Lymph node evaluation for colon cancer in routine clinical practice: a population-based study

Background Guidelines recommend that 12 or more lymph nodes ( ln s) be evaluated during surgical resection of colon cancer. Here, we report ln yield and its association with survival in routine practice. Methods Electronic records of treatment were linked to the population-based Ontario Cancer Registry to identify all patients with colon cancer treated during 2002–2008. The study population ( n = 5508) included a 25% random sample of patients with stage ii or iii disease. Modified Poisson regression was used to identify factors associated with ln yield; Cox models were used to explore the association between ln yield and overall ( os ) and cancer-specific survival ( css ). During 2002–2008, median ln yield increased to 17 from 11 nodes ( p < 0.001), and the proportion of patients with 12 or more nodes evaluated increased to 86% from 45% ( p < 0.001). Lymph node positivity did not change over time (to 53% from 54%, p = 0.357). Greater ln yield was associated with younger age ( p < 0.001), less comorbidity ( p = 0.004), higher socioeconomic status ( p = 0.001), right-sided tumours ( p < 0.001), and higher hospital volume ( p < 0.001). In adjusted analyses, a ln yield of less than 12 nodes was associated with inferior os and css for stages ii and iii disease [stage ii os hazard ratio ( hr ): 1.36; 95% confidence interval ( ci ): 1.19 to 1.56; stage ii css hr : 1.52; 95% ci : 1.26 to 1.83; and stage iii os hr : 1.45; 95% ci : 1.30 to 1.61; stage iii css hr : 1.54; 95% ci : 1.36 to 1.75]. Conclusions Despite a temporal increase in ln yield, the proportion of cases with ln positivity has not changed. Lymph node yield is associated with survival in patients with stages ii and iii colon cancer. The association between ln yield and survival is unlikely to be a result of stage migration.


INTRODUCTION
The importance of lymph node (ln) yield in surgery for colon cancer lies in the reported association between the number of lns evaluated and survival 1 . That association is independent of whether the lns are involved with metastatic disease 2 . Given that association, international guidelines recommend that at least 12 lns be evaluated with resection of the primary tumour [3][4][5][6] .
Since the publication of those clinical guidelines in the early 2000s, ln yield has increased in routine practice. In their analysis of data from the Surveillance, Epidemiology, and End Results Program in the United States, Parsons et al. 2 reported that the proportion of surgeries in which 12 or more lns were evaluated increased from 35% in 1988-1990 to 74% in 2006 -2008. Similarly, a population-based study from the Netherlands found that the rate of adequate ln yield increased to 59% in 2009-2011 from 13% in 2000-2002 7 . Population-based data from Ontario have show n similar temporal trends: during 1991-1993, 10 or more lns were resected in only 26% of patients; by 2004, the proportion of resections with more than 12 lns evaluated had increased to approximately 70% 8,9 .
We undertook a population-based study to explore ln yield in a contemporary cohort, to identify factors associated with ln yield, and to evaluate the association between ln yield and survival in routine clinical practice.

Study Design and Population
This population-based retrospective cohort study describes nodal yield and outcomes for patients with stages ii and iii colon cancer in the Canadian province of Ontario. Ontario has a population of approximately 13.5 million people and a single-payer universal health insurance program. The study population included patients who underwent resection of stage ii or iii colon cancer in Ontario during 2002-2008. Using the Ontario Cancer Registry (ocr), we identified a study cohort consisting of all incident patients with colorectal cancer (crc) diagnosed in Ontario during 2000-2008. We then identified all patients who underwent primary tumour resection within 6 months of diagnosis. The ocr does not capture disease stage for all patients; we therefore obtained surgical pathology reports for a random sample of 25% of the patients. Reports for patients undergoing surgery in 2005 were not available; the study cohort was thus restricted to patients who underwent surgery in 2002-2004 and 2006-2008. Patients with rectal cancer and non-adenocarcinoma histology were excluded, as were patients for whose surgeries the number of lns was unspecified. The study was approved by the Research Ethics Board of Queen's University.

Data Sources
The ocr is a passive, population-based cancer registry that captures diagnostic and demographic information for at least 98% of all incident cases of cancer in the province of Ontario 10 . The ocr also provides information about vital status and cause of death. Records of hospitalization from the Canadian Institute for Health Information provided information about surgical procedures; those records are known to have a very high level of completeness for crc surgery 11 . Provincial physician billing records from the Ontario Health Insurance Plan, treatment records from regional cancer centres, and provincial records of chemotherapy delivery (New Drug Funding Program and the Ontario Drug Benefit) were used to identify chemotherapy use. The datasets were linked using unique encoded identifiers and were analyzed at the Institute for Clinical Evaluative Sciences. Surgical pathology reports were obtained from the ocr. A team of trained data abstractors reviewed the pathology reports and entered information about extent of disease and ln yield into an electronic database.

