Abstract
Purpose:
There is a need to improve current risk stratification of stage II colorectal cancer (CRC) to better inform risk of recurrence and guide adjuvant chemotherapy. We sought to examine whether integration of QuantCRC, a digital pathology biomarker utilizing hematoxylin and eosin-stained slides, provides improved risk stratification over current American Society of Clinical Oncology (ASCO) guidelines.
Experimental Design:
ASCO and QuantCRC-integrated schemes were applied to a cohort of 398 mismatch repair proficient (MMRP) stage II CRCs from three large academic medical centers. The ASCO stage II scheme was taken from recent guidelines. The QuantCRC-integrated scheme utilized pT3 vs. pT4 and a QuantCRC-derived risk classification. Evaluation of recurrence free survival (RFS) according to these risk schemes was compared using the log-rank test and hazard ratios.
Results:
Integration of QuantCRC provides improved risk stratification compared to the ASCO scheme for stage II MMRP CRCs. The QuantCRC-integrated scheme placed more stage II tumors in the low-risk group compared to the ASCO scheme (62.5% vs. 42.2%) without compromising excellent 3-year RFS. The QuantCRC-integrated scheme provided larger hazard ratios (HR) for both intermediate-risk (2.27, 95%CI 1.32–3.91, P=0.003) and high-risk (3.27, 95%CI 1.42–7.55, P=0.006) groups compared to ASCO intermediate-risk (1.58, 95%CI 0.87–2.87, P=0.1) and high-risk (2.24, 95%CI 1.09–4.62, P=0.03) groups. The QuantCRC-integrated risk groups remained prognostic in the subgroup of patients that did not receive any adjuvant chemotherapy.
Conclusions:
Incorporation of QuantCRC into risk stratification provides a powerful predictor of RFS that has potential to guide subsequent treatment and surveillance for stage II MMRP CRCs.
Keywords: colorectal cancer, digital pathology, adjuvant chemotherapy, QuantCRC, recurrence free survival
INTRODUCTION
Pathologic variables derived from resection specimens are currently used to guide selection of adjuvant chemotherapy (ACT) in colorectal carcinoma (CRC) (1,2). Patients with stage I disease are not offered ACT given the high rate of cure with surgery alone. Patients with stage III disease are offered ACT given that approximately 40% will develop tumor recurrence (3). In stage III tumors, the length of ACT is determined by pT4 and number of positive lymph nodes (1,4). In stage II CRCs, 5-year recurrence rates range from 17–32% and determining which patients would benefit from adjuvant therapy remains a clinical challenge (5). The morbidity and financial costs associated with chemotherapy underscores the need for robust risk stratification in this group of tumors to identify those patients that would receive the most benefit from ACT (4,6,7).
The American Society of Clinical Oncology (ASCO) recently updated ACT guidelines for stage II CRC based on pathologic features that have been shown to predict recurrence (2). It is now well established that patients with stage II mismatch repair deficient (MMRD) tumors should not be offered ACT (8–10). According to ASCO guidelines, stage II mismatch repair proficient (MMRP) tumors with pT4 stage should be offered fluoropyrimidine-based ACT as these tumors are considered to have the highest risk of recurrence (3). In one study, the 5-year survival rate for pT4 tumors receiving ACT was 70.9% compared to 59.8% for those not receiving ACT (11). Tumors with venous, lymphatic, or perineural invasion (VELIPI), high tumor budding (Bd3), obstruction, or poor tumor differentiation have an increased risk of recurrence in most but not all studies (3,12–15). Patients with tumors that exhibit one or more of these features may be offered fluoropyrimidine-based ACT due to an intermediate risk of recurrence (2). Finally, ACT is not recommended for stage II MMRP tumors without the above adverse pathologic features given the low risk of recurrence in this group (2).
However, there is increasing recognition that this risk stratification scheme is suboptimal. The need to improve risk stratification is illustrated by several recent studies proposing alternate schemes by incorporating digital biomarkers (16) and circulating tumor (ct) DNA analysis (17). The recently published DYNAMIC study used ctDNA to direct ACT in stage II CRCs and demonstrated reduced rates of chemotherapy with no impact on 2-year recurrence free survival (RFS). However, ctDNA is expensive to perform and time consuming, which resulted in treatment delays in the DYNAMIC trial (17,18). Furthermore, in their post hoc analysis, a negative ctDNA result was less predictive of RFS for pT4 as compared to pT3 CRCs. Commercial molecular tests such as Oncotype DX colon (19) are available to risk stratify stage II tumors; however, the impact of such scores on clinical practice is limited (20). Immunohistochemistry has also been used to improve risk stratification, most notably by measuring CD3 and CD8 T-cell infiltration using the immunoscore (21–23). More recently, an AI-based immunoscore has been developed using a panel of immunohistochemical stains including CD4, CD8, CD68, and CD20 (24). However, these schemes require additional immunohistochemical stains and have not been widely adopted in the United States (25). Recently, Kleppe et al incorporated a digital biomarker, DoMore-v1-CRC, using hematoxylin and eosin slides (H&E) to better risk stratify stage II CRC illustrating the potential of AI-integrated risk schemes (16).
We recently developed a prognostic digital biomarker, QuantCRC, based on analysis of hematoxylin and eosin (H&E) images of 2,411 CRCs from the Colon Cancer Family Registry (26,27). QuantCRC segments a digital image of CRC and extracts 15 features. The 15 QuantCRC features can be combined with stage and MMR status to estimate risk of recurrence. In this study, we compared the ASCO risk stratification scheme to a novel integrated risk scheme for stage II MMRP CRC that incorporates QuantCRC risk classification in a validation cohort of 398 stage II MMRP CRCs from three institutions that have not been used to train the QuantCRC segmentation algorithm or the subsequent prognostic model. The goal was to improve upon current risk stratification to better predict RFS using readily available and inexpensive H&E slides.
