SEARCH

SEARCH BY CITATION

Keywords:

  • colorectal cancer;
  • prognosis;
  • tumor border configuration;
  • staging

Abstract

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. Conflict of Interest Disclosures
  8. References

BACKGROUND:

Reproducible and well characterized prognostic histomorphologic criteria added to current pathological staging could have an immediate effect on refining prognosis in colorectal cancer. The aim of this study was to determine the additive effect of tumor border configuration and peritumoral lymphocytic infiltration on the selection of patients for adjuvant therapy classified by TNM.

METHODS:

A total of 1420 primary colorectal cancers with complete clinicopathological data from multiple treatment centers were analyzed. The prognostic effect of tumor border configuration (pushing or infiltrating) and peritumoral lymphocytic infiltration was assessed, validated by resampling of the data, and compared with TNM staging. All P values were 2-sided.

RESULTS:

Multivariate analysis confirmed the adverse prognostic value of the tumor border configuration (P < .001), but not of peritumoral lymphocytic infiltration. The addition of tumor border configuration to T and N category identified 2 major prognostic subgroups (relative risk of death, 4.75; 95% confidence interval [CI], 2.53-8.94). Moreover, stage II patients with a pushing border had a 5-year survival rate of 82.1% (95% CI, 71.8%-90.3%), whereas an infiltrating border resulted in a significantly more adverse outcome (5-year survival rate, 62.7%; 95% CI, 48.0-76.2%), closely resembling that of stage III patients. Similar results were obtained after adjusting for adjuvant therapy (P < .001).

CONCLUSIONS:

The classification of patients into prognostic subgroups is improved with the addition of tumor border configuration to TNM stage. In particular, patients with stage II disease characterized by an infiltrating tumor border have poor clinical outcome and represent a subset of lymph node-negative patients who could be considered for adjuvant therapy. Cancer 2009. © 2009 American Cancer Society.

Strategies for improving the TNM staging system for patients with colorectal cancer is a subject of intense debate.1 Although the detection of prognostic molecular and immunohistochemical markers capable of identifying high-risk patients constitutes a promising area of research, no biomarkers are currently recommended as prognostic factors for patients with colorectal cancer by the American Society of Clinical Oncology or by the European Group on Tumor Markers.2, 3 Reproducible and well characterized prognostic histomorphologic criteria added to current pathological staging would have a more immediate effect on refining prognostic subgroups, which could be complemented by biomarkers in the future.

In 1987, Jass and coworkers proposed a new prognostic classification system for rectal cancers based on 4 histomorphologic parameters.4 By evaluating the primary tumor extent (T category), lymph node metastasis (N category), tumor border configuration (pushing or infiltrating) and the presence of conspicuous peritumoral lymphocytic (PTL) infiltration, the Jass classification was shown to significantly improve the prognostication of patients, particularly for those with Dukes B and C tumors. In the interval after its proposal, the Jass staging system has been discussed, revisited, validated, and also challenged,5-10 yet neither of its more distinctive features, tumor border configuration and PTL infiltration, have been integrated into standard practice, despite the body of evidence supporting their prognostic value.11-13

The aim of this study was to determine the prognostic value of the tumor border configuration and PTL infiltration and to investigate to what extent these 2 features could improve on the current TNM staging system to better classify patients into prognostic subgroups at the time of resection. To this end, we used 2 different groups of colorectal cancer patients from multiple treatment centers with complete clinicopathological data (n = 931 and 417, respectively).

MATERIALS AND METHODS

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. Conflict of Interest Disclosures
  8. References

Patient Collective

A total of 1420 primary colorectal cancer resections were retrospectively collected from 1987 to 1996. Pathology was systematically reevaluated by 1 pathologist in all cases. The patient collective was derived from multiple centers and included the Institute of Pathology, University Hospital of Basel, Switzerland; the Institute of Clinical Pathology, Basel, Switzerland; and the Institute of Pathology, Stadtspital Triemli, Zurich, Switzerland. Missing clinical information was assumed to be at random. All patients were preoperatively untreated. On average, 5 tumor blocks were evaluated per case. The use of tissue was approved by the local ethics committee.

