SupportInstitute

Accepte

ReprintFletcherlington,

0039-60

� 2010

doi:10.1

358 S

Staging error does not explain therelationship between the number oflymph nodes in a colon cancerspecimen and survivalJesse Moore, MD,a Neil Hyman, MD,a Peter Callas, PhD,b and Benjamin Littenberg, MD,c Burlington, VT

Background. Survival in colon cancer is greater in those patients who have more lymph nodes identifiedat resection and may be due to stage migration, confounding by treatment, social, or clinicalcharacteristics. Identifying factor(s) responsible for the effect may represent an opportunity to improvequality of care for patients with colon cancer by increasing node counts in specimens.Methods. Cox proportional hazards models were created to analyze survival of 11,399 patients withstage I-III colon cancer from the Surveillance, Epidemiology and End Results (SEER)-Medicaredatabase. The primary predictor variable was the number of lymph nodes identified. The models allowedadjustment for patient factors, use of chemotherapy, surgical specialty, and the average number of nodesidentified by surgeon and hospital pathologist.Results. The number of nodes identified was related to survival. Compared to those with less than 7 nodes,patients with 7 to 11 nodes had a 13% lesser risk of death (hazard ratio [HR], 0.87; 95% confidenceinterval [CI], 0.76--0.99; P = .037). Patients with more than 12 nodes had a 17% lesser risk (HR, 0.83;95% CI, 0.73--0.95; P = .005). Adjusting for selected patient demographic characteristics, receipt ofchemotherapy, surgical specialty, and the average number of nodes identified per specimen by the surgeon orhospital did not significantly alter the relationship between number of nodes and survival.Conclusion. These findings argue against understaging or confounding as the explanation for theinferior survival observed in patients with fewer nodes identified. National initiatives to increase thenumber of nodes identified in colon cancer specimens may not improve substantially the cancer-specificoutcomes. (Surgery 2010;147:358-65.)

From the Department of Surgery,a Department of Biostatistics,b and General Internal Medicine,c University ofVermont College of Medicine, Burlington, VT

NUMEROUS STUDIES have appeared in the recent liter-ature documenting an impressive survival advan-tage for patients who have more nodes identifiedin their pathologic specimen after colectomy forcolon cancer. In stages I, II, and III colon cancer,there is a 21% decrease in mortality when 15 ormore nodes are analyzed compared to when 1 to7 nodes are examined.1 A secondary analysis of alarge adjuvant chemotherapy trial for stage II andstage III colon cancer demonstrated that for bothnode-positive and node-negative patients, survival

ed in part by research grants from the Nationals of Health (K30 RR22260 and K24 DK68380).

d for publication October 2, 2009.

requests: Neil Hyman, MD, Department of Surgery,464, University of Vermont College of Medicine, Bur-VT 05401. E-mail: Neil.Hyman@vtmednet.org.

60/$ - see front matter

Mosby, Inc. All rights reserved.

016/j.surg.2009.10.003

URGERY

improved as the number of nodes analyzed in-creased.2 In patients with stage IIIB and stageIIIC disease, a greater number of negative nodesis an independent predictor of better outcome.3

These findings have generally been attributedto understaging due to suboptimal cancer surgeryor inadequate pathologic examination. Patientswith a lesser number of nodes analyzed may bedesignated falsely as stage I or stage II, becausenone of the examined nodes harbors malignancy.If more nodes are examined, the chance of missinga positive node may be decreased, allowing thepatient to be staged properly and receive adjuvanttherapy when appropriate. This concern has led torecommendations for a minimum of 12 nodes tobe analyzed in every colon cancer specimen.4-6 Inthis context, the number of nodes in a colon can-cer specimen has been proposed as a measure ofquality of colon cancer care.7

It may be, however, that the number of nodesidentified in the colectomy specimen is actually an

SurgeryVolume 147, Number 3

Moore et al 359

indicator of patient-specific factors such as a morerobust immunologic response to the cancer orother tumor-related characteristics not reflected bystage.8-11 These factors would not be amenablenecessarily to process improvements in pathologicor surgical care, and therefore node count may notbe a meaningful quality measure.

A recent study by Wong et al12 found that pa-tients treated in hospitals at which a greater num-ber of nodes were examined per specimen did nothave a greater survival rate after colectomy. Al-though hospitals differed greatly in the numberof nodes examined, they were equally likely tofind node-positive neoplasms. These data refutestage migration as an important factor explainingthe relationship between greater node count andimproved oncologic outcome; however, this studyby Wong et al12 did not include the surgeon per-forming the operation as a variable, leaving extentof resection as a possible explanation for the ob-served survival benefit associated with greaternode harvest.