MATERIALS AND METHODS
Study Populations
The validation cohort consisted of consecutively resected stage II (pT3N0 or pT4N0) CRCs from the University of Pittsburgh Medical Center (UPMC) between 2010–2015, Mount Sinai Hospital Toronto between 2011–2016, and Mayo Clinic Arizona and Rochester between 2017–2020 (Table 1). Additional inclusion criteria included age > 18 years, available H&E slides, and known MMRP. MMR immunohistochemistry was used to determine retained expression of all four MMR proteins (MLH1, MSH2, MSH6, and PMS2) as per institutional protocols. Neoadjuvant treated tumors were excluded. The following data were extracted from the pathology report: age, sex, tumor location, pT-stage, lymph node count, histologic grade, and venous/lymphatic/perineural invasion (VELIPI). Re-review of the tumor was performed to obtain grade of tumor budding and were categorized according to the international tumor budding consensus recommendations (28). Presence of bowel obstruction, pre-operative carcinoembryonic antigen (CEA) levels, recurrence, and chemotherapy data were assessed by review of the medical records at the respective institutions. The study was approved by the Mayo Clinic institutional review board (IRB 18–11309 and 22–001404) and was conducted in concordace with the U.S. Common Rule. This study complies with the “Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis” (TRIPOD) statement (29). No patients were involved in any part of the study, including concept and study design, data collection, analysis and interpretation, drafting of the manuscript and critical revision.
Table 1.
Clinical and pathologic parameters of the study cohort of stage II MMRP colorectal cancers
| Clinical and Pathologic Parameters | Total | Mayo Clinic | UPMC | Mount Sinai | P-value |
|---|---|---|---|---|---|
| Time frame | 2017–2020 | 2010–2015 | 2011–2016 | -- | |
| Female Sex (%) | 172 (43.7) | 56 (37.8) | 65 (48.5) | 51 (45.5) | 0.2 |
| Median Age (IQR) | 68 (18) | 68 (22) | 68 (16) | 66 (19) | 0.3 |
| Tumor location, N (%) | |||||
| Proximal colon | 206 (51.8) | 76 (51.4) | 80 (59.7) | 50 (43.1) | 0.03 |
| Distal colon | 192 (48.2) | 72 (48.6) | 54 (40.3) | 66 (56.9) | |
| Tumor grade, N (%) | |||||
| Low-grade | 382 (96.0) | 136 (91.9) | 132 (98.5) | 114 (98.3) | 0.006 |
| High-grade | 16 (4.0) | 12 (8.1) | 2 (1.5) | 2 (1.7) | |
| pT stage | |||||
| pT3 | 337 (84.7) | 131 (88.5) | 105 (78.4) | 101 (87.1) | 0.04 |
| pT4 | 61 (15.3) | 17 (11.5) | 29 (21.6) | 15 (12.9) | |
| pT4a | 45 (11.3) | 12 (8.1) | 18 (13.4) | 15 (12.9) | |
| pT4b | 16 (4.0) | 5 (3.4) | 11 (8.2) | 0 (0) | |
| VELIPI N (%) | |||||
| Absent | 239 (60.1) | 121 (81.8) | 74 (55.2) | 44 (37.9) | <0.001 |
| Present | 159 (39.9) | 27 (18.2) | 60 (44.8) | 72 (62.1) | |
| Tumor budding grade, N (%) | |||||
| Bd1 | 258 (64.8) | 105 (70.9) | 97 (72.4) | 56 (48.3) | <0.001 |
| Bd2 | 81 (20.4) | 28 (18.9) | 22 (16.4) | 31 (26.7) | |
| Bd3 | 59 (14.8) | 15 (10.1) | 15 (11.2) | 29 (25.0) | |
| Bowel Obstruction | |||||
| Absent | 353 (88.7) | 143 (96.6) | 110 (82.1) | 100 (86.2) | <0.001 |
| Present | 45 (11.3) | 5 (3.4) | 24 (17.9) | 16 (13.8) | |
| Macroscopic tumor perforation | |||||
| Absent | 377 (94.7) | 143 (96.6) | 124 (92.5) | 110 (94.8) | 0.3 |
| Present | 21 (5.3) | 5 (3.4) | 10 (7.5) | 6 (5.2) | |
| Pre-operative CEA | |||||
| Not elevated | 203 (51.0) | 84 (56.8) | 78 (58.2) | 41 (35.3) | 0.08 |
| Elevated | 96 (24.1) | 53 (35.8) | 27 (20.1) | 16 (13.8) | |
| Not done | 99 (24.9) | 11 (7.4) | 29 (21.6) | 59 (51.0) | |
| Median CEA (ng/ml) (IQR) | 2.5 (3.2) | 2.5 (3.25) | 2.6 (3.9) | 2.6 (2.8) | 0.8 |
| Stage II Adjuvant chemotherapy | |||||
| No | 325 (81.7) | 127 (90.1) | 106 (81.5) | 92 (79.3) | <0.001 |
| Yes | 62 (15.9) | 14 (9.5) | 24 (17.9) | 24 (20.7) | |
| FOLFOX | 20 (5.0) | 2 (1.4) | 15 (11.5) | 3 (2.6) | |
| CAPOX | 7 (1.8) | 7 (5.0) | 0 (0) | 0 (0) | |
| 5-FU/Capecitabine | 32 (8.0) | 4 (2.8) | 9 (6.9) | 19 (16.4) | |
| Unknown chemotherapy | 3 (0.8) | 1 (0.7) | 0 (0) | 2 (1.7) | |
| Unknown | 11 (2.9) | 7 (4.7) | 4 (3.0) | 0 (0) |
5-FU, 5-fluoropyrimidine; FOLFOX, 5-fluoropyrimidine and oxaliplatin, CAPOX, capecitabine and oxaliplatin; MMRP, mismatch repair proficient; VELIPI, venous, lymphatic or perineural invasion; Bd, tumor budding grade
Application of QuantCRC and Risk Schemes
