Introduction

Colorectal cancer (CRC) is one of the most common malignancies, resulting in 16 800 deaths in the UK annually.1 2 Tumour stage (table 1) is an important determinant of outcome, with many early-stage tumours possessing an excellent prognosis (table 2).3–5 The NHS Bowel Cancer Screening Programme (BCSP) was initiated in 2006 and resulted in the identification of CRC at earlier stages (table 3).6 7 While many pT1 CRCs (ie, those in which adenocarcinoma extends into the submucosa) possess a low risk of regional lymph node metastasis (LNM), these still occur in approximately 5%–16%.8 9 In pT1 CRC, patients with LNM have a worse overall and cancer-free survival than those without LNM.10 Patients at significant risk of LNM are considered for surgical resection to identify tumour-involved nodes. This has highlighted the requirement for accurate risk assessment in pT1 CRC.

Table 1

TNM staging of CRC3

Table 2

Survival of CRC according to stage at diagnosis4

Table 3

CRC stage in screening versus symptomatic individuals6

The likelihood of LNM in pT1 CRC needs to be balanced against other patient factors when determining the best management. For example, age and comorbidities influence the risk–benefit ratio of surgical resection. The 2014 UK National Bowel Cancer Audit found a 2.9% 30-day mortality rate following major colorectal resection.11 Colorectal resection is resource intensive, with the cost estimated at £5500/patient. Therefore, the ability to make the best patient management decisions has important implications for health providers.12

LNM risk in pT1 CRC has been linked to a broad range of histopathological and clinical features, some of which are well established and others novel in nature. This review will discuss these and provide an assessment of their utility for clinical practice. In our experience, the ability to quantify LNM risk in this setting is valuable during multidisciplinary team (MDT) discussions and clinician–patient interactions. Where relevant, the Royal College of Pathologists (RCPath), World Health Organisation (WHO), International Collaboration on Cancer Reporting (ICCR) and Bowel Cancer Screening Programme (BCSP) guidelines will be referenced. The review will not discuss the distinction between epithelial misplacement and pT1 CRC, which is covered elsewhere.13 14

Histopathological risk factors

There is a wealth of literature on this topic, with groups investigating different combinations of proposed risk factors for LNM in pT1 CRC. It is common for factors to be presented in groups that are assigned a risk grade (high, medium, low, etc) allowing the incidence of LNM to be compared between groups. This can make interpreting the impact of individual variables challenging. While it is tempting to compare the relative power of each risk factor in predicting LNM, this varies between studies and types of analysis performed. For this reason, individual risk factors are judiciously discussed here and a ranking from least-to-most important is not absolutely stated. Meta-analyses are best placed to present data such as odds ratios (ORs), a relative measure of effect, for key risk factors for LNM in pT1 CRC (table 4).15–19

Table 4

Meta-analysis of LNM risk in pT1 CRC

The nature of histopathological examination means that microscopic features are commonly more difficult to assess in routine diagnostic practice than published criteria may initially suggest. Factors hampering histopathological examination include sampling variability, tangential cutting and other difficulties with orientation, fragmentation, diathermy artefact and tumour-related issues. While it is likely that a pT1 CRC would be fully embedded into tissue blocks, each examined 4-µm section is a tiny proportion of the whole block thickness (typically 3–4 mm). The examination of deeper levels can often provide useful additional information and is recommended practice.

Lymphovascular invasion

The invasion of lymphatic channels in the submucosa is the primary mechanism behind LNM in pT1 CRC. A recent study of 801 patients who received surgical resection and lymph node dissection for pT1 CRC confirms using a multivariate analysis that lymphatic invasion (LI) is significantly associated with LNM.8 This is seen in other studies, notably in a meta-analysis including 7066 patients where LI was found to be one of the most powerful predictors of LNM, OR 6.91 (95% CI 5.40 to 8.85, p<0.001).16 LI in the absence of other adverse histopathological features is rare.20 These other adverse factors, at least to some degree, are likely proxy measures for the probability of LI and therefore LNM.

Although the invasion of blood vessels is causally linked to distant metastases, there is evidence to support venous invasion (VI), sometimes called vascular invasion, as a predictive factor for LNM.8 15 21 This finding is not unanimous, however.22 LI and VI are often pragmatically grouped under the umbrella of lymphovascular invasion (LVI). Across the literature, LVI, that is, the presence of tumour cells within any endothelial-lined space, is strongly associated with LNM.15 17 19 20 Since the predictive power of VI is weaker than that of LI, grouping the two may be less informative and could lead to underestimation of LNM risk.16 However, identification of VI is still important as a risk factor for distant, blood-borne metastasis and therefore as an adverse prognostic factor for this reason.