We sought to examine the relationship betweenthe number of nodes analyzed and survival whilecontrolling for the extent of resection and thepathologic processing of the specimen. We hypoth-esized that the positive relationship between thenumber of nodes analyzed and survival wouldpersist even after controlling for the demographiccharacteristics of the patient, the operating sur-geon, and the pathologic search for nodes. If so,the relationship between number of nodes andsurvival may be a surrogate for other prognosticfactors and not primarily a reflection of subopti-mal operative resection or pathologic evaluation.

METHODS

This study was approved by the University ofVermont Committees on Human Research.

Data sources. The Surveillance, Epidemiology,and End Results (SEER) database is a population-based registry maintained by the National CancerInstitute (Bethesda, MD). The SEER programcollects data on cancer incidence, patient charac-teristics, and survival from 18 population-basedregistries comprising more than 25% of the U.S.population. Information on each case includestumor site, cancer stage, number of nodes, andpatient demographics.13 Survival data are currentthrough December of 2003.14

Medicare provides health insurance to disabledand senior patients including 97% of the U.S.population over the age of 65 years. To providedetailed population-based information on elderlypatients with cancer, the SEER-Medicare linked

database was created. A total of 93% of patients inthe SEER database 65 years of age and older havebeen matched to their Medicare files.3,15

Information about in-patient claims is includedin the Medicare Provider Analysis and Review files(MEDPAR). Information on claims for outpatientcare, including cancer chemotherapy, is includedin the physician/supplier bills (known as theNational Claims History [NCH]) and the outpa-tient file. SEER files are linked to the MEDPAR,NCH, and the outpatient files and are updated ev-ery 3 years. Unique identifiers for the treating sur-geon and hospital are located in the NCH andMEDPAR files.

We considered the mean number of nodesreported per surgeon as a marker of that surgeon’stypical extent of resection. Because there are nounique identifiers for pathologists in the SEER-Medicare database, we used the hospital as asurrogate indicator for the typical diligence ofthe pathologic processing at an institution.

Patients. We identified 34,762 patients who werediagnosed with colon cancer between January 1994and December 1998, had a complete date ofdiagnosis, and were enrolled in Medicare Parts Aand B at the time of diagnosis. Patients in HealthMaintenance Organizations were excluded be-cause detailed claims were not submitted to Medi-care for these patients. Patients undergoing openor laparoscopic partial colectomy or hemicolec-tomy for colon cancer were identified usingcommon procedural terminology (CPT) codes(44140-44146, 44160, or 44204-44208) and Inter-national Classification of Diseases, 9th Revision,Clinical Modification (ICD9-CM) codes (153.x or154.0). The MEDPAR, NCH, and outpatient fileswere searched for chemotherapy claims (ICD-9-CM diagnostic codes V581, V662, V672, and9925) and Healthcare Common Procedure CodingSystem (HCPCS) codes for fluorouracil (J9190),leucovorin (J0640), irinotecan (J9206), and floxur-idine (J9200) associated with the ICD-9-CM codesfor colon cancer. If a surgeon operated at morethan 1 hospital, only cases from their primary hos-pital (defined as the hospital where the most casesoccurred) were included.

In an effort to exclude factors independent ofthe surgeon or the hospital that may have alteredthe number of nodes in a specimen, we excludedthose patients who had synchronous or metachro-nous lesions (n = 2,302; 7%) or who underwent op-erations other than partial colectomy orhemicolectomy (n = 4,376; 13%). Patients withstage 0 or stage IV disease (n = 3,771; 11%) wereexcluded, because the very early or late nature of

SurgeryMarch 2010

360 Moore et al

the disease may have been readily apparent at op-eration and could have affected a surgeon’s ten-dency toward lesser or palliative resections.Patients not operated on at a surgeon’s primaryhospital (n = 962) were excluded. Patient recordsmissing the identifier of the operating surgeon(n = 4,321) or the hospital where the operationtook place (n = 1,009), with an unknown numberof nodes examined (n = 971), or with unknownor incorrect staging (n = 1,333) were excluded.There were 4317 patients (12%) treated nonoper-atively who were excluded. In addition, 1 patientwith a recorded date of death occurring prior tothe date of diagnosis was excluded.