One representative H&E slide was digitized using Leica Aperio GT450 or AT2 at 40X. All cases were stained with H&E and scanned at their respective institutions. Digitized H&E images were uploaded to the Aiforia Create deep learning cloud-based platform (Aiforia Technologies, Helsinki, Finland). Each image was manually reviewed, and the entire tumor bed was digitally outlined. QuantCRC was applied to the tumor bed to segment the image as previously described (26,27). In brief, QuantCRC was trained on 24,157 annotations made on 559 CRCs (using both Aperio GT450 and AT2 scanners) not used in this study. The segmentation algorithm employs four convolutional neural networks (CNN) to segment the image in a stepwise manner. First, the tumor bed is segmented into carcinoma, tumor budding/poorly differentiated clusters (TB/PDC), stroma, mucin, necrosis, fat, and smooth muscle. The second CNN layer uses the stromal segmentation performed in CNN layer 1 to further segment stroma into immature (loose, often myxoid stroma with haphazardly arranged plump fibroblasts and collagen fibers), mature (densely collagenous areas with scattered fibroblasts, often with parallel collagen fibers), and inflammatory (dense clusters of chronic inflammatory cells obscuring stromal cells) subtypes. The third CNN layer uses the carcinoma region performed in CNN layer 1 to further segment into low-grade, high-grade, and signet ring cell carcinoma (SRCC). The fourth CNN layer identifies TILs within the carcinoma and TB/PDC (tumor epithelium) regions from CNN layer 1. Fat and smooth muscle were subtracted from the tissue area to generate the tumor bed area. The following fifteen QuantCRC parameters were recorded for each tumor: %tumor, %stroma, tumor:stroma ratio, %TB/PDC within the tumor, %mucin within the tumor, %necrosis within the tumor bed, %high-grade, %SRCC, TILs per mm2 of tumor epithelium, %immature stroma (tumor bed), %inflammatory stroma (tumor bed), %mature stroma (tumor bed), %immature (stromal region), %inflammatory (stromal region), and %mature (stromal region). These parameters were chosen as they corespond to well known pathologic features that have been studied qualitatively or semi-quantitatively (28,30–36).
QuantCRC risk classification (low-risk, intermediate-risk, and high-risk) was determined for each tumor using our previously developed prognostic model that incorporates 15 features from QuantCRC and predefined cutoffs from the predicted 36 month RFS that this model provides based on analysis of 2,411 Colon Cancer Family Registry CRCs with known recurrence data (26). Stage II ASCO risk stratification schemes were taken from recent guidelines (1,2). The stage II MMRP QuantCRC-integrated risk scheme utilizes pT3 vs. pT4 stage and QuantCRC risk classification. A diagramatic overview of this study is provided in Supplementary Figure S1.
Statistical Analysis
Chi-squared and Kruskal-Wallis were used to characterize the relationships between categorical and continuous variables respectively. Agreement between the QuantCRC-integrated and ASCO schemes was determined using kappa statistics and interpretated based on Landis and Koch (37). Survival analysis was conducted using Kaplan-Meier and Cox proportional hazards regression. 3-year RFS estimates are provided for risk groups within each risk scheme. Performance metrics included log-rank test and Hazard ratios (95% confidence intervals). R version 4.1.2 was used for statistical analysis.
Data Availability Statement
The data generated in this study are not publicly available due to information that could compromise patient privacy but are available upon reasonable request from the corresponding author.
RESULTS
Study cohort characteristics
Clinicopathologic variables for the cohort are shown in Table 1. The median age was 68 and 51.8% were located in the proximal colon. The majority of tumors were low-grade (382/398, 96%). The presence of pT4 was identified in 61/398 (15.3%) of tumors with 5.3% also demonstrating macroscopic tumor perforation. Bowel obstruction was identified in 11.3% (45/353). The presence of VELIPI was identified in 159/398 (39.9%) with considerable variation across the three institutions (P<0.001). High-grade tumor budding (Bd3), another high-risk feature, was identified in 59/398 (14.8%) tumors. Tumor bud grade also demonstrated significant variation across the three sites (P<0.001). Pre-operative CEA data was available in 299/398 patients and was elevated in 96/299. Adjuvant therapy was given in 62/398 (15.9%) of cases with most receiving single agent 5-FU/capecitabine.
Pathologic and QuantCRC variables associated with recurrence free survival
A flow diagram outlining the steps in this study is provided in Supplementary Figure S1. Recurrence free survival (RFS) was evaluated according to pathologic risk factors currently used to guide adjuvant chemotherapy in stage II MMRP cancers including pT stage (Figure 1A), tumor budding (Figure 1B), VELIPI (Figure 1C), and obstruction (Table 2). In this cohort, there was no difference in RFS by pT-stage, obstruction, number of lymph nodes examined, or presence of VELIPI. Pathologist derived tumor budding grades Bd2 and Bd3 were associated with worse RFS compared to low tumor budding (Bd1).
Figure 1.
Recurrence free survival in stage II MMRP colorectal carcinomas according pathologic and QuantCRC features using Kaplan-Meier analysis. A. pT-stage. B. Pathologist tumor budding grade. C. VELIPI. D. QuantCRC %TB/PDC. E. QuantCRC %Immature stroma (tumor bed). F. QuantCRC tumor:stroma ratio. Abbreviations: Bd, tumor budding grade; TB/PDC, tumor budding/poorly differentiated clusters; VELIPI, venous/lymphatic/perineural invasion.
Table 2.
