The lymph node status as a prognostic factor in colon cancer: comparative population study of classifications using the logarithm of the ratio between metastatic and nonmetastatic nodes (LODDS) versus the pN-TNM classification and ganglion ratio systems

Colon cancer is the most frequent malignancy in both sexes in Western countries, with an incidence of approximately 471,000 cases per year and a mortality of 228,000 cases per year in Europe [1]. Lymph node involvement is the prognostic factor most directly related to survival and the disease-free interval. Thus, patients with stage I or II disease have a 5-year survival rate of more than 75% compared to 30–60% in patients at stage III or IV. In addition, the total number of lymph nodes analysed, both in cases at lymph node stage pN0 and in patients with tumour infiltration, has been shown to be a prognostic factor [2‐4].

Until now, the pN stage in the TNM classification has been the “gold standard” for lymph node staging of colorectal carcinomas, but this system recommends collecting at least 12 lymph nodes for the staging to be reliable. However, new prognostic staging systems have been devised, such as the ganglion quotients or lymph node ratios (LNR) and natural logarithms of the lymph node odds (LODDS) methods, which aim to refine the construction of risk groups within the lymph node stage of this tumour and thus, provide better individualised oncological treatments for them.

The aim of this study was to establish and validate the predictive and prognostic ability of the LNR and LODDS staging systems and compare them to the pN nodal classification of the TNM system in a population sample of patients with colon cancer.

During the 4 years of the study, 944 patients were diagnosed with colon cancer in our area, 396 patients were excluded from our work for different reasons, as shown in Fig. 1. Thus, 548 patients were finally analysed in this present study. The median follow-up period and range was 51 (30–64) months. The clinical and histopathological characteristics of the sample are shown in Table 1. The overall survival results and disease-free survival, according to the univariate and multivariate analyses, are shown in Table 2.

Overall survival and disease-free survival, comparing the pN classification with the LNR and LODDS, are shown in Figs. 2 and 3. For the overall survival (Fig. 2), groups with well-differentiated survival rates and with non-overlapping survival curves were observed in the three groups and there were significant differences (p <  0.0001) between all of them. However, the LNR and LODDS curves were easier to discriminate because their separation was clearer and more balanced. For disease-free survival (Fig. 3), the discrimination between the pN0 and pN1 groups was poor, but this phenomenon was adequately corrected for the LNR and LODDS curves which could be sufficiently discriminated to be able to estimate the survival prognosis.

An analysis of the 3 pN groups, according to their corresponding LODDS and LNR results, lymph node analysis results, and overall survival is shown in Table 3. As shown, in the LODDS, the analysis of fewer lymph nodes was associated with a worse prognosis, and this relationship was the same for the number of positive lymph nodes. Similarly, the results shown in Table 3, and visually represented in Fig. 4 (where the prognosis was B, ‘better’, W, ‘worse’, or M, ‘much worse’) correspond to the different LODDS subgroups. In the pN0 group, LODDS differentiated groups with different prognoses while the LNR was unable to identify these differences. According to the different prognosis groups established by LODDS, within both the pN0 and pN1 groups there were prognosis subsets, and subsets with a very different prognosis were also distinguished in the pN2 group.

LNR and LODDS analysis can distinguish risk subgroups with different survival rates more precisely than the condensed pN category alone. In the multivariate study, LODDS was identified as an independent risk factor for mortality and disease-free interval in non-metastatic colon cancer patients. After the determining presence of distant metastasis (M factor) and the condition of the primary tumour (T factor), the most important element in predicting the prognosis of colon cancer is lymph node involvement (N factor). Given its enormous importance, especially in terms of prognostic and therapeutic decisions, gaining a detailed picture of the lymph node status of patients with colon cancer should be a priority for clinicians involved in the diagnostic-therapeutic process of these patients.

Firstly, our results confirm that it is possible to create prognostic TNM-classification subgroups, based on the patient lymph node stage (pN), as shown in Figs. 2 and 3. Subgroups with well-differentiated survival rates can be identified using various classification systems. Nonetheless, in colon cancer, the total number of nodes analysed is a fully demonstrated prognostic factor [2‐4]. In fact, this is the main drawback of the TNM classification because it only considers the number of affected lymph nodes and ignores the influence associated with the total number of lymph nodes analysed. Therefore, the new lymph node staging systems (LNR and LODDS) were created to try to provide a more detailed view of the lymph node status in patients by analysing both the total number of lymph nodes analysed and the total number affected.

