Introduction

Recent progress in endoscopic resection, including the innovative technique of endoscopic submucosal dissection, has technically resulted in lesser dependence on laparotomy for patients with early invasive (T1) colorectal cancer (CRC) [1–3]. However, indicators for accurate assessment of the risk of lymph node metastasis (LNM) in patients with resected malignant polyps and identification of the parameters that should be integrated as criteria to determine requirement of additional laparotomy for lymph node dissection remain controversial.

Initial series of study to identify histopathological risk factors for LNM in early invasive CRC first appeared in the 1980s, and since then, unfavorable tumor grade [4–9] and vascular invasion [4, 6–10] have been regarded as important adverse features. According to the National Comprehensive Cancer Network (NCCN) guidelines [11], grade 3 and positive lymphovascular invasion are the only risk factors indicating the necessity of additional surgery for patients undergoing local excision of T1 CRC with a histopathologically confirmed negative margin. However, the issue of appropriate selection of patients at risk of LNM is not simply resolved by evaluating these two parameters because, in 1–5 % of T1 CRC cases, LNM reportedly occurs in tumors with neither unfavorable tumor grade nor vascular invasion [12–14]. Until the 1990s, it was justified to overlook such false-negative incidences because of the high rate of postoperative in-hospital mortality [15], which was reportedly around 7 % for curatively intended colectomies [16]. However, recent improvements in perioperative management has reduced the mortality rate associated with additional colectomy, which is now almost insignificant [2, 17, 18]. Based on this background, many studies have been conducted to attempt to identify novel parameters other than tumor grade and vascular invasion as standards for identifying patients in whom the safety of applying this observational policy will be warranted.

Consequently, new promising risk factors for CRC have been reported (Table 1) and some have already been adopted in recent guidelines as diagnostic parameters. In particular, tumor budding is listed in the guidelines of the European Society for Medical Oncology (ESMO) guidelines [19] and the Japanese Society for Cancer of the Colon and Rectum (JSCCR) [20]. Submucosal invasion depth (the 1000-µm rule) is recognized by the JSCCR guidelines [20] as a parameter indicating the necessity of additional laparotomy. Furthermore, recent studies have reported that a histological feature of poorly differentiated clusters (PDCs) reflects the metastatic potential of CRC more accurately than conventional histopathological features such as tumor grade and vascular invasion or tumor budding [21, 22]. Likewise, PDC assessment is reportedly a potential useful index of LNM risk in early invasive CRC [23].

Table 1 Review of articles reporting a significant value of new parameters for assessment of early invasive colorectal cancers
Full size table

The growing pursuit of novel indicators over the last decade has raised expectations of further advancement in the treatment of early invasive CRC. However, most proposed parameters have not been implemented because of the lack of reproducibility, validation, and/or standardization. Furthermore, few studies have compared the value of these new factors in a large multi-institutional case series of early invasive CRC. Therefore, in the present study, we compared the value of conventional and novel histopathological risk factors, including PDC, to predict the risk of LNM on the basis of Akaike’s information criterion (AIC). Our aim in this study involving 30 institutions specialized in treatment of CRC was to evaluate the potential value of new risk factors and clarify potential issues resulting from the use of these factors in routine practice.

Patients and methods

Patients

A total of 3556 early invasive CRCs were reviewed for pathology in 30 institutions belonging to the JSCCR. These institutions included university and major regional hospitals as well as cancer centers. All patients were consecutively treated in each institution within a voluntary period of time between 1980 and 2011 through laparotomy with lymph node dissection, and all tumors had their pathology confirmed as invading the submucosa. The average patient age was 64.7 years (range 23–97), and 2174 and 1382 tumors were resected from male and female patients, respectively. The median number of patients per institution enrolled in this study was 74 (range 10–478). All data except for those in interobserver study were collected from each institution by means of questionnaires in unlinkable anonymizong status. The ethics committee of the JSCCR approved the study protocol.

Evaluation of LNM parameters

In addition to the conventional parameters of tumor grade and vascular invasion, submucosal invasion depth, tumor budding, and PDC were evaluated in each institution without knowledge of the nodal status and clinical outcome to determine a potential correlation with LNM.