Univariate Cox analysis of pathologic and QuantCRC features for prediction of patient RFS in stage II MMRP colorectal cancers
| Variable | Hazard Ratio (95%CI) | P-value |
|---|---|---|
| Pathologic variables | ||
| pT stage | ||
| pT3 | 1 (ref) | 0.08 |
| pT4 | 1.72 (0.93–3.20) | |
| Grade | ||
| Low-grade | 1 (ref) | 0.27 |
| High-grade | 0.05 (0.0–11.19) | |
| VELIPI | ||
| Absent | 1 (ref) | 0.13 |
| Present | 1.48 (0.90–2.47) | |
| Tumor budding grade, N (%) | ||
| Bd1 | 1 (ref) | -- |
| Bd2 | 2.19 (1.19–4.02) | 0.01 |
| Bd3 | 2.60 (1.38–4.90) | 0.003 |
| Bowel Obstruction | ||
| Absent | 1 (ref) | 0.2 |
| Present | 1.65 (0.81–3.36) | |
| Lymph nodes examined | ||
| ≥ 12 | 1 (ref) | 0.9 |
| < 12 | 1.10 (0.27–4.51) | |
| QuantCRC variables (P<0.05 only) | ||
| %Tumor | ||
| Low | 1 (ref) | -- |
| Intermediate | 0.57 (0.32–1.02) | 0.06 |
| High | 0.33 (0.16–0.65) | 0.001 |
| %Stroma | ||
| Low | 1 (ref) | -- |
| Intermediate | 1.95 (0.93–4.1) | 0.08 |
| High | 3.04 (1.52–6.09) | 0.002 |
| Tumor:Stroma ratio | ||
| Low | 1 (ref) | -- |
| Intermediate | 0.65 (0.37–1.15) | 0.14 |
| High | 0.30 (0.15–0.62) | 0.001 |
| %TB/PDC | ||
| Low | 1 (ref) | -- |
| Intermediate | 1.66 (0.79–3.47) | 0.18 |
| High | 3.12 (1.60–6.11) | 0.0009 |
| %Immature Stroma (tumor bed) | ||
| Low | 1 (ref) | -- |
| Intermediate | 2.21 (1.03–4.75) | 0.04 |
| High | 3.88 (1.89–7.95) | 0.0002 |
| %Inflammatory stroma (stromal area) | ||
| Low | 1 (ref) | -- |
| Intermediate | 0.48 (0.26–0.90) | 0.02 |
| High | 0.47 (0.25–0.87) | 0.02 |
MMRP, mismatch repair proficient; VELIPI, venous, lymphatic or perineural invasion; Bd, tumor budding grade
QuantCRC was applied to one digitized image from each of the 398 CRCs to segment the image and extract quantitative data for 15 features. None of the CRCs in this study were used to train the QuantCRC segmentation model or the subsequent prognostic model that predicts 36-month risk of recurrence. To evaluate the QuantCRC features associated with RFS, the quantitative data from each of the 15 features were divided into low, intermediate, and high groups based on the 33rd and 66th percentiles. When grouped in this manner, six features were associated with RFS including %tumor (Table 2), %stroma (Table 2), %TB/PDC (Figure 1D), %immature stroma (tumor bed) (Figure 1E), tumor:stroma ratio (Figure 1F), and %inflammatory stroma (stromal area) (Table 2). High %tumor, tumor:stroma ratio, and %inflammatory stroma (stromal area) were associated with improved RFS, whereas, high %stroma, %TB/PDC, and %immature stroma (tumor bed) were associated with decreased RFS. Tumors with high (>66th percentile) immature stroma (tumor bed) exhibited the worst overall RFS with a hazard ratio of 3.88 (1.89–7.95, P=0.0002).
Comparison of ASCO and QuantCRC-integrated risk schemes
We previously developed a prognostic model that combines the 15 QuantCRC features with MMR status and overall stage to predict 36-month RFS based on analysis of 2,411 CRCs from the Colon Cancer Family Registry (26). We applied this prognostic model to this validation cohort of MMRP stage II tumors. Using predefined risk score cutoffs determined from our prior study, each tumor was classified as low-risk (N=101), intermediate-risk (N=181) and high-risk (N=116). In a multivariable model that included pT-stage, any high-risk pathologic feature, and QuantCRC risk classification, only tumors classified as QuantCRC high-risk were associated with RFS with a hazard ratio of 2.35 (1.15–4.82, P=0.019) (Supplementary Table S1).
We developed a novel risk stratification scheme that utilizes pT-stage and QuantCRC risk classification (Figure 2A) to test if this scheme can improve prediction of RFS compared to the ASCO risk scheme (Figure 2B) in this cohort. The concordance between the ASCO and QuantCRC-integrated schemes are shown in Supplementary Table S2. Agreement between the two schemes was only fair with kappa of 0.25 (95%CI 0.17–0.33). In the ASCO risk scheme, 168/398 (42.2%) were categorized low-risk compared to 249/398 (62.6%) in the QuantCRC-integrated scheme with concordance in only 136 tumors. Of the 168 low-risk CRCs in the ASCO scheme, 32 (19.0%) were classified as intermediate-risk in the QuantCRC scheme. Concordance was seen in only 56 of the 169 CRCs categorized as intermediate-risk in the ASCO scheme with 113/169 (66.9%) categorized as low-risk in the QuantCRC-integrated scheme. Of the 61 CRCs categorized as high-risk in the ASCO scheme, 33 (54.1%) were categorized as intermediate-risk in the QuantCRC-integrated scheme. Notable differences in the percentage with elevated pre-operative CEA were observed in the QuantCRC-integrated high-risk group (68.4%) compared to intermediate-risk (28.6%) and low-risk (30.2%) groups (P=0.002, Supplementary Table S3). Differences in median CEA levels were also observed. No difference in percentage with elevated pre-operative CEA or median CEA levels was observed among the risk groups in the ASCO scheme. Clinical and pathologic features were analyzed according to concordant and discordant risk groupings in the ASCO and QuantCRC-integrated schemes. Compared to concordant tumors, no differences in clinicopathologic features such as age, sex, tumor location, VELIPI, tumor grade, tumor budding grade, or lymph node recovery were identified in tumors that shifted between risk groups, (Supplementary Table S4). However, those categorized as ASCO high-risk but QuantCRC-integrated intermediate-risk had a lower number of patients with elevated pre-operative CEA as well as lower median CEA levels compared to those categorized as high-risk in both schemes.