Measures and Outcomes
Indicators of the socioeconomic status of the community in which a patient resided at diagnosis were linked as previously described 12 . Quintiles of median household income were based on the household income distribution for the full province of Ontario. Geographic regions reflect the catchment areas for Ontario's regional cancer centres 12 . Comorbidity was classified using the modified Charlson index and was based on all non-cancer diagnoses recorded during any hospital admission within the 5 years preceding surgery 13 . Each patient was assigned a hospital volume index based on the total number of colon cancer resections performed at their respective hospital in the preceding 12 months. We used the same approach to derive a surgeon volume index for each patient. Laparoscopic surgical resections were identified using Ontario Health Insurance Plan physician billing records.
Overall (os) and cancer-specific survival (css) were determined from the date of surgery. To account for the potential of cause-of-death miscoding, css included death from any cancer. Complete information about vital status in the ocr was available up to 31 December 2012; cause of death was available up to 31 December 2010.

Statistical Analysis
The chi-square test was used to compare proportions between study groups. Lymph node yield was evaluated as a continuous variable and a categorical variable-that is, 12 or more lns compared with fewer than 12 lns, and by quartile. Factors associated with ln yield were evaluated by modified Poisson regression. The associations of patient-, disease-, and treatment-related factors with os and css were evaluated using Cox proportional hazards regression models stratified by disease stage. Results were considered statistically significant at a p value less than 0.05. All analyses were performed using the SAS software application (version 9.3: SAS Institute, Cary, NC, U.S.A.).

Study Population
Linked administrative data sets identified 25,613 potentially eligible patients who underwent resection of primary colon cancer during 2002-2008. Surgical pathology reports were reviewed for 7519 randomly selected patients. The age, sex, comorbidity, and survival of the randomly selected patients did not differ substantially from those of the 18,094 unselected cases ( Table i). Of the 7519 randomly selected patients, 270 (4%) were excluded. Of the 5519 surgical pathology reports for patients with stage ii or iii disease, 11 (0.2%) lacked an explicit nodal count and were therefore excluded. Accordingly, the study cohort consisted of 5508 patients. Table ii shows the characteristics of the study population. Median age in this cohort was 72 years; 51% of the patients were men; and 47% (n = 2593) had stage ii disease.

Practice Patterns and Factors Associated with Nodal Yield
Median ln yield in the study cohort was 15 nodes. In 72% of the patients (n = 3941), 12 or more lns were evaluated. Figure 1 shows temporal trends for ln yield. Mean and median ln yield increased, respectively, to 20 and 17 from 13 and 11 (p < 0.001). The proportion of patients with a ln yield of 12 or more increased to 89% (717 of 809 patients) in 2008 from 48% in 2002 (301 or 623 patients), p < 0.001. The proportion of patients with node-positive disease remained stable over the study period (53% in 2008 vs. 54% in 2002, p = 0.357).
Table iii shows factors associated with ln yield. The factors associated with a ln yield of 12 or more nodes were younger age, higher socioeconomic status, lesser comorbidity, more recent year of surgery, a right-sided tumour, and a higher-volume hospital. Those factors remained significant when ln yield was analyzed as a continuous variable; however, in the model, female sex, higher T stage, and laparoscopic resection were also associated with a greater ln yield. Surgeon volume was not associated with ln yield in either model.