Histopathological differentiation between small tumour deposits, lymphatic and venous channels can be challenging. An international concordance study of LI and VI in a cohort of 20 patients with pT1 CRC demonstrated only moderate interobserver agreement between 22 pathologists, using Japanese diagnostic criteria (VI—kappa 0.399, LI—kappa 0.386).23 The lumen and endothelial lining of vessels can be obliterated by tumour. Retraction artefact can mimic LI. The presence of luminal red blood cells may suggest that a vessel is venous in nature, but these can be seen within lymphatic spaces and therefore the identification of either surrounding muscle or elastic lamina is currently recommended by the RCPath when defining VI.24 Data suggest that the standard use of elastic stains increases the detection rate of VI.25 26 Immunohistochemical staining for CD31 and CD34 (endothelial cells) or D2-40 (lymphatic-specific endothelial lining) to aid diagnosis is inconsistent and is therefore not recommended as a routine procedure in all cases. However, these stains may aid the identification of LI and VI in cases where their presence is suspected but not confirmed on H&E staining (figure 1). It is highly likely that the detection of LI and VI is underestimated on histopathological assessment.

Figure 1

Lymphatic invasion by adenocarcinoma within the submucosa. (A) H&E stain. (B) CD31 immunohistochemistry. (C) CD34 immunohistochemistry. (D) D2-40 immunohistochemistry.

Perineural invasion

Perineural invasion (PNI) is widely regarded as having adverse prognostic significance in CRC. In pT1 CRC, PNI is recognised as an independent risk factor for LNM.27–29 In a study of 2697 pT1 CRC resections, a significant association was found between PNI and LNM, OR 6.21 (95% CI 3.50 to 11.0, p<0.001).10 This suggests a role for PNI in metastasis as well as local tumour progression.

Hypotheses surrounding the pathogenesis of PNI attribute the interaction between the nerve microenvironment and tumour.30 Aspects of the nerve microenvironment, such as tumour-associated macrophages and the neurotrophic regulatory protein L1 cell adhesion molecule, are associated with LNM in CRC.31 32 PNI represents a relatively understudied aspect of CRC within research and an entity that, as is with LVI, is likely underestimated in prevalence in diagnostic pathology reports.

Tumour extent

Increasing tumour size is broadly associated with worse prognosis and metastasis. At least one measure of the extent of submucosal invasion is included in most analyses of LNM risk in pT1 CRC. Many methods of standardising the measurement of tumour depth in pT1 cancers have been proposed (figure 2).

Figure 2

Tubular adenoma with extensive epithelial misplacement and small focus of pT1 adenocarcinoma (Haggitt level 1). (A) Low power image demonstrating area of submucosal invasion. (B) High power image demonstrating width (1.6 mm) and depth (0.8 mm) of adenocarcinoma.

The current consensus for the definition of ‘deep submucosal invasion’ (DSI) is ≥1 mm below the muscularis mucosae, Kikuchi level of SM2 or deeper in sessile lesions33 or Haggitt level 4 in pedunculated lesions,34 although this is variable between studies. In practice, accurate microscopic assessment of tumour depth is limited by common pitfalls such as specimen fragmentation, suboptimal orientation and tangential cutting. For example, the Kikuchi level cannot be assessed when the full thickness of the submucosa is not present or the muscularis mucosae is not visible. Novel full-thickness endoscopic resection and local resection methods that include resection of the inner circular layer of the muscularis propria can allow assessment of the Kikuchi level in sessile lesions since the entirety of the submucosa is present. Care should also be taken to only measure true carcinoma, rather than mimics such as epithelial misplacement. Systematic reviews and meta-analyses have widely shown an association between greater depth of invasion and LNM.15 16 19 35 A recent meta-analysis, however, found that in the absence of poor differentiation, LVI and high-grade tumour budding (TB) the absolute risk of LNM in DSI cancers is only 2.6%. Using a multivariate analysis, DSI was not found to be a significant predictor for LNM.19

