancer patients typically receive neoadjuvant chemoradiotherapy. However, the 2 blacks with rectal cancer comprised a cohort too small to account for stage-specific racial disparities in occult metastases in this study.
One principal factor contributing to racial disparities in overall mortality in colorectal cancer is the advanced stage of disease at which blacks compared with whites present at the time of diagnosis.5-8, 11 In turn, late stage at presentation reflects a delay in detection and diagnosis.5, 6, 8, 39 Factors contributing to delayed cancer diagnosis are manifold and include access to quality health services, cultural beliefs, education level, and the ability to comply with diagnostic and treatment instructions, to name a few.5, 11, 39 The higher frequency of occult metastases in blacks compared with whites might simply reflect this diagnostic lag. However, most of the factors contributing to delays in detection and diagnosis are related to socioeconomic status,5-8, 11, 39 which appears to provide only a partial contribution to stage-specific disparities in outcomes.8, 11, 12 Moreover, a temporal relation between delays in diagnosis and the incidence of nodal micrometastases, beyond standard staging, has not been observed.
The partial contribution of socioeconomic status to stage-specific racial disparities8, 11, 12 and the unlikely contribution of mis-staging or undertreatment in the current study raises the possibility that biology might contribute to differences in occult tumor burden in blacks compared with whites. Biology may, in part, underlie differences in clinical presentation, for example, the earlier onset, greater undifferentiated histology, and right-sided predominance of tumors in black compared with white patients.5, 11, 40 In turn, clinical differences may reflect molecular disparities in DNA damage sensing and repair pathways underlying microsatellite, chromosomal, and epigenetic stability in colorectal tumors from blacks compared with whites.41-44 Also, race-related biological differences may be acquired, instead of inherited, reflecting lifestyle, rather than genetic, disparities.5-8, 11 For example, rates of obesity are higher in the blacks compared with whites, which is associated with an increased risk of developing and dying from colon cancer.11 Conversely, it is unlikely that an increase in inherited germline mutations underlies the escalating racial disparity in stage-specific outcomes, which has expanded over the past 4 decades.5 Moreover, the greater genetic variation within, rather than between, races makes a genetic basis for racial disparities in outcomes unlikely.45, 46 The biologic contribution to racial disparities in outcomes in colorectal cancer will continue to be debated.5, 11, 40-44 However, the present observations associate increased occult tumor burden with poorer outcomes in blacks compared with whites. They suggest future studies exploring the relation between occult metastases and molecular determinants of tumor aggressiveness, including specific cancer-related gene mutations, microsatellite instability, chromosomal instability, and DNA methylation.47
The current study demonstrates that stage-specific racial disparities in outcomes in pN0 blacks compared with whites with colorectal cancer is associated with greater occult tumor burden in regional lymph nodes. These results demonstrate, for the first time to our knowledge, the impact of occult tumor burden on racial disparities in clinical prognosis. They highlight the importance of validating this emerging approach to identify colorectal cancer patients of all races at increased prognostic risk. The current study was limited to a cohort comprising approximately 8% African Americans, which is representative of the colon cancer population but relatively small, demonstrating the need for validation. That validation will be provided by an ongoing study with equally balanced black and white colon cancer patients, which will be critical to understanding the utility of GUCY2C to reduce racial disparities in colorectal cancer outcomes. Moreover, the presence of tumor cells in regional lymph nodes directs therapy in patients with colon cancer. Although adjuvant chemotherapy provides a survival benefit to patients with stage III disease, its utility in patients with pN0 colon cancer remains uncertain, with marginal survival benefits in stage II patients in some, but not all, clinical trials.3, 48, 49 Heterogeneous responses to therapy in pN0 patients may reflect, in part, the variable presence of occult metastases. It is tempting to speculate that black and white pN0 patients with excess occult tumor burden might benefit from adjuvant therapy. These considerations underscore the importance of refining the predictive utility of occult tumor burden to eliminate the widening5 racial gap in stage-specific mortality in colorectal cancer patients.
This study was supported by grants from the National Institutes of Health (CA75123, CA95026, CA112147, CA146033), the Pennsylvania Department of Health, and Targeted Diagnostic and Therapeutics Inc. S.A.W. is the Samuel M.V. Hamilton Endowed Professor of Thomas Jefferson University.
CONFLICT OF INTEREST DISCLOSURES
S.A.W. is the Chair of the Data Safety Monitoring Board for the C-Cure Trial™ sponsored by Cardio Biosciences, and the Chair (uncompensated) of the Scientific Advisory Board to Targeted Diagnostics and Therapeutics, Inc., which provided research funding that, in part, supported this work and has a license to commercialize inventions related to this work.
- 3AJCC cancer staging manual. 7th ed. New York, NY: Springer; 2009., , , , , .
- 11Racial disparities in colorectal cancer survival: to what extent are racial disparities explained by differences in treatment, tumor characteristics, or hospital characteristics? Cancer. 2010; 116: 4622-4631., , , .
- 28Evaluation of guanylyl cyclase C lymph node status for colon cancer staging and prognosis [published online ahead of print May 1, 2011]. Ann Surg Oncol. PMID: 21533822., , , et al.
- 31An introduction to recursive partitioning using the rpart routines. Technical report 61. Rochester, MN: Mayo Clinic; 1997., .