Secondly, LNR analysis can identify subgroups with distinct prognoses in patients with non-metastatic colon cancer based on their lymph node staging, as widely reported in the international literature and as confirmed by the results we obtained in this work. This system has been shown to be superior to standard TNM classification, thus in 2010, Ceelen et al. performed a meta-analysis that included 16 studies with a total of 33,984 patients diagnosed with colon or rectal cancer and identified that the results of LNR analysis were an independent prognostic factor for stage III patients [6]. Furthermore, Moug et al. found evidence that the LNR results may be prognostically significant for every tumour stage [7]. Based on our results (Figs. 2 and 3) we can conclude that LNR analysis can more precisely pinpoint subgroups than the TNM classification, both in terms of overall and disease-free survival because the curves it produces are more differentiated from each other than those produced by the condensed N category. However, the main disadvantage of LNR-based grouping is that it cannot classify the prognosis of patients who do not have any affected lymph nodes. Thus, in the same way as the TNM classification for the pN0 group, every case with no positive lymph nodes is included in the same group (LNR 0%) regardless of the total number of lymph nodes analysed. This characteristic is very important for colon cancer because about 75% of patients submitted to surgical treatment do not have any affected lymph nodes when studied anatomopathologically [8].

Like the LNR curves, the LODDS analysis (Figs. 2 and 3) can also distinguish risk subgroups with different survival rates more precisely than the condensed pN category. The advantage of the LODDS-based grouping over LNR-based grouping is that it can differentiate risk subgroups within the N0 category as well as also differentiating risk groups within the pN1 and pN2 categories.

The LODDS lymph node classification system has been used in different types of cancer. It was first applied in breast and gastric cancers and is now being studied in colon cancer. In 2004, Vinh-Hung et al. were the first to use this novel statistical technique in a group of 83,686 patients with breast cancer [9]. In terms of prognostic discrimination, the results using this technique were better than those for the TNM classification and similar to the LNR grouping technique results among patients with affected lymph nodes. Thus, the LODDS method was considered superior because it can be used to study lymph node involvement in these patients at all classification levels. Following on from studies on breast cancer, Yildirim et al. questioned the usefulness of the LODDS method, because the results obtained from it were similar to those from the LNR technique, yet its mathematical calculation is more complicated [10]. However, after first using LODDS in breast cancer samples, Sun et al. subsequently also tested these techniques in a sample of 2547 gastric cancer cases and demonstrated the superiority of the LODDS technique over the TNM and LNR methods [11]. Nonetheless, several authors [12, 13] favour the LNR method with which they obtained results similar to those from the LODDS. Some authors, maintain that the results obtained with the LODDS are lower than expected and that this could reflect a false correlation between the N classification and the result [14]. Finally, Wang et al. designed a population study of 24,477 patients with stage III colon cancer in which they demonstrated the importance of the LODDS method [15]; they obtained better results both against the TNN and LNR classifications and so they considered it a better risk stratification technique. Chang et al. obtained similar results with a sample of 9644 patients with stage I–III cancer [16]. However, another study in a sample of 1297 patients with stage I–III cancer obtained better results with the LNR than with the LODDS [17]. In summary, the LODDS technique has multiple advantages over the other methods used for lymph node staging in various cancers, including in colon cancer [18, 19].

In our multivariate analysis of the multiple factors involved in the overall survival of colon cancer (Table 2), of those implicated in overall colon cancer survival, age was independently-weighted, the number of lymph nodes analysed was limiting, and the number of positive lymph nodes (pN) in pT3–T4 cases and the LODDS results themselves were all involved. Of note, when these other factors were present, the LNR method lost its status as an independent risk factor. Regarding the disease-free interval, only LODDS, the number of affected lymph nodes, and the advanced T category were identified as independent risk factors in our multivariate analysis. Chemotherapy is usually indicated in N+ patients or in T3 N0 patients with additional risk factors (such as obstruction or perforation). Thus, it is logical that in the multivariate study, chemotherapy is not a decisive factor for survival because its effect is already be encompassed within the pN.