DISCUSSION
In this population-based study, we described ln yield for patients with colon cancer in routine clinical practice in the contemporary era. Several important findings emerged. First, ln yield increased substantially over the study period. Second, despite the increase in ln yield, the proportion of patients with node-positive disease did not change. Third, we found that age, socioeconomic status, comorbidity, year of surgery, laterality, and hospital volume were associated with ln yield. Surgeon volume was not associated with ln yield. Finally, our data demonstrated a substantial association between ln yield and survival for both stage ii and stage iii colon cancer in the general population.
Ontario data from the early 1990s revealed that, in 74% of patients, fewer than 10 nodes were resected 8 ; by the late 1990s, fewer than 12 lns were being evaluated in 73% of patients with stage ii disease 9 . After a multifaceted knowledge translation intervention study designed by Wright et al. in 2004 14 , the proportion of patients having 12 or more lns resected increased to 76%. Our study demonstrates that, in 2006-2008, more than 12 lns were being evaluated in more than 80% of patients; that finding is comparable with nodal yield data in the United States during 2006-2009, as reported by Parsons et al. 15 .
Our study also provides insight into factors that are associated with ln yield. Understanding the factors that affect ln yield is important if optimizing yield has the potential to improve patient outcomes. The association between ln yield and survival has been established in numerous studies. A large systematic review by Chang et al. 1 that included seventeen studies and 61,371 patients reported a consistent association between ln count and survival for both stage ii and iii crc. The existing literature has identified three broad factors that are associated with nodal yield: the operating characteristics of the surgeon, the facilitating techniques of the pathologist, and patient-related factors 16 . The former two factors are "modifiable" in the sense that interventions to alter ln yield can be established; in contrast, patient-and disease-related factors are not modifiable.
With respect to modifiable factors, our study reveals that hospital volume is associated with ln yield, but that surgeon volume is not. Those two modifiable factors have had inconsistent associations with ln yield in the literature [17][18][19][20][21][22][23][24] . Nevertheless, although variations in hospital-associated ln yield are larger than those for individual pathologists and surgeons, patient-related variations are largest 18 . Of the non-modifiable patient factors, senescence has consistently been shown to be associated with lower ln yield 18,20,22,[24][25][26] . That observation might reflect decreased immune response with aging 27 and comorbidity 28 . Right-sided tumours are also known to be associated with higher ln counts [18][19][20][21][22][24][25][26]29 , likely a result of a larger mesenteric surface and the presence of multiple vascular trunks 30 .
It must be emphasized that this retrospective study confirms the known association between ln yield and survival; it does not establish causality. The current debate is focused on the mechanism behind the association: whether ln yield itself leads to improved survival (for example, because of a direct therapeutic effect or appropriate stage classification) or whether it is a surrogate for other factors associated with outcome, such as quality of care or variation in host or tumour biology.
At present, studies have failed to establish that a more extensive ln dissection confers a direct therapeutic benefit by removal of occult micrometastases 31,32 . Studies have also refuted what is perhaps the most debated mechanism: stage migration, whereby detection of node positivity "migrates" a patient from a lower stage to a higher one, transforming that patient from a "high-risk" stage ii individual to a "lowrisk" stage iii individual, thereby spuriously improving survival in both groups 33 . The largest retrospective data collections in both the United States and Europe have shown that, despite increases in ln yield over time, and increased survival with increased ln yield, the yield of positive lns has not increased 2,7,34 . Our results are consistent with that finding. Moreover, stage migration would not explain the observed association between ln yield and survival of patients with stage iii disease.
Finally, the literature does not support the assertion that ln yield is a marker for quality care. There is no association between patients who receive an adequate ln yield and guideline-recommended postsurgical crc care 35 . There is also a strong association between ln yield and outcome in single centres, where provider factors would be held constant 36 .
Some of the factors that we found to be associated with a higher ln yield-younger age, lower comorbidity,   right-sided colectomies-raise the question of whether ln yield might reflect prognosis rather than drive it. Patients with histologic evidence of a more vigorous immune response experience improved outcomes 37,38 . Similarly, patients with a depleted or under-stimulated lymphocytic response to their crc, either within their lymphatics 39 or at the tumour edge 40 , have a relatively unfavourable prognosis. Galon et al. 41 coined the term "immune contexture" to describe the effect of the local host immune reaction in crc. There is a negative association between senescence and decreased nodal count 27 , and possibly a stronger association between comorbidity and diminished immunity 28 ; also, patients with crcs that genomically harbour microsatellite instability have a propensity for right-sided malignancies 42 , higher ln yields 43 , and better overall prognosis 44 . Given those findings, it is entirely plausible that ln yield is the surrogate marker for an immune response that reflects inherent prognosis. Although our study provides detailed data about ln yield in early-stage colon cancer in a contemporary population-based cohort, several methodology limitations merit comment. The study population was identified using linked administrative health databases. Although the ocr and the Canadian Institute for Health Information dataset are known to be consistent and complete, it is possible that our results might be biased by misclassification. As with The analysis excluded 147 patients for "both" or unstated laterality.
any retrospective cohort study, the observed association between ln yield and survival is vulnerable to confounding by incompletely controlled differences between patient groups. For example, our results suggest inferior outcomes for patients treated with open surgery rather than with laparoscopic surgery. It is highly likely that that observation is itself driven by residual confounding rather than by a true difference in outcome.
Our study population was identified from the ocr and is, therefore, unselected and includes patients of all ages. Because of the work required to manually review surgical pathology reports, our study population consists of a random sample of approximately 25% of all cases treated in Ontario during the study time period. No substantial differences were observed between the randomly selected patients and those not included in the study cohort ( Table i). We did not obtain pathology reports for patients who underwent surgery in 2005; we feel that this omission is unlikely to bias our results in any significant way. Finally, our study does not establish whether there is a causal relationship between ln yield and outcome, or whether ln yield is a surrogate for other elements of care or for variation in host-tumour biology.

CONCLUSIONS
In this contemporary population-based study, we found that, despite international guidelines and numerous knowledge translation initiatives, fewer than 12 lns are resected in one quarter of patients with stages ii and iii colon cancer. Lymph node yield is associated with age, comorbidity, socio economic status, tumour laterality, and hospital volume. Greater ln yield is associated with improved survival in routine clinical practice and is unlikely to be explained by stage migration. Future work should further explore the extent to which this survival benefit is driven by the complex relationship between tumour biology and host response.

ACKNOWLEDGMENTS
Parts of this material are based on data and information provided by Cancer Care Ontario. However, the analysis, conclusions, opinions, and statements expressed herein are those of the authors and not necessarily those of Cancer Care Ontario.
This study was supported by the Institute for Clinical Evaluative Sciences (ices), which is funded by an annual grant from the Ontario Ministry of Health and Long-Term Care (mohltc). The opinions, results, and conclusions reported in this paper are those of the authors and are independent from the funding sources. No endorsement by ices or mohltc is intended or should be inferred. This work was supported by the Canada Foundation for Innovation and the Canadian Institutes of Health Research. CMB is supported as a Canada Research Chair in Population Cancer Care.

CONFLICT OF INTEREST DISCLOSURES
We have read and understood Current Oncology's policy on disclosing conflicts of interest, and we declare that we have none.