Submucosal invasion depth was evaluated according to the JSCCR guidelines [20]. In brief, the invasion depth was measured from the lower border of the muscularis mucosae (MM; when it was possible to identify or estimate the location of the MM) or from the surface of the polyp. For pedunculated lesions with a tangled MM (Peutz–Jeghers-type polyp), Haggitt’s level 2 [24] was used as the reference line of measurement.

Tumor budding was assessed by hematoxylin and eosin (H&E) staining and defined as an isolated single cancer cell or a cluster comprising <5 cancer cells, as reported previously [25], and after choosing a microscopic field with intense budding, a count was performed (magnification, 20× objective lens). Tumors with <5 budding foci were classified as low-grade and those with ≥5 budding foci as high-grade [20, 25].

PDC assessment

PDCs are defined as cancer clusters of ≥5 cancer cells infiltrating the stroma and lacking a glandular formation (Fig. 1) [21, 23]. H&E-stained tumor tissues were evaluated for the presence of PDCs, and positive judgment was made regardless of whether they are located inside or outside the vascular space. With regard to assessment in mucinous carcinoma, malignant clusters exhibiting the abovementioned features (i.e., stromal infiltration with minimal extracellular mucin formation) are classified as PDCs. On the other hand, cancer cell clusters within a large mucin pool (a so-called mucinous lake) are not classified as PDCs.

Fig. 1
figure 1

Poorly differentiated clusters (PDCs). a Cancer cell clusters located in the stroma, comprising ≥5 cancer cells, and lacking a glandular formation (arrows) are classified as PDCs (H&E staining; ×40 objective lens); b PDCs with intracytoplasmic lumina (ICL) or structures morphologically similar to ICL, such as cytoplasmic vacuoles and intracytoplasmic mucin (arrow head) (H&E; ×40 objective lens). a, b Extracts from the website for the multi-institution study for PDC at the 75th JSCCR meeting (http://jsccr.umin.jp/75/enq1.html)

Full size image

Detailed descriptions concerning the definition and positive judgment criteria of PDCs appear on the website for this multi-institution study (http://jsccr.umin.jp/75/enq1.html). No direct instructions were provided to the observers regarding microscopic evaluations of PDCs.

Interobserver study

We conducted a study to examine the interobserver reproducibility of PDC evaluations. In brief, H&E-stained glass slides of 50 randomly selected cases of early invasive CRCs treated with endoscopic resection (R0) between 1987 and 2001 were prepared at the National Defense Medical College (NDMC). They were circulated among 10 institutions (Keiyukai Sapporo Hospital, Takano Hospital, Kurume University Faculty of Medicine, NDMC, Yamagata Prefectural Central Hospital, Hiroshima City Asa Hospital, Teikyo University School of Medicine, Aichi Cancer Center Hospital, Tokyo Women’s Medical University Medical Center East, and Osaka Medical Center for Cancer and Cardiovascular Diseases) and the diagnostic results were compared. These 10 institutions were selected in order of the number of the subjects they enrolled from 22 institutions who entered themselves for the interobserver study.

Statistical analyses

Stratification of the risk of LNM according to conventional and new risk factors was evaluated using AIC [26], which was analyzed in a logistic regression model to identify the risk factor with the highest ability to identify the risk of LNM. The optimum model (i.e., the simplest effective model with the smallest information loss when predicting outcome) will have the lowest AIC value. To assess variability of the interobserver assessments of PDCs, Cohen’s kappa statistic for all possible pairs of 10 institutions and Fleiss’ kappa statistic were calculated. For comparison, corresponding statistics for tumor grade and vascular invasion were also obtained from the same sample set.

Statistical analyses were performed using three software packages: SPSS (SPSS, Inc., Chicago, IL, USA), Stata/SE 10 (StataCorp LP, College Station, TX, USA), and R (version 2.8.1, http://www.r-project.org/) [27].

Results

Incidence of PDC

PDC was observed in 1401 tumors (39.4 %), including 94 (17.8 %) with submucosal invasion <1000 µm and 1307 (43.2 %) with ≥1000 µm submucosal invasion (P < 0.0001) (Table 2). Although there was a strong correlation between tumor grade and PDC (P < 0.0001), PDCs were observed in 32.6 % of grade 1 tumors. The incidence of PDCs was also significantly associated with vascular invasion and the grade of tumor budding (both, P < 0.0001).