Similar to our finding in this series, in the international scientific literature there is, on the one hand, a lot of evidence that both the T category of the TNM and the nodal stage are directly related to survival, as demonstrated in various studies [20‐22]. On the other hand, the usefulness the LODDS and the LNR remains controversial. Several other studies agree with our results and also show a direct relationship between the results from LODDS and survival, without showing a similar relationship with the LNR technique. Thus, Arslan et al. demonstrated this relationship, only with the LODDS results, in a series of 444 patients with colon cancer [23]. Similarly, Persiani et al. used Cox regression to demonstrate the same relationship with the LODDS results, but not with those from LNR, in a series of 258 patients with non-metastatic colon cancer [24]. In a study including 17,632 cases, Huang et al. demonstrated that the LODDS method was more accurate than the LNR when calculating specific survival rates [25]. On the other hand, other authors have demonstrated that both methods are directly related to survival [9, 10]. But, in a study of 372 patients with gastric cancer, Liu et al. showed that the NRL technique was a better predictor of overall survival than the LODDS method [12]. Many of these discrepancies may be the result of statistical problems derived from the data collection itself and from the design and definition of the different variables.

According to our findings in this present work, the TNM classification should be complemented with the LODDS risk sub-group in clinical practice because the precision of this method has been demonstrated in the staging of risks. Thus, a patient with an increased risk according to LODDS will have a worse predicted survival rate than one only assessed according to the TNM classification. This fact becomes even more important in patients in which an insufficient number of lymph nodes could be obtained and analysed.

The main limitation of the work we present here is its retrospective population-study design. The great disadvantage of population studies is their inherent variability. This variability is derived both from differences in the study centres (5 centres in our case) and in the possible variation in the surgical techniques used, surgeons’ specialisations, and the anatomopathological sample analysis techniques used. In contrast, the advantage of population studies is that, by analysing all the patients in a specific area without any selection, they provide a global view of the real situation within a given period. This also implies that the results are representative only of the population studied, in our case the province of Castellon in Spain. Therefore, it is very important to compare the similarity of the characteristics of this study population (shown in Table 1) with those of other populations in order to assess the external validity of our results. As stated in the literature, a median of 10 lymph nodes are usually analysed, whereas different scientific societies recommend the analysis of at least 12 nodes. However, several articles, especially those such as ours in which a population analysis has been performed, were unable to reach this threshold number of lymph nodes [2, 8, 26]. Despite the analysis of a low number of lymph nodes, pN2 is not underestimated because at least 7 lymph nodes must be found to be positive in order to stage a patient as pN2. Nonetheless, the collection of fewer than 12 nodes can affect the confidence in the diagnosis of a true pN0 stage. Furthermore, obtaining only 10 lymph nodes does not dramatically affect this confidence level: the risk of misclassifying a pN0 after having analysed 10 lymph nodes is less than 2% [27].

Another limitation was our use of the T category from the 6th rather than the 7th edition of the TNM classification; this was because it was the version in effect during the period of data collection. However, to help reduce this bias, we allocated the patients either into a group with earlier tumours (stages T1 and T2) or into one with locally more advanced tumours (stages T3 and T4). Thus, minimising the effect of any modifications to the T factor classification implemented between these two editions of the TNM. In our province, the tumour registry for colon cancer was started in 2004 and here in this study we used the whole database dating back to its origin. However, because the validation of the cases we analysed was complex and we needed 5-year post-surgery follow-ups, the complete database could only be obtained up to 2012. The methods of diagnosis and anatomopathological analysis have not substantially changed in our area and so the results should be comparable to those more recently obtained. However, it would be interesting perform the comparison using a more up-to-date database.

The pN classification is useful as a prognostic factor and remains the standard for lymph node staging because its measurement is very easy, and its groups have a clear and generally well-differentiated prognostic significance. However, the LNR and LODDS techniques can more precisely differentiate risk subgroups from within the pN groups.

Of the three methods tested in this study, the LODDS was the most accurate for staging non-metastatic colon cancer. This technique is slightly more difficult to calculate but, in turn, it is better able to distinguish patient prognoses than that derived from the TNM pN classification or the LNR method for N0 cases, especially in cases where an insufficient number of lymph nodes had been analysed. Therefore, the LODDS system complements the pN classification by improving the discrimination of the colon cancer prognosis in different groups, thus helping to more precisely adjust and individualise the indication for adjuvant treatments in these patients.