Introduction

Colon cancer is a leading cause of cancer death in the Western world [1]. A number of systems have been developed for the staging of colonic cancers, allowing prognosis and the need for adjuvant therapy to be more readily determined. The metastatic involvement of regional lymph nodes plays a significant role in all major classifications [2].

Patients with lymph node metastases (Dukes C) have lower 5-year survival rates and may benefit from adjuvant systemic chemotherapy [3]. Long-term studies suggest mortality due to cancer recurrence in Dukes B patients is approximately 25–30% [4]. This may be the result of understaging due to inadequate lymph node harvest, inadequate identification of lymph nodes during histopathological examination of the resected specimen, micro-metastases not detected by histopathological examination of resected lymph nodes or vascular invasion of the tumour with haematologic spread.

The precise number of lymph nodes, which must be examined in order to stage a cancer with reasonable accuracy, is debated (data varies considerably between different studies, ranging from 6 to 17), with 12 being generally accepted as optimum [5–8]. However, an almost universal finding is that long-term outcome improves in direct relation to the number of lymph nodes retrieved and examined. Thus, although it is clinically impractical to count out lymph nodes at the time of surgery, it was suggested in 1996 by Goldstein et al. [5] that the surgeon should attempt to retrieve as many nodes as possible during curative resection.

All study participants came from the catchment area of Middlemore Hospital—a large tertiary care institution that services a unique multi-ethnic population (European 41%, Pacific Islander 28%, Maori 15%, Other 16%) [9]. International guidelines may not be accurate in this context. The aim of this study was to assess the impact of the number of lymph nodes examined on recurrence and mortality in our population following attempted curative resection of Dukes B colon cancer.

Materials and methods

A retrospective database was constructed of 328 consecutive patients who underwent major colorectal resection for histopathologically defined Dukes B cancer between January 1993 and December 2001 (inclusive). All surgery was performed and pathological specimens analysed at a single tertiary care facility—Middlemore Hospital.

An online pathology database was accessed to allow identification of all patients who had had a Dukes B colorectal neoplasm removed. Individual patient National Health Index’s (NHI) were then used to acquire information for the following parameters:

  • Age (at the time of surgery)

  • Gender

  • Procedure performed

  • Acuity of Surgery, i.e. acute or elective

  • Number of lymph nodes examined

  • Local or distant cancer recurrence within 5 years of index operation. Disease activity was determined by clinical examination, regular sigmoidoscopy, colonoscopy, CEA levels and radiology. There was no standard follow-up protocol in the time frame studied.

  • Mortality within 5 years of index operation. This was subsequently sub-classified as cancer or non-cancer related.

All information was entered into an electronic worksheet. Patients with incomplete data (this included those with inadequate documentation of the parameters outlined above or those lost to follow-up), previous colorectal neoplasms or perioperative mortality (death during or within 30 days of the index operation) were excluded from this study. Additionally, all patients with a rectal neoplasm (defined as occurring distal to the rectosigmoid junction) were excluded as the method of invasion, spread and management of such cancers varies from colonic neoplasms (Table 1). Data for the remaining 216 patients were then subjected to multivariate and logistic regression analysis. Statistical analyses were carried out using SAS version 9.3 [10]. Categorical variables were compared using the χ 2 test, and the Student’s t test was used for continuous variables. Two separate multivariate regression models were built using the number of lymph nodes harvested as outcome, with each containing either ‘patient death’ or ‘cancer recurrence’. This was because it was found that by placing these similar endpoints in the same model, the level of statistical significance was erroneously lowered. A multiple logistic regression was used to determine predictors of death.

Table 1 Patients who met the exclusion criteria
Full size table

A graph depicting the cancer recurrence rates as broken down by lymph node strata was constructed. Visual representation of these different rates allowed for easy identification of a drop. A one-tailed χ 2 test was used to determine if any observed drop in the recurrence rate was statistically significant.

Additionally, receiver operator characteristic (ROC) curves were constructed for lymph nodes as a predictor for ‘patient death’ and ‘cancer recurrence’. The ROC curve determines how good a predicted risk value is at discriminating a bivariable outcome (in this case ‘death’ or ‘no death’, or ‘recurrence’ or ‘non-recurrence’) by constructing a graph with sensitivity on the Y-axis and one specificity on the X-axis. It depicts the inverse relationship between sensitivity and specificity (an increase in sensitivity will be followed by a decrease in specificity) [11].

Results

Patient demographics are shown in Table 2. The mean number of lymph nodes examined was 16.0 (median 14; range 2–48) (Fig. 1).

Fig. 1
figure 1

Graphic depiction of the number of lymph nodes retrieved during surgery

Full size image
Table 2 Patient demographics
Full size table

The number of patients with cancer recurrence or mortality within 5 years of the index operation is shown in Table 3. The mean number of lymph nodes examined in those who died within 5 years was 12.8 vs. 17.5 in those who remained alive (p = 0.0027) (Fig. 2). Similarly, the mean number of lymph nodes examined in those with evidence of recurrence within 5 years was 11.8 vs. 17.1 in those without recurrence (p = 0.0007).

Fig. 2
figure 2

A boxplot illustrating the number of lymph nodes harvested vs mortality at 5 years. The mean is depicted by the bold horizontal line within each plot. The notches represent the upper and lower confidence intervals. The notches on each plot do not overlap indicating a significant difference between the groups

Full size image
Table 3 Patient outcome at 5 years as defined by Cancer Recurrence and Mortality (both cancer and non-cancer related)
Full size table

The graph depicting cancer recurrence rates as broken down by lymph node strata showed that the recurrence rate tended to decrease as the number of lymph nodes examined increased (Fig. 3). Focussing on these specifically, it can be seen that the recurrence rate in the 1–4 node group is 46.2%. This figure is misleading, however, because the number of patients in this group was relatively small. In the 5–8, 9–12 and 13–16 node groups, the recurrence rate was around 25–30%, which is what we would expect for a Dukes B tumour. However, after the 16th node mark, it dropped off sharply to 6.5% and remained at a low level for the remaining strata. This new recurrence rate is similar to that seen in Dukes A tumours. Analysis of the drop from the 13–16 node group to the 17–20 node group reveals a significant difference between the two groups (p = 0.0368).