Variables. The main independent or predictorvariable---the number of nodes examined---was di-vided into 3 groups (<7 nodes, 7--11 nodes, and$12 nodes).1,16 The sex of each patient was re-corded, and age was recorded to the nearest year.Patient race was dichotomized to white and non-white. Income was estimated as the median 1990per capita income for each patient’s residentialcensus tract. The use of colon cancer--directed che-motherapy after the index operation was alsorecorded. Surgical specialty of the operatingphysician was categorized as general surgery, colo-rectal surgery, or ‘‘other’’ (surgical oncology, ob-stetrics/gynecology, urology, thoracic surgery,vascular surgery, or cardiac surgery). The meannumber of nodes per specimen was calculatedfor each hospital and for each surgeon.

For each patient, 2 outcomes were analyzed:(1) overall survival, which was represented as thetime from date of diagnosis to date of death fromany cause; and (2) disease-specific survival, whichwas defined as time from date of diagnosis to date ofdeath from colorectal cancer with censoring on thedate they died of another cause. For both outcomes,patients who had not suffered the outcome werecensored on the date the dataset was closed(December 31, 2003). The cause of death wasunknown for 125 patients. These patients wereincluded in the overall survival analyses but wereexcluded from the disease-specific survival analyses.

Statistical approach. We described continuousvariables using the mean (± standard deviation),median and range, and dichotomous variablesusing percentages. Subgroups were comparedwith either Chi-square tests or analysis of variance(ANOVA). All analyses were 2-tailed. We definedfindings with P < .05 to be statistically significant.Survival analyses were performed using theKaplan-Meier method.

We created 3 Cox proportional hazards models.Model A was unadjusted with the number of nodes

(categorized as <7, 7--11, or $12) as the independentvariable. Model B included the number of nodes and5 patient characteristics (age, sex, race, per capitaincome by census tract, and receipt of chemother-apy). Model C included the number of nodes and thepatient factors as well as the surgical specialty, thesurgeon’s mean number of nodes per specimen, andthe hospital’s mean number of nodes per specimen.Because patients cared for by the same surgeon maynot be completely independent of one another, weadjusted the variances by clustering on surgeon.17

We clustered on surgeon rather than hospital,because all surgeons operated within a single hospitalin this data set and because the intraclass correlationof survival within surgeons (0.024) was greater thanthat within hospital (0.011).

We ran each model for both overall and disease-specific survival on the total population and thenstratified based on the American Joint Commissionon Cancer (AJCC) tumor-node-metastasis (TNM)staging system. We designated Model C of overallsurvival as the primary analysis and the others assecondary or exploratory analyses to confirm theconsistency of the findings and search forexceptions.

If the hazard ratios associated with number ofnodes observed in Model A are attenuated by theinclusion of the patient descriptors in Model B, itsuggests the effect of number of nodes on survivalis confounded by demographic or treatment fac-tors. If the hazard ratios seen in Model B arediminished in Model C (controlling for the extentof the operative and pathologic search for nodes),it would suggest that the previously observed rela-tionship between number of nodes and survival isan artifact of understaging or other treatment-related factors. If the hazard ratios are similaracross all 3 models, then the effect of the numberof nodes on survival may be due to unmeasuredcharacteristics not present in the models.

RESULTS

The demographic and clinical characteristics ofthe 11,399 patients who met the selection criteriaare shown in Table I. Operations were performedby 2101 surgeons at 488 hospitals. General sur-geons made up the largest proportion (85.2%),followed by colorectal surgeons (11.1%), andothers (3.7%). Mean follow-up time was 7.5 years(range, 1 day--10 years) during which there were6,564 deaths (57.6%). The mean number of nodesexamined was 11.6; the median was 10, with arange from 0 to 78. The mean number of nodesidentified by surgeon was 11.6 ± 4.8. The meannumber of nodes identified by hospital was 11.6

Table I. Demographic, treatment, and tumor characteristics

Characteristic All patients Stage I Stage II Stage III P value

N 11,399 2,206 5,436 3,757Age, y 77.8 ± 7.6 77.4 ± 7.3 78.2 ± 7.6 77.4 ± 7.8 <.001Female 6,430 (56.4) 1,200 (54.4) 3,054 (56.2) 2,176 (57.9) .009Race, white 9,974 (87.5) 1,948 (88.3) 4,824 (88.7) 3,202 (85.2) .009Income* 17.8 ± 9.9 17.9 ± 10.0 17.9 ± 9.9 17.6 ± 9.7 .33Chemotherapy 3,306 (29.0) 102 (4.6) 1,113 (20.5) 2,091 (55.6) .007Nodes examined 11.6 ± 8.1 9.8 ± 8.0 11.7 ± 8.2 12.3 ± 7.9 <.001