Evaluation of Tumor Border Configuration and PTL Infiltration

Tumor border configuration (Fig. 1), and PTL infiltration at the invasive tumor front were diagnosed according to the method proposed by Jass at low magnification.4, 13 Briefly, tumor margins were identified as infiltrating when there was no recognizable margin of growth and a “streaming dissection” between normal structures of the bowel wall was present. Tumors with focal infiltrating growth patterns were considered infiltrating. Margins were considered pushing when they were reasonably well circumscribed (or expanding) and often associated with a well developed inflammatory lamina. Peritumoral lymphocytic infiltration was regarded as present when there was a distinctive connective tissue mantle cap at the invasive margin in which lymphocytes and other inflammatory cells were scattered.

thumbnail image

Figure 1. Cytokeratin 22 staining highlights different tumor border configurations in colorectal cancer (original magnification, ×5): infiltrating tumor border (A) and pushing tumor border (B).

Download figure to PowerPoint

Test Groups

Table 1 summarizes the clinicopathological information for the 931 patients included in test Group 1. Because M stage was not available for these patients, test Group 1 was used to evaluate the prognostic performance of the tumor border configuration and PTL infiltration in univariate and multivariate analysis. A total of 417 patients with complete TNM stage and follow-up information were included in test Group 2 (Table 2). This subgroup of patients was used to evaluate the additional prognostic contribution of the tumor border configuration and/or PTL infiltration to TNM stage.

Table 1. Clinicopathological Characteristics of Patients in Test Group 1 (n = 931)
CharacteristicsFrequency, No. (%)
  1. SD indicates standard deviation.

Age, y, mean±SD69±11.3
Sex 
 Women498 (53.5)
 Men433 (46.5)
T category 
 pT147 (5.1)
 pT2125 (13.7)
 pT3606 (66.4)
 pT4135 (14.8)
N category 
 pN0480 (53.0)
 pN1237 (26.2)
 pN2189 (20.8)
Tumor grade 
 G127 (3.0)
 G2823 (90.2)
 G362 (6.8)
Vascular invasion 
 Absence656 (71.8)
 Presence257 (28.2)
Tumor border configuration 
 Pushing276 (30.3)
 Infiltrating635 (69.7)
Mismatch repair status 
 Proficient810 (87.0)
 Deficient121 (13.0)
Peritumoral lymphocytic infiltration 
 Absence702 (76.9)
 Presence211 (23.1)
Tumor location 
 Left-sided275 (29.6)
 Right-sided289 (31.1)
 Rectum364 (39.2)
Histologic subtype 
 Adenocarcinoma852 (88.0)
 Medullary5 (0.5)
 Mucinous59 (6.1)
 Other2 (5.4)
Survival rate 
 5-y %54.2
 No. of deaths407
Table 2. Clinicopathological Characteristics of Patients in Test Group 2 (n = 417)
CharacteristicsFrequency, No. (%)
  1. SD indicates standard deviation.

Age, y, mean±SD71.2±11.4
Sex 
 Women201 (48.2)
 Men216 (51.8)
T category 
 pT19 (2.2)
 pT267 (16.1)
 pT3265 (63.4)
 pT476 (18.2)
N category 
 pN0217 (52.0)
 pN1106 (25.4)
 pN294 (22.5)
Tumor grade 
 G12 (0.5)
 G2310 (74.3)
 G3105 (25.2)
Vascular invasion 
 Absence307 (73.8)
 Presence109 (26.2)
Tumor border configuration 
 Pushing209 (50.1)
 Infiltrating208 (49.9)
Mismatch repair status 
 Proficient328 (78.7)
 Deficient89 (21.3)
Peritumoral lymphocytic infiltration 
 Absence343 (82.3)
 Presence74 (17.8)
Distant metastasis 
 Absence346 (83.0)
 Presence71 (17.0)
Local recurrence 
 Absence248 (60.0)
 Presence165 (40.0)
Tumor location 
 Left-sided177 (43.5)
 Right-sided136 (33.4)
 Rectum94 (23.1)
Histologic subtype 
 Adenocarcinoma359 (86.1)
 Medullary0 (0.0)
 Mucinous56 (13.4)
 Other0 (0.5)
Postoperative therapy 
 Untreated324 (78.3)
 Chemotherapy77 (18.5)
 Radiotherapy7 (1.7)
 Chemotherapy/radiotherapy9 (2.2)
Survival time 
 5-y %62.8
 No. of deaths284