Figure 2.
ASCO and QuantCRC-integrated risk schemes used in this study (blue = low-risk, green = intermediate-risk, red = high-risk). A. Flow chart of the stage II QuantCRC-integrated risk scheme (blue = low-risk, green = intermediate-risk, red = high-risk). B. Flowchart of the stage II ASCO risk scheme (blue = low-risk, green = intermediate-risk, red = high-risk). C. Recurrence free survival in stage II MMRP colorectal carcinomas regardless of ACT according to the QuantCRC scheme. D. Recurrence free survival in stage II MMRP colorectal carcinomas regardless of ACT according to the ASCO scheme. E. Recurrence free survival in stage II MMRP colorectal carcinomas in those receiving no ACT according to the QuantCRC scheme. F. Recurrence free survival in stage II MMRP colorectal carcinomas in those receiving no ACT according to the ASCO scheme. Abbreviations: ASCO, American Society of Clinical Oncology; HR, high-risk; IR, intermediate-risk; LN, lymph node; LR, low-risk.
When including all cases regardless of ACT, both QuantCRC-integrated intermediate-risk (hazard ratio 2.27, 1.32–3.91, P=0.003) and high-risk (hazard ratio 3.27, 1.42–7.55, P=0.006) groups demonstrated worse RFS compared to the QuantCRC-integrated low-risk group (Figure 2C). In contrast, only the ASCO high-risk group demonstrated worse RFS compared to ASCO low-risk CRCs with a hazard ratio of 2.24 (1.09–4.62, P=0.03) (Figure 2D). ACT was given to 21/61 (34.4%), 29/169 (17.2%), and 12/168 (7.1%) in ASCO high-risk, intermediate-risk, and low-risk groups respectively (P<0.001) (Supplementary Table S5). According to the QuantCRC-integrated scheme, 9/28 (32.1%), 26/121 (21.5%), and 27/249 (10.8%) received ACT in the high-risk, intermediate-risk, and low-risk groups respectively (P=0.004). In the subgroup of patients categorized as ASCO intermediate-risk but QuantCRC-integrated low-risk (N=113), 16 (14.2%) received ACT compared to 1/32 (3.1%) patient categorized as ASCO low-risk but QuantCRC-integrated intermediate-risk and 13/56 (23.2%) categorized as intermediate-risk in both schemes (Supplementary Table S5).
In order to adjust for the effects of chemotherapy on RFS, we analyzed RFS according to the ASCO and QuantCRC-integrated schemes in the subgroup (N=325) that did not receive any ACT. Significant differences in RFS were seen in the QuantCRC-integrated scheme between risk groups in the absence of ACT (Figure 2E). In the ASCO scheme, only the high-risk group demonstrated a statistically significant difference in RFS compared to the ASCO low-rosk group (Figure 2F).
The estimated 3-year RFS for ASCO and QuantCRC integrated schemes in the entire stage II cohort as well as those that received no ACT are shown in Table 3. Despite many more tumors categorized as low-risk in the QuantCRC-integrated scheme, the 3-year RFS was 92% (95% CI 88–96%), similar to the predicted 3-year RFS in the ASCO low-risk group (91%, 95%CI 87–96%). The QuantCRC-integrated intermediate-risk and high-risk groups had lower estimated 3-year RFS at 78% (95% CI 71–87%) and 69% (95%CI 51–95%) respectively compared to 85% (95%CI 78–92%) and 80% (95%CI 70–92%) for ASCO intermediate-risk and high-risk groups.
Table 3.
Estimated 3-year recurrence free survival according to ASCO and QuantCRC-integrated risk schemes by adjuvant chemotherapy.
| Risk Groups | ASCO risk scheme | QuantCRC-integrated risk scheme |
|---|---|---|
| 3-year RFS (95% CI) | 3-year RFS (95% CI) | |
| Stage II all | ||
| Low-risk | 91% (87–96%) | 92% (88–96%) |
| Intermediate-risk | 84% (79–90%) | 78% (71–87%) |
| High-risk | 80% (70–92%) | 69% (51–95%) |
| Stage II without ACT | ||
| Low-risk | 91% (87–96%) | 92% (88–96%) |
| Intermediate-risk | 85% (78–92%) | 77% (68–88%) |
| High-risk | 78% (65–94%) | 64% (43–96%) |
Abbreviations: ACT, adjuvant chemotherapy; ASCO, American Society of Clinical Oncology; RFS, recurrence free survival
QuantCRC features stratified by ASCO and QuantCRC-integrated risk schemes are shown in Figure 3 and Supplementary Figure S2. Figure 3A shows the QuantCRC features in two tumors that were classified differently in the ASCO and QuantCRC-integrated schemes. Differences in %TB/PDC (Figure 3B), tumor:stroma ratio (Figure 3C), %immature stroma (tumor bed) (Figure 3D), %inflammatory stroma (stromal area) (Figure 3E), %tumor (Supplementary Figure S2A), %stroma (Supplementary Figure S2B), TILs per mm2 tumor (Supplementary Figure S2C), %high-grade (Supplementary Figure S2D), %SRCC (Supplementary Figure S2E), %inflammatory stroma (tumor bed) (Supplementary Figure S2F), and %immature stroma (stromal area) (Supplementary Figure S2G) was seen among the QuantCRC-integrated risk groups. Differences in %TB/PDC (Figure 3B), tumor:stroma ratio (Figure 3C), %tumor (Supplementary Figure S2A), %stroma (Supplementary Figure S2B), TILs per mm2 tumor (Supplementary Figure S2E), %mature stroma (Supplementary Figure S2H and S2I), and %necrosis (Supplementary Figure S2J) were seen in the ASCO scheme for the QuantCRC features. No difference in %mucin were seen in either the QuantCRC-integrated or ASCO schemes (Supplementary Figure S2K). These results help provide biologic interpretability to the risk schemes.