Using the width of submucosal invasion (WSI) as a measure of tumour size negates some limitations of depth assessment. In a review of eight studies, half found WSI to be a significant predictive factor for LMN. While no consensus ‘cut-off’ value for higher risk of LNM is yet agreed, 2 mm or 3 mm have been suggested as the most discriminatory cut-off value.36 In a study of 207 pT1 CRCs, the association between WSI and the presence of LNM did not reach statistical significance on multivariate analysis, although the width cut-off used was >11.5 mm.37 It is recognised that there is significant interobserver variability in assessing the dimensions of a pT1 CRC and therefore any measure of LNM risk that is based on a particular cut-off value of tumour dimension used in calculation models should be interpreted with caution.38

With digital pathology, the area of submucosal invasion can be estimated on a histological section with increasing ease and accuracy. Although less frequently used as a measure of tumour size in the literature, it has been shown to be a significant predictor of LNM, OR 15.4 (95% CI 1.80 to 132.3, p=0.01) (cut-off area of submucosal invasion>35 mm2).39

Although it is unlikely that a histology report would ever lack a measure of tumour dimension unless this could not be determined, for example, due to fragmentation or poor orientation, the current evidence suggests that the strength of tumour extent as an independent risk factor for predicting LNM is low. One hypothesis is that tumour size is a surrogate marker for LVI when assessing the risk of LNM. As tumours expand, there is an increased likelihood that vessel walls will be disrupted by pushing and infiltrative tumour growth. Interestingly, 3D models of the large bowel vasculature have shown that Kikuchi level sm1 contains significantly more vessels than deeper layers sm2 and sm3. The chance of LVI occurring is likely to reflect a cumulative risk as tumour depth progresses from sm1 to sm3. This perhaps favours the area of submucosal invasion as the most biologically representative measure of tumour size when considering LNM risk.8 It is also possible that associations between larger tumours and LVI reflects tumour biology and mechanisms of LVI such as lymphangiogenesis and the induction of lymphatic permeability.

Tumour grade

Poor differentiation in pT1 CRC, assessed on the degree of glandular formation in the invasive component, has been shown to increase LNM risk.8 10 15 19 21 22 28 37 40 In earlier studies, high tumour grade was regarded as the most important indicator of LNM,20 although this has not been consistently replicated. Poor differentiation has a prevalence of approximately 2%–7% in pT1 CRC,28 although interobserver variation when grading tumours is likely high. Dichotomous grading into either well/moderately differentiated or poorly differentiated adenocarcinoma is recommended by the RCPath to try to reduce variation compared with using a 1–3 or 1–4 scale.24 Almost all of the papers cited above used dichotomous grading; either well/moderate and poor, or low (grades 1–2) and high (grades 3–4). Interestingly, Yamaoka et al 40 also found that moderately differentiated pT1 CRC had statistically higher rates of LNM compared with well-differentiated cancers.40 It should also be noted that although the RCPath, WHO and ICCR now recommend that tumour grading is based on the least differentiated rather than the predominant component,24 41 42 this may not be reflected in older studies.

Histological subtypes of adenocarcinoma

While the majority of colorectal adenocarcinomas have no specific subtype, histological subtypes associated with poor prognosis include mucinous carcinoma (MAC), signet ring cell carcinoma (SRC); each diagnosed when the percentage of the respective tumour component represents >50%; and micropapillary adenocarcinoma, for which at least 5% of the tumour must show this pattern. Multiple further subtypes are listed in the current WHO classification, but the proportion of a tumour that is required to show the characteristic features in order to diagnose these is not always defined.41 In many studies of pT1 CRC, adverse subtypes are commonly grouped with poor differentiation, given their relatively low prevalence.8 10 20 40 However, MAC has been shown to have significantly higher LNM risk than patients with conventional adenocarcinoma by Ebbehoj et al, 21 OR 2.19 (95% CI 1.01 to 4.76, p=0.049) and Guo et al, 43 OR 2.19 (95% CI 1.70 to 2.80, p<0.001), although this was not found by Barel et al.28 The presence of MAC or SRC is associated with higher rates of LNM than adenocarcinoma, with SRC having a higher rate than MAC. However, on multivariate analysis, histological subtype was not found to be an independent prognostic factor for survival.44 Micropapillary subtype in pT1 CRC is observed to correlate with more frequent LNM than adenocarcinoma,45 46 although case numbers are low, and this finding is not unanimous across studies.47