Table 2 Correlation between poorly differentiated clusters and other histopathological characteristics in early invasive colorectal cancer
Full size table

Risk factors for LNM

LNM was observed in 393 tumors (11.1 %) and submucosal invasion depth, tumor grade, vascular invasion, grade of tumor budding, and PDC were all significantly associated with the incidence of LNM (Table 3). The incidence of LNM was 17.4 and 6.9 % in tumors with and without PDCs, respectively (P < 0.0001). PDCs had an adverse impact on LNM irrespective of the degree of submucosal invasion (Table 3).

Table 3 Correlation between histopathological risk factors and lymph node metastasis in early invasive colorectal cancer
Full size table

Comparison of risk factors in terms of their ability to identify the risk of LNM.

AIC was used to compare the ability of all risk factors to identify the risk of LNM. As shown in Table 4, AIC was favorable in the order of vascular invasion, PDC, tumor budding, and tumor grade. Submucosal invasion depth was demonstrated to have the lowest ability to identify the risk of LNM. Odds ratio (OR) for LNM associated with grade 3 was high [OR, 7.2; 95 % confidence interval (CI), 4.4–11.8]; however, the results of AIC indicated that other risk factors (vascular invasion, budding, and PDC) had better ability to identify the possibility of LNM irrespective of the degree of submucosal invasion.

Table 4 Statistical analysis using the logistic regression model to assess an association between risk factors and lymph node metastasis in early invasive colorectal cancer
Full size table

Interobserver reproducibility of PDC

Fleiss’ kappa value for risk factors among all 10 institutions was 0.51 for PDC, 0.48 for tumor grade, 0.33 for vascular invasion, 0.29 for tumor budding, and 0.21 for submucosal invasion depth.

Figure 2 shows the range of Cohen’s kappa value for all possible pairs of 10 institutions regarding PDC, tumor grade, and vascular invasion. The range was 0.33–0.80 for PDC and 0.01–0.77 for vascular invasion. None of the 50 circulated tumor specimens was diagnosed as grade 3 tumor differentiation in one institution. Thus, the range of Cohen’s kappa value for all possible pairs of the remaining nine institutions was −0.03 to 1.00 for tumor grade.

Fig. 2
figure 2

Kappa values for PDC for all possible combinations of institutional pairs. Calculated Cohen’s kappa values for all possible pairs of 10 institutions were plotted for PDC and vascular invasion in order of the values. With regard to tumor grade, none of the 50 tumor specimens circulated were diagnosed as grade 3 in one institution; thus, the kappa values for all possible pairs of the remaining nine institutions were calculated. PDCs, poorly differentiated clusters

Full size image

Discussion

Over the last decade, various histopathological parameters have been proposed as potential risk factors for LNM in early invasive CRC, including those associated with the degree of submucosal invasion (the level of the polyp [24], relative level of the submucosal layer [28], measured depth [29] or width [30] of submucosal invasion, and the status of the muscularis mucosae [31]), cancer morphology (budding [32], growth pattern [33], and PDC [21]), and host response against tumor growth (inflammatory cell infiltration [34], myofibroblast proliferation [35], and microvessel count [36]). However, systematic review of articles disclose that different assessment criteria were often used for each parameter, and there have been a limited number of parameters (e.g., Haggitt’s level, relative depth of submucosal invasion, calculated depth (µm) of submucosal invasion, and tumor budding) for which positive results were validated in other studies on the basis of same assessment criteria (Table 1).

As a quantitative risk factor, the actual depth of submucosal invasion has been regarded as an important parameter of LNM in Japan [25, 30, 31]. That is because Haggitt’s classification criteria [24] poses problems for sessile-type cancers, which are always classified as level 4, and relative submucosal invasion levels defined by Kudo’s classification system [28] are difficult to apply to endoscopic resection specimens, which usually do not include the muscularis propria [37]. Based on the results of a multi-institution study to determine the cut-off point of the depth of submucosal invasion [38], the 1000-µm rule, which includes the concept of Haggitt’s level 2 as a yardstick for measurement of pedunculated polyps, has been adopted as a risk factor in the JSCCR guidelines [20]. In the present study, we found that submucosal invasion depth >1000 µm was significantly relevant to the incidence of LNM, and in particular, the false-negative rate (the incidence of positive LNM in the no-risk group) was most favorable among the parameters examined. However, it should be noted that the statistical power of identifying the risk of LNM was the lowest. Furthermore, a previous study demonstrated that approximately 80 % of malignant polyps treated with laparotomy meet the criterion of submucosal invasion >1000 µm, and this criterion was related to the LNM rate with an increase of only around 2 % compared with the average incidence of LNM in early invasive CRCs [38]. Consequently, we concluded that the 1000-µm rule has excellent power to identify cases with a very low risk of LNM; however, this criterion could increase the incidence of excessive indication of laparotomy.