Figure 3.
QuantCRC features according to ASCO and QuantCRC-integrated risk schemes. A. QuantCRC was applied to the stage II MMRP cohort which segments the image in a stepwise manner. First the image is segmented into carcinoma (green), stroma (light blue), mucin (dark blue), TB/PDC (red), necrosis (brown), smooth muscle (purple), and fat (yellow). Next the stroma is segmented into immature (teal), mature (green), and inflammatory (gray). The carcinoma is segmented into low-grade (purple), high-grade (orange), and signet ring cell (light green). Finally, TILs are recognized as objects (blue dots) within the tumor epithelium. After this segmentation, fifteen features are calculated from each image as shown. Two tumors with different classifications in the ASCO and QuantCRC-integrated scheme are shown. B. Scatter plot of %TB/PDC according to ASCO and QuantCRC-integrated risk shemes. C. Scatter plot of tumor:stroma ratio according to ASCO and QuantCRC-integrated risk shemes. D. Scatter plot of %immature stroma (tumor bed) according to ASCO and QuantCRC-integrated risk shemes. E. Scatter plot of %inflammatory stroma (stromal area) according to ASCO and QuantCRC-integrated risk sheme. Shown are the P-values for comparisons between LR, IR, and HR groups within the QuantCRC-integrated and ASCO risk schemes. Abbreviations: ASCO, American Society of Clinical Oncology; B, tumor bed; ; HR, high-risk; IR, intermediate-risk; LR, low-risk, SRCC, signet ring cell carcinoma; ST, stromal region; TB/PDC, tumor budding/poorly differentiated clusters; TIL, tumor infiltrating lymphocytes.
DISCUSSION
The majority of patients with stage II CRC will be cured with surgery and will receive no benefit from ACT (2,11,13). Current recommendations rely on high-risk features to guide administration of ACT; however, such schemes are inadequate as many patients with high-risk features will not have disease recurrence (2,15). Given the modest benefit in stage II CRCs, it is essential to spare many patients from the unnecessary side effects and increased costs associated with ACT (6,7). As over 150,000 new CRC cases are diagnosed each year in the United States with approximately 25% presenting as stage II, it is imperative to improve risk stratification schemes (38). Most research has focused on identifying new pathologic features that better predict recurrence (e.g. desmoplastic stromal subtypes (31)), immunohistochemical evaluation of the tumor immune microenvironment (e.g. immunoscore (22)), or expensive and time consuming molecular testing (e.g. ctDNA (17)).
In this study, we applied our recently developed quantitative segmentation algorithm, QuantCRC (26), to digitized H&E images of stage II MMRP CRCs to better understand the features that are associated with tumor recurrence. The tumors we analyzed has similar clinical and pathologic characteristics seen in other recent stage II CRC cohorts evaluating prognosis (14,17). We demonstrate that high %TB/PDC, %stroma, and %immature stroma (tumor bed) were associated with decreased RFS whereas high %tumor, tumor:stroma ratio, and %inflammatory stroma (stromal area) were associated with improved RFS in this subgroup of stage II tumors. Importantly, these findings are consistent with published literature. The presence of TB and PDC have been shown to provide important prognostic information in all stages of colorectal carcinoma (28,32,39,40). Increasing attention has also been given to the quantity and type of stroma associated with the invasive carcinoma. In particular, stroma-rich tumors with predominately immature stroma are associated with poor outcomes (31,36,41,42). In contrast, the presence of inflammatory cells within the stroma is associated with improved outcomes (22,43).
In our prior study, we trained a prognostic model that incorporates 15 QuantCRC features with MMR status and stage to directly predict the 36-month probability of RFS (26). The predicted RFS at 36-months from an individual CRC can provide risk classification using predefined cutoffs. We applied the QuantCRC prognostic model to develop a QuantCRC-integrated risk stratification scheme in hopes of improving upon ASCO guidelines. Most studies evaluating new prognostic markers fail to integrate the new marker into a useful clinical algorithm. Kleppe et al. was the first to incorporate a digital biomarker into a clinical decision support tool for stage II and III CRC (16). In their risk stratification scheme for stage II CRC, low-risk, intermediate-risk and high-risk groups were identified based on pT stage (pT3 or pT4), number of lymph nodes recovered (>12 or ≤ 12), and classification from their DoMore-v1-CRC AI-algorithm (good, uncertain, or poor prognosis). MMR status was not incorporated into this scheme, and the stage II cohort analyzed included both MMRD and MMRP tumors. High-risk stage II were defined by ≤ 12, pT4, and DoMore-v1-CRC uncertain or poor prognosis. Low-risk tumors were defined as either pT3, > 12 lymph nodes, and DoMore-v1-CRC good/uncertain prognosis or pT4, > 12 lymph nodes, and DoMore-v1-CRC good prognosis. All other combinations were regarded as intermediate-risk. In the 374 stage II CRCs analyzed, 198 (53%) were low-risk, 161 (43%) were intermediate-risk, and 15 (4%) were high-risk. Overall, this risk scheme provided a log rank P-value of 0.035 for cancer-specific survival in this cohort. Compared to the low-risk group, the intermediate-risk group had decreased cancer-specific survival with a hazard ratio of 2.88 (95%CI 1.19–7.01). The high-risk group provided a hazard ratio of 3.75 (95%CI 0.78–18.08).