Pedunculated versus non-pedunculated morphology

The morphology of pT1 CRC may affect the impact of prognostic factors. For example, a Danish study of 1167 patients with pT1 CRC found that tumour budding (TB) (see below) was only an adverse factor in pedunculated tumours, while age (see below) was only prognostic in non-pedunculated tumours. Tumour morphology itself was not significantly associated with LNM as an independent factor in this study.21 However, another group found that pedunculated tumours were less likely to be associated with LNM than sessile tumours.22

Tumour budding and poorly differentiated clusters

TB, that is, single or clusters of up to four neoplastic cells in the tumour stroma at the advancing tumour front, is recognised as a poor prognostic factor in CRC. Buds do not imply poor differentiation but are graded in their own right on the number of buds per high power field (Bd1 ≤5 buds, Bd2=5–9 buds, Bd3 = ≥10 buds) on recommendation by the International Tumour Budding Consensus Conference.47 In pT1 CRC, there is a significant correlation between high grade (Bd2-3) TB and LNM on meta-analysis, OR 2.83 (95% CI 2.06 to 3.88).19 Budding is strongly associated with LVI in pT1 cancers.17 It is hypothesised that TB is a visual representation of the epithelial–mesenchymal transition, a mechanism where neoplastic cells acquire the motility to dissociate from the tumour mass and invade the extracellular matrix, which precedes the process of VI.20

Assessment of TB is limited in areas of active inflammation, gland rupture and mucinous differentiation, which should be avoided. The use of cytokeratin immunohistochemistry when grading TB is controversial. A study of 163 patients with pT1 CRC found fair interobserver agreement between two pathologists (kappa 0.235) for TB quantification using H&E-stained slides alone but very good agreement (kappa 0.842) using cytokeratin immunohistochemistry.28 The RCPath recommends that TB assessment is performed on H&E for all forms of local excision specimen for pT1 CRC, although cytokeratins can be used to help identify the area of highest budding activity (figure 3).24

Figure 3

Tumour budding by adenocarcinoma. The tumour buds (1–4 cells in size) are present within the centre of the field in both images. The cytokeratin immunohistochemistry preparation also highlights two central islands of tumour cells that are better classified as poorly differentiated clusters. (A) H&E stain. (B) CK8/18 immunohistochemistry.

In contrast, a poorly differentiated cluster (PDC) consists of five or more tumour cells at the advancing tumour front, lacking glandular formation. There is no consensus on the maximum size of a PDC. There is emerging evidence of a significant association between PDC and LNM, although the studies are small in number, variable in sample size, design and PDC grading.48 Examining additional levels may aid the distinction between TB and PDC. The optimum number of levels required for this purpose will vary between cases, but as a general rule, we would perform three extra levels in the first instance.

Proximity to surgical margin

The proximity of adenocarcinoma to the surgical margin is an adverse prognostic factor in pT1 CRC, both for the presence of residual disease and for LNM.49 The definition of a ‘positive’ margin in this setting varies between studies, from the presence of tumour directly at the margin (ie, a clearance of 0 mm), up to a distance of 2 mm. The RCPath, ICCR and BCSP guidelines all define a positive margin in pT1 CRC as the presence of tumour less than 1 mm from the margin.24 42 50 A 1 mm cut-off for the definition of a positive margin in CRC was used by the Medical Research Council (MRC) CRO7 trial and has since been adopted for routine diagnostic practice.51 In a retrospective review of 162 cases of pT1 CRC with no other well-characterised risk factors for LNM (ie, poor differentiation, LI/VI or TB), where resection was performed, a 0 mm margin was associated with a 22% risk (4/18 cases) of LNM, while a margin greater than 0 mm was associated with a 3% risk (1/37 cases) of LNM.49 This suggests that the requirement for a 1 mm margin in pT1 CRC may be conservative.

Measurement of the proximity of a pT1 CRC to the surgical margin can be challenging or impossible, for example, if the specimen is fragmented or poorly orientated. Diathermy artefact and associated tissue retraction at the margin can further impair this assessment. If adenocarcinoma extends into the diathermy zone and precludes formal measurement of the distance from the margin, the latter is best assessed as positive.24 50 Tumour-related factors such as attenuation/loss of the muscularis mucosae and the juxtaposition of adenomatous and invasive components can also lead to difficulties in assessment since the boundary of the invasive tumour may be difficult to define. The ability to identify an ‘intramucosal’ (ie, non-submucosal-invasive) component within a lesion can help to avoid overestimation of the depth of a pT1 CRC.52