Statistical Analysis

Survival analysis on test Group 1 was performed using Cox proportional hazards regression in univariate analysis and subsequently in a multivariate setting with an automated forward selection procedure. Variables entered into the model included age, sex, tumor location, tumor grade, vascular invasion, mismatch repair status, histologic subtype, T category, N category, tumor border configuration, and PTL infiltration. The assumption of proportional hazards was verified before each Cox regression analysis, and sample size calculations confirmed an appropriately powered test group. To determine the reliability of our prognostic models, the forward selection procedure was carried out in conjunction with resampling of the dataset with replacement by bootstrapping the data 1000 times.14 This process resulted in 1000 resamples of the original dataset. We determined the number of times T category, N category, tumor border configuration, and PTL infiltration were selected as independent prognostic factors over the 1000 resamples, adjusting for the prognostic effects of other variables in the analysis. In addition, the most frequently selected prognostic model over all 1000 resamples was identified. The relative risk of death, or hazard ratio (HR), and 95% confidence interval (CI) were used to determine the prognostic effect of each feature in the final prognostic model, in addition to P values. For our 4 features of interest, the baseline hazard of 1.0 was attributed to the more favorable outcome for each feature, namely earlier T and N category, a pushing tumor margin, and presence of PTL infiltration. The prognostic factors of the final model were combined in test Group 2. Kaplan-Meier survival curves were plotted using the guidance of Pocock et al, and 5-year survival rates and 95% CI were analyzed by the product-limit method.15P values < .05 were considered statistically significant (2-sided). The goodness-of-fit of different prognostic models was compared by the Akaike Information Criterion (AIC) and the Schwarz Bayesian Criterion (SBC). The better the model fit to the data, the smaller the values of both AIC and SBC. The inter- and intraobserver variability was evaluated with the kappa statistic (k).

RESULTS

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. Conflict of Interest Disclosures
  8. References

Test Group 1

Univariate survival analysis

An infiltrating tumor border configuration led to a 5-year survival rate of 41.9% (95% CI, 33%-50%), compared with 81.2% (95% CI, 74%-87%) for patients with a pushing tumor margin (P < .001). Absence of PTL infiltration resulted in adverse prognosis compared with the presence of PTL infiltration (5-year survival rate 60.1% [95% CI, 53%-66%] and 72.8% [95% CI, 59%-83%], respectively) (P = .019).

Selection of most reproducible prognostic features in multivariate analysis

In multivariate survival analysis, tumor border configuration (P < .001) but not PTL infiltration was found to be an independent prognostic factor after adjusting for age, sex, T category, N category, tumor grade, tumor location, mismatch repair status, and vascular invasion. The reproducibility of the prognostic effects of T category, N category, tumor border configuration, and PTL infiltration are observed in Figure 2 (Top). After resampling of the data 1000 times , T category, N category, and tumor border configuration were selected 1000, 996, and 918 times, respectively, as independent prognostic factors, whereas PTL infiltration contributed additional information in only 49 of 1000 cases.

thumbnail image

Figure 2. After resampling of the data and multivariate Cox proportional hazards regression analysis, T category, N category, and tumor border configuration had the most reproducible prognostic effects, whereas peritumoral lymphocytic (PTL) infiltration was selected few times as an independent prognostic factor (Top). The model containing T, N, and tumor border configuration was identified as having the most reproducible prognostic effects (Bottom).

Download figure to PowerPoint

Selection of the most reliable prognostic model

Combinations of T and N categories, tumor border configuration, and PTL infiltration were investigated in all 1000 resamples of the data. The multifeature combination of T and N categories along with tumor border configuration was observed in 880 of 1000 resamples, clearly underlining the contribution of these 3 combined features to prognosis (Fig. 2, Bottom). Together, more advanced T category (P < .001; HR, 2.24; 95% CI, 1.6-3.1), N category (P < .001; HR, 2.47; 95% CI, 2.0-2.9), and an infiltrating tumor border configuration (P < .001; HR, 1.53; 95% CI, 1.2-1.9) had a significant negative impact on survival time.

Rectal cancer

Similar results were obtained when focusing on tumors located in the rectum, with more advanced T category (P < .001; HR, 2.46; 95% CI, 1.6-3.8), N category (P < .001; HR, 1.92; 95% CI, 1.4-2.6), and an infiltrating margin (P = .007; HR, 1.65; 95% CI, 1.1-2.4) highly linked to unfavorable prognosis.