<7 3,328 (29.2) 909 (41.2) 1,538 (28.3) 881 (23.5) .008y7--11 3,419 (30.0) 643 (29.2) 1,610 (29.6) 1,166 (31.0) .008y$12 4,652 (40.8) 654 (29.7) 2,288 (42.1) 1,710 (45.5) .008y

Values presented as n (%) or mean ± SD.*1990 census tract per capita income in thousands of dollars.yChi-square analysis of the proportion in each node category by stage.

SurgeryVolume 147, Number 3

Moore et al 361

± 3.8. Stage II made up the largest proportion ofpatients (48%). Stage III patients were the mostlikely to have 12 or more nodes examined, whereasstage I patients were the most likely to have fewerthan 7 nodes examined (P = .008). The Kaplan-Me-ier estimated overall survival at 5 years was 54.7%.Across all stages, as the number of nodes in-creased, there was significant improvement in over-all survival at 5 years (Fig 1). Evaluating each stageseparately showed this trend to be true for eachstage (data not shown).

Disease-specific survival at 5 years was 85.7%.Colon cancer was the cause of death in 7.6% ofstage I patients, 16.0% of stage II patients, and35% of stage III patients. When all stages wereincluded, there was a significant improvement indisease-specific survival as the number of nodesfound was increased (Fig 2). When each stage wasevaluated separately, there was significant improve-ment in disease-specific survival for patients withstage II and stage III disease with greater nodecounts (Figs 3 and 4).

Overall survival was associated with the numberof nodes in Cox proportional hazards analyses(Table II, Model A). Those patients with 7 to 11 no-des identified had an 11% less hazard than thosewith fewer nodes (unadjusted hazard ratio [HR],0.89; 95% confidence interval [CI], 0.84--0.95; P< 0.001) and those with at least 12 nodes hadeven lesser risk (unadjusted HR, 0.80; 95% CI,0.75--0.85; P < .001). While controlling for patientfactors (Model B), patients with 7 to 11 nodes ex-amined had a 12% lesser risk of overall mortalitycompared to patients with fewer than 7 nodes (ad-justed HR, 0.88; 95% CI, 0.83--0.94; P < .001). Pa-tients with 12 or more nodes examined had an18% (adjusted HR, 0.82; 95% CI, 0.77--0.87; P <.001) lesser risk of all-cause mortality. When

surgeon and hospital factors were added (ModelC), the change in hazard ratio for patients with 7to 11 nodes was minimal (adjusted HR, 0.89;95% CI, 0.84--0.95; P < .001). The change was sim-ilarly negligible in patients with 12 or more nodes(adjusted HR, 0.85; 95% CI, 0.79--0.90; P < .001).

Analyses by AJCC stage found similar effects instages I, II, and III (Table II). In stage I patients,the number of nodes examined was significantlyassociated with survival only in those patientswho had 12 or more nodes examined. We foundno difference in patient’s overall survival betweencolorectal and general surgeons (data not shown).Patients operated on by other surgeons had a sig-nificantly worse overall survival than those oper-ated on by a general surgeon (adjusted HR, 1.23;95% CI, 1.05--1.45, P = .011).

Disease-specific survival was also associated withthe number of nodes identified (Table III, ModelA). There was a 13% decrease in mortality with7 to 11 nodes (unadjusted HR, 0.87; 95% CI,0.76--0.99; P = .037) and 17% with 12 or morenodes (unadjusted HR, 0.83; 95% CI, 0.73--0.95;P = .005). While controlling for patient factors(Model B), the risk of colon cancer death was16% less in patients with 7 to 11 nodes examinedcompared to patients with fewer than 7 nodes ex-amined (adjusted HR, 0.84; 95% CI, 0.74--0.96;P = .010). Patients with 12 or more nodes exam-ined had a 20% less risk of colon cancer deathcompared to those with fewer than 7 nodes exam-ined (adjusted HR, 0.80; 95% CI, 0.71--0.91;P = .001). Adding the operative and hospital fac-tors had minimal effect on the hazard ratio in pa-tients with 7 to 11 nodes (adjusted HR, 0.85; 95%CI, 0.75--0.97; P = .019) or patients with more than12 nodes (adjusted HR, 0.84; 95% CI, 0.73-- 0.97;P = .019).