Tumor budding is a histological feature of relatively small cancer foci detached from the tumor body at the invasive frontal region. The morphological features of tumor budding are put into context with the expanding knowledge of the epithelial–mesenchymal transition, which allows a polarized cell to assume a more mesenchymal phenotype with increased migratory capacity [39]. Although different assessment methods have been reported [40], budding has been recently defined in many studies as an isolated single cancer cell or a cluster of <5 cancer cells and as a significant risk factor for LNM in early invasive CRC when positive judgment was applied for ≥5 budding foci in a microscopic field of 20× objective lens [25, 31, 41–46]. This method was recently adopted by the JSCCR guidelines to evaluate tumor budding as an indicator of additional laparotomy following endoscopic treatment [20]. As expected, the present large-scale multi-institution study confirmed the adverse impact of high-grade budding on LNM on the basis of this assessment method.

More recent studies have highlighted the potential importance of PDC defined as ≥5 cancer cells with no gland formation as a new histological benchmark of CRC metastatic behavior [22, 23], and the present study confirmed the significant impact of PDC on LNM in early invasive CRC, demonstrated previously in a single-institution study [21]. The results showed that the statistical power of PDC to help in identification of the risk of LNM was the second best among the variables examined, i.e., after vascular invasion. It is generally recommended that the feature of leading edge of tumor is not included in grading tumor [47–50]. However, of note, AIC for identifying LNM risk in PDC was superior to that in tumor grade diagnosed when ignoring the finding of focal dedifferentiation at the invasive margin. Our data were consistent with the hypothesis proposed by a group of Italian surgical pathologists who stated that focally undifferentiated components, which invade fibrous tissue or the vasculature, probably play a fundamental role in lesion development [9].

With regard to judgment reproducibility, PDC was shown to be more favorable compared with conventional tumor grade [22, 23]. Our multicenter interobserver study also indicated that interinstitutional judgment disparities were smaller in PDC than that in cases of tumor grade or vascular invasion. In addition, the cut-off number of cancer cells in a cluster was arbitrarily determined for the purpose of deriving a rigorous definition for this parameter. Because PDCs do not include single cancer cells or very small clusters, which are reportedly difficult to assess in tumors with inflammatory cells using H&E-stained slides [31], improved interobserver agreements could be expected in the assessment of PDC compared with that of tumor budding, for which the requirement of cytokeratin staining for accurate identification of tumor buds has been repeatedly discussed [44, 51–55].

We believe that there is room for improvement with regard to the interobserver variation of PDC, which has not been evaluated in research trials or routine practice in any of the institutions participating in this study. Educational and training programs, as well as learning curves based on the experience gathered in routine practice expectantly make this parameter a useful tool in deciding treatment for patients with early invasive CRC.

Genetic molecular biomarkers including MSS, KRAS, or BRAF have recently been included as prognostic markers in the international guidelines [19]. With regard to the choice of treatment for early invasive CRC, however, it is unlikely that a molecular biological approach is very useful at present, and histopathological-based approaches will continue to be mainstream for a while. Compared with studies before the 1990s, in which most included <100 cases of early invasive CRC [25], we increased the number of cases analyzed within each institution (Table 1). Nevertheless, we believe a multi-institutional approach involving a large number of pathologists is essential to evaluate the value of new risk factors. In addition, the prevalence of LNM might be higher in retrospective surveys on only CRCs treated by surgery than the true prevalence because it seems possible that CRCs, which were less likely to have LMN, were excluded from the analysis. This would be likely to have contributed to a bias for the interpretation of the diagnostic accuracy for LNM. We need well-designed prospective cohort studies and on the basis of such an approach, a robust international standard for the risk assessment of locally excised early invasive CRC will be established.