Fig. 3
figure 3

A bar graph depicting the percentage of patients with evidence of cancer recurrence as broken down by lymph node strata. Each stratum contains the recurrence rate for a four-node group. This number was chosen because it ensured that each group contained a comparable number of patients whilst simultaneously allowing for a clear distinction to be made at smaller group intervals

Full size image

The ROC curve for cancer recurrence within 5 years (Fig. 4) showed that the examination of 12 lymph nodes provided maximal sensitivity (0.60) and specificity (0.64). Similarly, the curve for mortality at 5 years (Fig. 4) showed that the examination of 13 nodes provided a maximal sensitivity (0.62) and specificity (0.61). The area under the curve for recurrence was 0.68 and for mortality was 0.65. This shows that although the number of lymph nodes examined did predict recurrence and mortality, it was not very good at it.

Fig. 4
figure 4

Receiver operating characteristic (ROC) curve for cancer recurrence (circles) and mortality (triangles) after 5 years

Full size image

Increasing age was significantly associated with a decreasing number of lymph nodes harvested and examined (p = 0.0018). Female patients had a mean of 16.4 nodes examined compared to 15.4 in male patients (N.S.). Acuity of surgery (acute or elective) was also found to have no significant bearing on lymph node retrieval.

Discussion

Lymph node retrieval and examination is essential for determining the stage of colonic cancer and subsequently the need for further adjuvant therapy. It is also a good indicator of prognosis, with positive nodes being a strong predictor of cancer recurrence and death in patients with no evidence of distant metastatic disease [12]. The present study has found that cancer recurrence and death in patients with Dukes B colon cancer increase significantly as the number of lymph nodes examined decreases. Furthermore, our results have shown that examining more than 16 lymph nodes in patients with Dukes B colon cancer is associated with a significant reduction in cancer recurrence.

There has been a considerable controversy surrounding the issue of lymph node harvesting. Intuition dictates that no minimum number of lymph nodes must be examined in order to safely rule out the risk of recurrence, but rather there is a continuum where confidence of non-recurrence increases in concordance with the number of lymph nodes removed. This principle is reflected in most of the literature, and current international guidelines recommend removing as many nodes as possible at the time of surgical resection [5–8]. However, whilst this practice is clinically feasible and practical, determining the ideal number of lymph nodes to be harvested and examined is essential for two reasons. Firstly, it provides an indicator as to where it can be stated with reasonable certainty that the chance of recurrence (and death) is reduced. Secondly, it provides an important means of quality control. An institution that consistently fails to report this ‘ideal’ number of nodes may indeed be representative of a poor level of surgical or pathological care [13].

Studies published over the last few years have varied greatly in their establishment of an ideal number of nodes to be reported. The general consensus, initially proposed by Fielding et al. [14] in 1991, is that a minimum of 12 lymph nodes is needed to adequately stage colon cancer. Indeed, the American National Quality Forum has recently endorsed a 12-lymph node minimum as a baseline standard for hospital-based performance in the surgical treatment of colon cancer [15]. According to this recommendation, 62.5% of patients in our series were adequately sampled and staged. This compares relatively favourably to other studies including large series in Minnesota and Canada, where 37% of 116,995 patients and 27% of 1,789 patients, respectively, received adequate lymph node examination [16, 17]. However, smaller single-institution studies have been shown to achieve higher mean lymph node retrieval rates [8, 16, 18]. It is also interesting to note that even though the current guidelines recommend retrieving as many nodes as possible, several large organisations (including the College of American Pathologists) have described removing and examining a specific number [2].

The influence lymph node harvesting exerts on the staging and prognosis of colon cancer assumes a perfect surgical technique and pathological examination. Practically, there is much room for error. We acknowledge that factors such as variation in surgical technique and diligence of the pathology staff in identifying and correctly excluding negative nodes may all alter the consistency of the results (indeed, a recent study found that lymph node examination varied with reporting pathologist, but not with the operating surgeon) [19]. However, this reflects a typical clinical scenario, and thus standardising any error, which may have occurred, would be technically difficult and reduce applicability to other tertiary institutions.

Accurate staging of colon cancer is important in determining prognosis and the need for adjuvant chemotherapy [12]. The clinician’s certainty of the cancer being node-negative increases in accordance with the number of lymph nodes examined. However, it is still unclear whether a ‘cut-off’ number for nodes examined in Dukes B patients should form the basis of deciding whether adjuvant therapy is used., i.e. should patients in whom less than 12 nodes are examined be candidates for chemotherapy because they have a higher chance of having undetected nodal metastases? Similarly, should such patients be subjected to more intensive and frequent surveillance for cancer recurrence? Whichever the case, confidence in long-term prognosis and deciding the need for adjuvant chemotherapy increases in direct relation with the number of lymph nodes retrieved by the surgeon and examined by the pathologist.

In summary, this study has shown that long-term outcomes such as cancer recurrence and death after attempted curative resection for Dukes B colon cancer were significantly reduced by increasing lymph node retrieval and examination in our population. Statistically, our results suggest that examination of more than 16 lymph nodes is associated with a significant reduction in cancer recurrence. This supports the current clinical practice of harvesting and analysing as many nodes as possible during surgical resection and pathological analysis.