Fig 1. Kaplan-Meier estimates of overall survival, allstages. The 5-year overall survival for patients with fewerthan 7 nodes (dashed line) was 50.8%; for patients with7--11 nodes (solid line), it was 54.3%; and for patientswith 12 or more nodes (dotted line), it was 57.9%(P < .001 by log-rank test).

Fig 2. Kaplan-Meier estimates of disease-specific survival,all stages. The 5-year disease-specific survival for patientswith fewer than 7 nodes (dashed line) was 84.2%; for pa-tients with 7--11 nodes (solid line), it was 86.0%; and forpatients with 12 or more nodes (dotted line), it was86.5% (P = .011 by log-rank test).

Fig 3. Kaplan-Meier estimates of disease-specific survival,patients with stage II disease. The 5-year disease-specificsurvival for patients with fewer than 7 nodes (dashedline): 86.5%; for patients with 7--11 nodes (solid line), itwas 90.5%; and for patients with 12 or more nodes (dot-ted line), it was 92.2% (P < .001 by log-rank test).

Fig 4. Kaplan-Meier estimates of disease-specific survival,patients with stage III disease. The 5-year disease-specificsurvival for patients with fewer than 7 nodes (dashed line)was 66.0%; for patients with 7--11 nodes (solid line), it was73.7%; and for patients with 12 or more nodes (dottedline), it was 74.2% (P < .001 by log-rank test).

SurgeryMarch 2010

362 Moore et al

Analyses by AJCC stage showed little change tothe model when surgeon and hospital factors wereadded (Table III). The number of nodes examinedwas not significantly associated with disease-specificsurvival for stage I patients; however, the numberof nodes was significantly associated with disease-specific survival in stage II and stage III patients.Surgeon type was not predictive of disease-specificsurvival (data not shown).

In both the overall survival analyses and thedisease-specific analyses for all stages, the hazardratios for Models B and C were similar to those of

the corresponding Model A, never differing bymore than 6%. In other words, adjustment forneither patient factors nor surgeon and hospitalfactors significantly affected the relationship of thenumber of nodes to survival.

DISCUSSION

In the present study, controlling for patient age,sex, race, per capita income by census tract, receiptof chemotherapy, surgeon specialty, and the aver-age number of nodes identified per specimen bythe surgeon and hospital did not affect the

Table II. Results of Cox proportional hazards models for overall survival

Model ByNodes examined Model A* Hazard ratio (95% CI) Model Cz

All stages (n = 11,399, 6,564 deaths)<7 (referent) 1.0 1.0 1.07--11 0.89 (0.84–0.95) 0.88 (0.83–0.94) 0.89 (0.84–0.95)$12 0.80 (0.75–0.85) 0.82 (0.77–0.87) 0.85 (0.79–0.90)

Stage I (n = 2,206, 1,045 deaths)<7 (referent) 1.0 1.0 1.07--11 0.89 (0.77–1.03) 0.88 (0.79–1.02) 0.89 (0.77–1.03)$12 0.77 (0.66–0.89) 0.79 (0.68–0.91) 0.81 (0.69-0.96)

Stage II (n = 5,436, 2,983 deaths)<7 (referent) 1.0 1.0 1.07--11 0.80 (0.73–0.87) 0.81 (0.74–0.89) 0.81 (0.73–0.88)$12 0.67 (0.61–0.73) 0.70 (0.64–0.77) 0.70 (0.63–0.77)

Stage III (n = 3,757, 2,536 deaths)<7 (referent) 1.0 1.0 1.07--11 0.87 (0.79–0.97) 0.86 (0.78–0.96) 0.88 (0.79–0.97)$12 0.83 (0.75–0.91) 0.82 (0.74–0.91) 0.86 (0.77–0.98)

*Model A: unadjusted.yModel B: adjusted for patient factors: sex, age, race, per capita income by census tract, and receipt of chemotherapy.zModel C: adjusted for patient factors as well as surgeon type, surgeon’s mean nodes per specimen, and hospital’s mean nodes per specimen.

CI, Confidence interval.

SurgeryVolume 147, Number 3

Moore et al 363

relationship between the number of nodes exam-ined in a specimen and overall or disease-specificsurvival.