Similar to Kleppe et al., we attempted to develop a simple-to-use stratification scheme by combining QuantCRC-derived risk classification with pT-stage, given that pT4 is considered the highest risk factor in ASCO guidelines (16). In this scheme, pT4 and a high-risk classification in the QuantCRC prognostic model were required for a final classification of QuantCRC-integrated high-risk. Intermediate-risk tumors could be either pT3 and QuantCRC high-risk or pT4 and QuantCRC low-risk or intermediate risk. QuantCRC-integrated low-risk tumors were defined by pT3 and a classification of low-risk or intermediate-risk in the QuantCRC prognostic model. Our goal was to compare the QuantCRC-integrated scheme to the ASCO scheme that includes tumor grade, tumor budding, VELIPI, obstruction, pT-stage, and number of lymph nodes examined. Only MMRP tumors were examined as this is the subgroup where ACT would be considered in contrast to MMRD tumors. Number of lymph nodes examined was not included in our QuantCRC-integrated scheme given that the vast majority of pathology laboratories routinely harvest 12 or more lymph nodes.
The QuantCRC-integrated risk scheme resulted in robust prognostic risk groups that outperformed current ASCO risk schemes. In the QuantCRC-integrated scheme, both the intermediate-risk and high-risk groups had significantly worse RFS compared to low-risk tumor with a HR 2.27 (1.32–3.91, P=0.003) and 3.27 (1.42–7.55, P=0.006) respectively. Importantly, the QuantCRC-integrated scheme placed more tumors in the low-risk category (62.6%) compared to the ASCO scheme (42.2%) as well as the scheme developed by Kleppe et al where 53% in their stage II cohort were classified as low-risk (16). Excluding MMRD from their cohort would likely result in even less tumors being classified as low-risk. Despite more tumors being classified as low-risk in the QuantCRC-integrated scheme, the estimated 3-year RFS remained high at 92% (95%CI 88–96%) compared to 90% (95%CI 86–95%) in the ASCO scheme. The excellent 3-year RFS in the low-risk group was maintained when analysis was restricted to patients receiving no ACT.
There has been tremendous interest in utlizing ctDNA to guide adjuvant chemotherapy in stage II CRC with studies indicating a very high rate of disease recurrence in ctDNA positive patients after surgery (44,45). The recently published DYNAMIC clinical trial directly compared a ctDNA-scheme to standard management using the ASCO scheme and demonstrated utility of the ctDNA approach, which resulted in lower rates of ACT without decreased RFS (17). In the DYNAMIC study, the estimated 3-year RFS among ctDNA-negative patients was 92.5%, which is comparable to the QuantCRC-integrated low-risk 3-year RFS estimate (17). The 3-year estimated RFS among all ctDNA positive patients treated with ACT was 86.4% (76.0% for those treated with single agent fluoropyrimidine and 92.6% treated with oxaliplatin-based ACT). In comparison, the 3-year RFS estimate for QuantCRC-integrated intermediate-risk and high-risk was 78% and 69%, respectively, in our entire cohort and 77% and 64%, respectively, in those not treated with ACT. Given the small number of patients treated with ACT in our cohort, we are not able to determine RFS estimates in the treated group. We hypothesize that a risk sheme combining ctDNA with the QuantCRC-integrated scheme may provide the most robust risk stratification. For example, one could envision performing ctDNA testing only in 37.4% of the patients categoried as QuantCRC-integrated intermediate and high-risk given the excellent 92% 3 year RFS in the 62.6% of patients categorized as QuantCRC-integrated low-risk. Given that QuantCRC only requires an already created H&E slide, this strategy could be an attractive alternative to either the ASCO-only scheme or ctDNA-only scheme.
Based on these results, the QuantCRC-integrated scheme could provide a helpful framework to guide ACT in stage II CRC by substantially reducing the number of patients where ACT is considered and maximizing the benefit of ACT in subgroups with increased risk of recurrence. This scheme also eliminates reliance on features with known suboptimal interobserver agreement such as VELIPI, grade, and tumor budding (46–49). The variation among pathologists is illustrated by the significant differences observed between the three institutions in this study for high risk features. Despite reducing reliance on high-risk features, the QuantCRC-integrated scheme still provides interpretable quantitative pathologic data that explains the risk stratification results, which is an advantage over “black box” AI algorithms.
One limitation of this study is the use of cohorts with varied treatment regimens due to institutional differences. In addition, we did not assess the ability of the scheme to predict response to ACT. In particular, the ability of ACT to improve RFS in the QuantCRC intermediate-risk group is unknown. Lastly, we also did not compare the QuantCRC-integrated scheme with schemes that incorporate immunoscore or ctDNA; however, our goal was to utilize readily available and inexpensive H&E slides that were prepared for routine pathologic interpretation. Given that only an H&E slide and pT-stage is required, the risk scheme could be embedded within the pathology report and readily available to oncologists shortly after surgery in contrast to a scheme using the immunoscore or molecular testing. Additional evaluation of a larger number of patients is needed to confirm this observation. We hope to evaluate this risk scheme in clinical trial datasets to further validate this approach.
Supplementary Material
STATEMENT OF TRANSLATIONAL RELEVACE.
Current risk stratification of stage II colorectal carcinomas is suboptimal and new schemes are necessary to better guide adjuvant chemotherapy. In this study, we determined if inclusion of a digital pathology biomarker, QuantCRC, can improve risk stratification of stage II mismatch repair proficient colorectal cancers. Our analysis demonstrated that a risk scheme including QuantCRC improves prediction of recurrence free survival over the American Society of Clinical Oncology risk scheme. The QuantCRC-integrated scheme placed more tumors in the low-risk category compared to the American Society of Clinical Oncology scheme (65.2% vs. 42.2%), thus reducing the number of patients who would be considered for adjuvant chemotherapy. The QuantCRC-integrated intermediate-risk and high-risk groups also had a higher risk of recurrence compared to the American Society of Clinical Oncology intermediate-risk and high-risk groups. Our study demonstrates the potential of AI-integrated risk schemes utilizing readily available and inexpensive hematoxylin and eosin stained slides.