Immune response

A chronic inflammatory reaction may be seen in CRC and is particularly associated with tumours that are mismatch repair deficient (dMMR).53 Tumours exhibiting dMMR (eg, in patients with Lynch syndrome) tend to possess a higher tumour mutational burden, leading to the creation of tumour neo-antigens that can be recognised by immune cells and lead to immune-mediated tumour cell destruction.53 In pT1 CRC, dMMR is associated with the presence of tumour-infiltrating lymphocytes (TILs) and a lower incidence of LNM compared with mismatch proficient tumours.22 Similarly, greater numbers of CD8+ TILs, specifically in the tumour centre, are associated with lower rates of LNM.54 However, a similar association was not identified in a study of CD3+ TILs.17

PD-L1 expression may affect LNM risk in pT1 CRC. Multivariate analysis has demonstrated that higher PD-L1 expression, for example, a Combined Positivity Score (CPS) of>1.2 or an Immune Cell Score (ICS) of >1.3% was associated with lower LNM risk in these tumours, with ORs of −2.5 (95% CI −4.11 to 0.97) for the CPS method and −1.85 (95% CI −2.90 to 0.79) for the ICS method.9 The relationship, if any, between PD-L1 expression and factors such as the pattern of immune cell infiltration and MMR status in pT1 CRC is currently unknown. Interestingly, PD-L1 expression in this study was limited to immune cells, with no expression by tumour cells.

Tumour stroma

The nature of tumour-associated stroma may modulate disease progression and other factors, for example, the nature of the immune response. In pT1 CRC, a greater proportion of tumour stroma has been associated with an increased risk of LNM and this may correlate with Consensus Molecular Subtype (CMS) 4.37 CMS grouping are a recently described subdivision of CRC into four categories, based on genomic analysis. CMS4 has been associated with mutations in fibrogenesis and epithelial–mesenchymal transition pathways and has been associated with a worse prognosis.55 The desmoplastic stromal reaction to CRC can be divided into inflammatory, fibrotic or myxoid stroma. Immature, myxoid stroma at the tumour periphery is associated with absent TILs, a high degree of TB and, in pT1 CRC, LNM.56–58

A ‘deep learning’ (a technology within artificial intelligence (AI) that can identify subtle patterns in complex data) study has suggested inflammation within adipose tissue in pT1 CRC may be associated with a higher incidence of LNM.59 The peri-tumoural stroma score is another AI-generated parameter predictive of LNM but has not yet been validated in a pT1 cohort.60

Tools for risk calculation

Tools to aid LNM risk prediction in pT1 CRC include an online risk calculator created by the Dutch T1 CRC working group.69 Several nomograms have also been developed (table 5).43 58 61 63 64 70–72 The heterogeneity between predictive tools is striking, including the range of risk factors that are included in each model. Although most tools include a measure of tumour grade or differentiation, this is assessed differently between authors.43 58 61 63 64 69–71 Similarly, invasive tumour depth and TB are assessed through a mixture of different semiquantitative scoring systems and continuous scales.43 61 63 64 69–71 The concordance of each model with real-life patient data in validation sets presents an estimate of the predictive accuracy of the tool.

Table 5

Nomograms for LNM prediction in pT1 CRC

The Dutch online risk calculator is the only tool discussed here that was validated using a primarily Caucasian population, identifying patients who required surgery with an area under the curve value of 0.830, 95% CI 0.76 to 0.90.69 73 We cautiously recommend this tool in light of several methodological considerations. To date, the model has only been validated on a small cohort of cases of pedunculated pT1 CRC, of which only 23 patients had LNM detected during follow-up.69 There are no current guidelines to standardise the histopathological assessment of PDCs. Mimics of carcinoma such as epithelial misplacement represent a particular challenge when assessing the status of the muscularis mucosa. Nonetheless, the calculator shares all other editable fields with at least four other published and validated predictive tools for LNM in pT1 CRC (table 5).43 58 61 63 64 70–72

A nomogram recently published by Kajiwara et al comes from a large Japanese study, N=4673, and has an acceptable concordance statistic, 0.790; 95% CI 0.751 to 0.829. This nomogram includes measures of tumour depth, tumour grade, LVI and TB akin to the Dutch group’s calculator but also gender and tumour location within the colon.63

Notably, the two Chinese studies that tested nomograms against the largest validation sets had the lowest concordance.43 61 It would be of significant interest to validate all published nomograms and calculation tools for LNM in pT1 CRC using substantial and representative international cohorts.