The number of nodes examined from a coloncancer specimen had been found previously to bepredictive of overall and disease-specific survivalfor patients with stages I, II, and III disease. In asingle-center study, patients with fewer than 7nodes had a 5-year overall survival of 49% com-pared to 68% for patients with more than 7nodes.16 These findings have subsequently beenconfirmed in population studies. An examinationof the National Cancer Database found that pa-tients with 1--7 nodes had a 5-year overall survivalof 50% compared to 56% for patients with 8--12nodes and 63% for 13 or more nodes.18 Our studyconfirms this basic relationship between the num-ber of nodes and survival. Other studies have usedmultivariate statistics to control for both patientand hospital factors,2,18-20 but none, to our knowl-edge, have included the surgeon as a factor.

A frequently offered explanation for the differ-ences in survival based on the number of nodesexamined is understaging. According to this the-ory, a low number of nodes in a specimen is areflection of inadequate resection or pathologicprocessing. This inadequate resection and analysismay lead to missing positive lymph nodes inadver-tently and incorrect staging of a patient. Ourmodels adjusted for demographic, social, andeconomic factors at the patient level, as well as

the extent of the resection and the pathologicprocessing. Controlling for the factors that havebeen proposed to be causing understaging had noeffect on the relationship between nodes andsurvival. If understaging explains the relationshipbetween number of nodes and survival, then con-trolling for the responsible factors should haveattenuated the hazard ratios from model to model.

Because the observed relationship between num-ber of nodes identified and survival was not alteredsubstantially by the factors studied here, it is unlikelythat understaging is playing a primary role. Someother factors could be responsible. It may be thattumor subtypes that metastasize early also tend tosuppress lymph node development. Alternatively,finding more nodes may signify that the patient hasmustered a more robust cellular immune responseto the cancer. Indeed, patients with increased levelsof histiocytes in the lymph nodes, increased localinflammation, and increased T- and B-cell prolifer-ation have been shown to have a superior 5-yearsurvival and are more likely to have well-differenti-ated neoplasms.8-10

Patients with lymphoid hyperplasia survive sig-nificantly longer than patients without lymphoidhyperplasia,21 whereas patients whose histologicpattern is consistent with lymphocyte depletionor an unstimulated immune response have 5-yearoverall survival of 10% and 17%, respectively, com-pared to patients with either lymphocyte or germi-nal center predominance (68% and 53% 5-year

Table III. Results of Cox proportional hazards models for cancer survival

Model By*Nodes examined Model A* Hazard ratio (95% CI) Model Cz

All stages (n = 11,274, 1,446 deaths)<7 (referent) 1.0 1.0 1.07--11 0.87 (0.76–0.99) 0.84 (0.74–0.96) 0.85 (0.75–0.97)$12 0.83 (0.73–0.95) 0.80 (0.71–0.91) 0.84 (0.73–0.97)

Stage I (n = 2,185, 79 deaths)<7 (referent) 1.0 1.0 1.07--11 1.16 (0.68–1.98) 1.22 (0.72–2.05) 1.21 (0.7–2.09)$12 0.95 (0.55–1.65) 0.97 (0.56–1.71) 0.97 (0.47–2.0)

Stage II (n = 5,374, 478 deaths)<7 (referent) 1.0 1.0 1.07--11 0.68 (.54–0.85) 0.68 (0.54–0.85) 0.68 (0.55–0.86)$12 0.55 (0.45–0.69) 0.57 (0.46–0.71) 0.57 (0.44–0.74)

Stage III (n = 3,715, 889 deaths)<7 (referent) 1.0 1.0 1.07--11 0.75 (0.63–0.89) 0.75 (0.63–0.90) 0.76 (0.64–0.91)$12 0.73 (0.62–0.85) 0.73 (0.63–0.86) 0.77 (0.64–0.94)

*Model A: unadjusted.yModel B: adjusted for patient factors: sex, age, race, per capita income by census tract, and receipt of chemotherapy.zModel C: adjusted for patient factors as well as surgeon type, surgeon’s mean nodes per specimen, and hospital’s mean nodes per specimen.

CI, Confidence interval.