Test Group 2

Survival analysis

In test Group 2, 71 patients were found to have stage IV disease. Because adjuvant treatment in stage IV patients is typically discussed in the case of R0-resection after metastases, these patients were not included further in the analysis. For the remaining 346 patients, T category (T1 or T2 vs T3 or T4), N category (node-negative, N0 vs any number of positive lymph nodes, N+), and tumor border configuration were categorized into their 8 possible combinations of features. Four distinct prognostic subgroups (PG1 to PG4) could be identified based on the similarity of their 5-year survival rates (Table 3). PG1 was comprised of patients with T1N0 or T2N0 (TNM stage I) tumors with a 5-year survival rate of 100%. Tumor border configuration had no effect on prognosis in these cases. PG2 was comprised of a subgroup of TNM stage III patients, namely those with T1N+ or T2N+ tumors with a pushing tumor margin and a 5-year survival rate of 83.3% (95% CI, 46.5%-99%). This group also included a subset of TNM stage II patients, namely T3N0 or T4N0 patients with a pushing tumor margin, demonstrating a survival rate of 82.1% (95% CI, 71.8%-90.3%). PG3 patients had a considerably lower survival rate, and included stage II patients with an infiltrating tumor margin (5-year survival rate, 62.7%; 95% CI, 48.0%-76.2%), and those with stage III disease and pushing tumor border (5-year survival rate, 56.7%; 95% CI, 38.0%-74.0%). Finally, the most adverse outcome was noted for PG4, which encompassed stage III patients with an infiltrating tumor border, irrespective of T status.

Table 3. Five-year Survival Rates for Patients in Test Group 2 With Combinations of T Category, N Category, and Tumor Border Configuration
No.T CategoryN CategoryTumor Border Configuration5-Year Survival Rate (95% CI)PGTG
  1. CI indicates confidence interval for the 5-year survival rate; PG, prognostic group; TG, treatment group; N+, any number of positive lymph nodes.

6T1, 2N0Infiltrating100%PG1TG1
53T1, 2N0Pushing100%
8T1, 2N+Pushing83.3% (46.5-99)PG2
93T3, 4N0Pushing82.1% (71.8-90.3)
54T3, 4N0Infiltrating62.7% (48.0-76.2)PG3TG2
42T3, 4N+Pushing56.7% (38.5-74.0)
84T3, 4N+Infiltrating29.9% (19.8-40.9)PG4
6T1, 2N+Infiltrating22.9% (1-66.8)

These 4 prognostic subgroups could be further classified based on favorable or unfavorable outcome into 2 major subgroups of patients. Together, PG1 and PG2, with their prolonged survival time, were defined in this study as a “nonadjuvant treatment group” (Treatment Group TG1), and those in PG3 or PG4 with considerably more adverse prognosis were considered here as an “adjuvant treatment group” (Treatment Group TG2).

Comparison of TG1/TG2 with TNM stage I and II/III

By using TG1 as a baseline, the relative risk of death in patients with TG2 was 4.75 (95% CI, 2.53-8.94) (Fig. 3). This effect on survival was contrasted to that of TNM stages III versus TNM I and II, the former typically considered to indicate potential candidates for adjuvant therapy.16 The 5-year survival rate for stage I and II disease was 82.5% (95% CI, 77.1%-89.2%), whereas for stage III it was 56.9% (95% CI, 45.5%-68%). By using stage I and II as a baseline, the relative risk of death in patients with stage III disease was 3.04 (95% CI, 1.83-5.1) (Fig. 4). The considerably larger relative risk of death for patients in TG2 compared with TG1 versus those in TNM stage III compared with I or II suggests an overall improvement in prognostication with the addition of tumor border configuration.

thumbnail image

Figure 3. Kaplan-Meier survival curves for patients with stage II and stage III colorectal cancer are stratified by tumor border configuration.

Download figure to PowerPoint

thumbnail image

Figure 4. Kaplan-Meier survival curves represent patients in treatment groups (TG1 and TG2) derived from T category, N category, and tumor border configuration compared with patients with TNM stage I /II or stage III disease.

Download figure to PowerPoint

To further validate these findings, the AIC and SBC were used as measures of goodness-of-fit to compare the 2 competing prognostic models, the first with TNM stage and the second with tumor border configuration in addition to TNM stage. For TNM stage only, the AIC and SBC were 635.0 and 637.1, respectively, whereas for the TG1/TG2 model they were 623.7 and 625.8, respectively. The lower values of both AIC and SBC for TG1 of 2 suggest an improved model fit compared with TNM stage.