SurgeryMarch 2010

364 Moore et al

overall survival, respectively).22 Halvorsen andSeim23 showed a worse stage distribution andworse 5-year cancer-specific survival in patientswith less inflammatory reaction to their colorectalcancer. Neoplasms lacking signs of early metastaticinvasion had greater levels of immune cells andeffector memory T cells. These patients had morefavorable staging and increased survival.24,25

We are not advocating for inadequate oncologicresections or less than a meticulous pathologicsearch for lymph nodes. Rather, a low number ofnodes in a specimen may not necessarily representthe quality of care a patient has received andinstead may be a marker of tumor biology orpatient factors. Patients with a low number ofidentified lymph nodes may be harboring a partic-ularly virulent cancer or may have a suboptimalcancer-specific immunologic response. A low no-dal harvest may be important independently toprognosis, have implications for therapy, and con-tribute to our understanding of the biology ofcolon cancer. Further studies are needed to ex-plain the survival advantage observed in patientswith more lymph nodes identified in their colec-tomy specimen.

There are limitations to this study. The SEER-Medicare database lacks information regarding theidentity of the pathologist performing the patho-logic exam. Even if this information were available,key components of the pathologic processing(gross inspection and removal of nodes) are

performed frequently by assistants or residents.We used the hospital identifier as a proxy for thepathologic process. We feel this is robust, becauseit is likely a hospital employs a similar techniquefor each specimen. In a similar fashion, the hospi-tal setting in which the colectomy took place, ie,teaching or nonteaching, may have affected thelevel of involvement of assistants or residentsduring an operation. Other factors that couldhave affected the number of nodes in a specimeninclude location of the neoplasm, a patient’s bodymass index, and comorbidities. We did not adjustfor these factors in the present study.

Of our original cohort, 15% lacked either aunique identifier for the surgeon or the hospital.These patients were excluded from our analyses.Similar studies using SEER-Medicare data to eval-uate outcomes after operation have encountered8--14% of patient’s claims missing 1 or bothidentifiers.26-28 We also excluded patients if the op-eration occurred at a hospital other than the sur-geon’s primary hospital. Some of these patientsare likely to have been operated on by surgeonstaking night call and covering several hospitals.These patients may be more likely to require ur-gent laparotomy and bowel resection rather thana planned, elective colectomy guided by oncologicprinciples. Had these patient’s data been available,our conclusions may have differed. Our cohort istherefore a narrow sample of those available inthe SEER-Medicare database. Additionally the

SurgeryVolume 147, Number 3

Moore et al 365

SEER-Medicare database is limited largelyto patients over 65 years of age, which may limitthe applicability of our conclusions to youngerpopulations.

In conclusion, we have shown that, in a subset ofthe SEER-Medicare database patients, the numberof nodes in a colon cancer specimen remainspredictive of survival while controlling for selectedpatient demographic data, receipt of chemother-apy, the type of surgeon, a surgeon’s mean numberof nodes, and the hospital’s mean number ofnodes. Therefore, the relationship between thenumber of nodes and survival may not be dueprimarily to understaging. The literature supportsthe hypothesis that an interaction between thecancer and the patient’s immune response may beplaying a greater role in the relationship then hasbeen recognized in the past. Studies at the basicscience level are needed to explore this relation-ship further.

REFERENCES

1. Chen SL, Bilchik AJ. More extensive nodal dissection im-proves survival for stages I to III of colon cancer: a popula-tion-based study. Ann Surg 2006;244:602-10.

2. Le Voyer TE, Sigurdson ER, Hanlon AL, et al. Colon cancersurvival is associated with increasing number of lymph no-des analyzed: a secondary survey of intergroup trial INT-0089. J Clin Oncol 2003;21:2912-9.

3. Johnson PM, Porter GA, Ricciardi R, Baxter NN. Increasingnegative lymph node count is independently associatedwith improved long-term survival in stage IIIB and IIICcolon cancer. J Clin Oncol 2006;24:3570-5.

4. Nelson H, Petrelli N, Carlin A, et al. Guidelines 2000 for co-lon and rectal cancer surgery. J Natl Cancer Inst 2001;93:583-96.

5. Fielding LP, Arsenault PA, Chapuis PH, et al. Clinicopatho-logical staging for colorectal cancer: an InternationalDocumentation System (IDS) and an International Com-prehensive Anatomical Terminology (ICAT). J Gastroen-terol Hepatol 1991;6:325-44.

6. Otchy D, Hyman NH, Simmang C, et al. Practice parame-ters for colon cancer. Dis Colon Rectum 2004;47:1269-84.

7. Chang GJ, Rodriguez-Bigas MA, Skibber JM, Moyer VA.Lymph node evaluation and survival after curative resectionof colon cancer: systematic review. J Natl Cancer Inst 2007;99:433-41.

8. Pihl E, Nairn RC, Nind AP, et al. Correlation of regionallymph node in vitro antitumor immunoreactivity histol-ogy with colorectal carcinoma. Cancer Res 1976;36:3665-71.

9. Fisher ER, Saffer E, Fisher B. Studies concerning the re-gional lymph node in cancer. VI. Correlation of lymphocytetransformation of regional node cells and some histopatho-logic discriminants. Cancer 1973;32:104-11.

10. Patt DJ, Brynes RK, Vardiman JW, Coppleson LW. Mesocoliclymph node histology is an important prognostic indicator

for patients with carcinoma of the sigmoid colon: an immu-nomorphologic study. Cancer 1975;35:1388-96.

11. Murray D, Hreno A, Dutton J, Hampson LG. Prognosis incolon cancer: a pathologic reassessment. Arch Surg 1975;110:908-13.

12. Wong SL, Ji H, Hollenbeck BK, et al. Hospital lymph nodeexamination rates and survival after resection for coloncancer. JAMA 2007;298:2149-54.

13. National Cancer Institute. Surveillance, Epidemiology, andEnd Results (SEER) program. Available at http://seer.cancer.gov/manuals/2007/SPCM_2007maindoc.pdf. Accessed onMay 31, 2007.

14. National Cancer Institute. Surveillance, Epidemiology, andEnd Results. Available at: http://seer.cancer.gov. Accessedon May 31, 2007.

15. Schrag D, Cramer LD, Bach PB, et al. Influence of hospitalprocedure volume on outcomes following surgery for coloncancer. JAMA 2000;284:3028-35.

16. Caplin S, Cerottini JP, Bosman FT, et al. For patients withDukes’ B (TNM Stage II) colorectal carcinoma, examina-tion of six or fewer lymph nodes is related to poor progno-sis. Cancer 1998;83:666-72.

17. Wears RL. Advanced statistics: statistical methods for analyz-ing cluster and cluster-randomized data. Acad Emerg Med2002;9:330-41.

18. Swanson RS, Compton CC, Stewart AK, Bland KI. The prog-nosis of T3N0 colon cancer is dependent on the number oflymph nodes examined. Ann Surg Oncol 2003;10:65-71.

19. Prandi M, Lionetto R, Bini A, et al. Prognostic evaluation ofstage B colon cancer patients is improved by an adequatelymphadenectomy: results of a secondary analysis of a largescale adjuvant trial. Ann Surg 2002;235:458-63.

20. Bui L, Rempel E, Reeson D, Simunovic M. Lymph nodecounts, rates of positive lymph nodes, and patient survivalfor colon cancer surgery in Ontario, Canada: a popula-tion-based study. J Surg Oncol 2006;93:439-45.

21. Pihl E, Nairn RC, Milne BJ, et al. Lymphoid hyperplasia: amajor prognostic feature in 519 cases of colorectal carci-noma. Am J Pathol 1980;100:469-80.

22. Nacopoulou L, Azaris P, Papacharalampous N, Davaris P.Prognostic significance of histologic host response in can-cer of the large bowel. Cancer 1981;47:930-6.

23. Halvorsen TB, Seim E. Association between invasiveness, in-flammatory reaction, desmoplasia and survival in colorectalcancer. J Clin Pathol 1989;42:162-6.

24. Pages F, Berger A, Camus M, et al. Effector memory T cells,early metastasis, and survival in colorectal cancer. N Engl JMed 2005;353:2654-66.

25. Galon J, Costes A, Sanchez-Cabo F, et al. Type, density, andlocation of immune cells within human colorectal tumorspredict clinical outcome. Science 2006;313:1960-4.

26. Gilligan MA, Neuner J, Zhang X, et al. Relationship be-tween number of breast cancer operations performed and5-year survival after treatment for early-stage breast cancer.Am J Public Health 2007;97:539-44.

27. Schrag D, Earle C, Xu F, et al. Associations between hospitaland surgeon procedure volumes and patient outcomes afterovarian cancer resection. J Natl Cancer Inst 2006;98:163-71.

28. Schrag D, Panageas KS, Riedel E, et al. Surgeon volumecompared to hospital volume as a predictor of outcome fol-lowing primary colon cancer resection. J Surg Oncol 2003;83:68-78; discussion, 78-9.