Adjustment for postoperative treatment

Of the 21% of patients in Test Group 2 who had received adjuvant therapy, 63% were patients with T3N+ or T4N+ tumors (stage IIIB). Adjuvant therapy was not significantly linked to outcome (P = .0721; HR, 1.63; 95% CI, 0.95-2.8]); however, Treatment Group TG1 of 2 retained its significant effect on prognosis after adjustment for treatment effect (P < .0001; HR, 4.0; 95% CI, 2.05-7.78).

Inter- and intraobserver variability of the tumor border configuration

To determine the reproducibility of the tumor border configuration, 100 cases were randomly selected and re-evaluated by a second experienced pathologist. The percent concordance between observers was 85% (k = 0.63; 95% CI, 0.45-0.8; P < .001). Concordance for the same observer was 91% with a kappa value of k = 0.76 (95% CI, 0.63-0.9; P < .001).

DISCUSSION

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. Conflict of Interest Disclosures
  8. References

The findings of this study highlight a substantial improvement in the prognostic classification of colorectal cancer patients with the addition of tumor border configuration to TNM stage, a result with particular implications for patients with stage II disease.

The treatment of stage II colorectal cancer patients with adjuvant therapy remains a matter of controversy.17-19 The identification of high-risk patients with worse outcome among those with stage II disease would have a considerable impact on the selection of patients most likely to benefit from adjuvant treatment. Our results show a significant negative effect of tumor border configuration on 5-year survival in patients with stage II disease, decreasing from 80% for those with a pushing border to 62.7% in those with an infiltrating tumor growth pattern. Prognosis of patients with stage IIIA disease was also determined to be dependent on tumor border configuration, with 5-year survival rates considerably improved from 23% to 83.3% for those with infiltrating and pushing growth patterns, respectively, and independent of postoperative therapy. Although the sample size of these particular stage IIIA groupings is likely too small to draw definite conclusions from, these results suggest that despite lymph node positivity, patients with a pushing tumor margin experience prolonged survival, and could perhaps be reconsidered from undergoing adjuvant therapy. Taken together, the assessment of the tumor border configuration could reclassify 17.9% of patients in this study, namely 8 with stage IIIA and 54 with stage II, into more appropriate prognostic subgroups.

One of the limitations of our study is the relatively small number of patients in each of the prognostic groupings once they are subdivided by the tumor border configuration, particularly for patients with an infiltrating margin who have early T1N0 or T2N0 tumors. Second, survival time in patients with rectal cancer is known to be linked to surgical procedure. Because our rectal cancer specimens were collected from 1987 to 1996, many of the patients included here predate the total mesorectal excision.20 Our findings suggest that the infiltrating growth pattern remains a significant independent prognostic factor in our subgroup of preoperatively untreated rectal cancers. Tumor growth pattern cannot be deduced either from biopsy, or by imaging modalities, and one can only speculate as to the changes in this feature induced by preoperative therapy. It may be possible to identify protein markers, such as E-cadherin, which are strongly linked to the infiltrating margin and can be applied on biopsies to predict the tumor border configuration.21

Although it is generally agreed by pathologists that the assessment of PTL infiltration at the invasive tumor margin can be a subjective and often difficult assessment, the diagnosis of an infiltrating or pushing tumor margin at low magnification is reproducible and can be performed with ease. Jass et al reported an overall agreement of 85% in their analysis of more than 400 rectal cancers.13 Compton proposed guidelines for the assessment of tumor border configuration and underlined an interobserver agreement of 90%.22 In this study, similar percent concordance was achieved, namely 85% for interobserver and 91% for intraobserver agreement, underlining the reproducibility and ease with which this feature can be interpreted.

Our results will require not only validation by other research groups but also confirmation in a prospective cohort study. However, these preliminary findings indicate that the prognostic power of the tumor border configuration is substantial and appears to play a critical role in determining outcome. As suggested by Jass et al in 1987, classification of patients into prognostic subgroups at the time of resection is considerably improved by the addition of this feature to TNM stage.4 These findings may have important implications in the selection of patients for postoperative therapy. The tumor border configuration is a well accepted, simple, and reproducible histomorphologic feature, and should consistently be reported by pathologists, because it does not complicate standard reporting of TNM stage yet improves the prognostic classification of colorectal cancer patients, in particular those with stage II disease.

Acknowledgements

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. Conflict of Interest Disclosures
  8. References

We thank Prof. Dr. Jeremy Jass, whose unwavering support and interest has inspired this work.

References

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. Conflict of Interest Disclosures
  8. References