Acknowledgements:
Rish K. Pai is supported in part by funding from the National Cancer Institute (NCI), National Institutes of Health (NIH) (award U01 CA167551).
CONFLICT OF INTEREST:
CW has received research funding (as a clinical trial investigator) from Vaccinex, Boston Biomedical, Lycera, Seattle Genetics, Symphogen, RAPT Therapeutics, and INHBRX. She has also received honoraria from Array Biopharma, Signatera, PrecisCA, and the Oncology Learning Network and served as a consultant to Nova Research Company
RKP1, RK report consulting income from Alimentiv Inc.
RKP2 reports consulting income from Alimentiv Inc., Allergan, and Verily
MBS has received consulting income from Novartis
TBS has received consulting income from Boehringer Ingelheim, TreosBio, Sobi, Ipsen, Array Biopharma, Seattle Genetics, Bayer, Genentech, Incyte, Merck, Boston Biomedical, Bayer, Amgen, Merck, Celgene, Lilly, Ipsen, Clovis, Seattle Genetics, Array Biopharma, Genentech, Abgenomics, Incyte
FS is a consultant for Guardant Health and has received research funding from Ventana Medical Systems
TW-B is an employee of Aiforia Inc.
All other authors report no financial relationships
ABBREVIATIONS:
- ASCO
American Society of Clinical Oncology
- ACT
adjuvant chemotherapy
- CRC
colorectal cancer
- MMRP
mismatch repair proficiency
- MMRD
mismatch repair deficiency
REFERENCES
- 1.Lieu C, Kennedy EB, Bergsland E, Berlin J, George TJ, Gill S, et al. Duration of Oxaliplatin-Containing Adjuvant Therapy for Stage III Colon Cancer: ASCO Clinical Practice Guideline. J Clin Oncol 2019;37:1436–47. [DOI] [PubMed] [Google Scholar]
- 2.Baxter NN, Kennedy EB, Bergsland E, Berlin J, George TJ, Gill S, et al. Adjuvant Therapy for Stage II Colon Cancer: ASCO Guideline Update. J Clin Oncol 2022;40:892–910. [DOI] [PubMed] [Google Scholar]
- 3.Böckelman C, Engelmann BE, Kaprio T, Hansen TF, Glimelius B. Risk of recurrence in patients with colon cancer stage II and III: a systematic review and meta-analysis of recent literature. Acta Oncol 2015;54:5–16. [DOI] [PubMed] [Google Scholar]
- 4.André T, Meyerhardt J, Iveson T, Sobrero A, Yoshino T, Souglakos I, et al. Effect of duration of adjuvant chemotherapy for patients with stage III colon cancer (IDEA collaboration): final results from a prospective, pooled analysis of six randomised, phase 3 trials. Lancet Oncol 2020;21:1620–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Argilés G, Tabernero J, Labianca R, Hochhauser D, Salazar R, Iveson T, et al. Localised colon cancer: ESMO Clinical Practice Guidelines for diagnosis, treatment and follow-up. Ann Oncol 2020;31:1291–305. [DOI] [PubMed] [Google Scholar]
- 6.Kerr RS, Love S, Segelov E, Johnstone E, Falcon B, Hewett P, et al. Adjuvant capecitabine plus bevacizumab versus capecitabine alone in patients with colorectal cancer (QUASAR 2): an open-label, randomised phase 3 trial. Lancet Oncol 2016;17:1543–57. [DOI] [PubMed] [Google Scholar]
- 7.Quasar Collaborative Group, Gray R, Barnwell J, McConkey C, Hills RK, Williams NS, et al. Adjuvant chemotherapy versus observation in patients with colorectal cancer: a randomised study. Lancet 2007;370:2020–9. [DOI] [PubMed] [Google Scholar]
- 8.Sargent DJ, Marsoni S, Monges G, Thibodeau SN, Labianca R, Hamilton SR, et al. Defective mismatch repair as a predictive marker for lack of efficacy of fluorouracil-based adjuvant therapy in colon cancer. J Clin Oncol 2010;28:3219–26. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Kim JE, Hong YS, Kim HJ, Kim K-P, Lee J-L, Park SJ, et al. Defective Mismatch Repair Status was not Associated with DFS and OS in Stage II Colon Cancer Treated with Adjuvant Chemotherapy. Ann Surg Oncol 2015;22 Suppl 3:S630–637. [DOI] [PubMed] [Google Scholar]
- 10.Gkekas I, Novotny J, Fabian P, Nemecek R, Palmqvist R, Strigård K, et al. Mismatch repair status predicts survival after adjuvant treatment in stage II colon cancer patients. J Surg Oncol 2020;121:392–401. [DOI] [PubMed] [Google Scholar]
- 11.Teufel A, Gerken M, Fürst A, Ebert M, Hohenthanner I, Klinkhammer-Schalke M. Benefit of adjuvant chemotherapy in high-risk colon cancer: A 17-year population-based analysis of 6131 patients with Union for International Cancer Control stage II T4N0M0 colon cancer. Eur J Cancer 2020;137:148–60. [DOI] [PubMed] [Google Scholar]
- 12.Mirkin KA, Hollenbeak CS, Mohamed A, Jia Y, El-Deiry WS, Messaris E. Impact of perineural invasion on survival in node negative colon cancer. Cancer Biol Ther 2017;18:740–5. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Morris M, Platell C, McCaul K, Millward M, van Hazel G, Bayliss E, et al. Survival rates for stage II colon cancer patients treated with or without chemotherapy in a population-based setting. Int J Colorectal Dis 2007;22:887–95. [DOI] [PubMed] [Google Scholar]
- 14.Ueno H, Ishiguro M, Nakatani E, Ishikawa T, Uetake H, Matsuda C, et al. Prospective Multicenter Study on the Prognostic and Predictive Impact of Tumor Budding in Stage II Colon Cancer: Results From the SACURA Trial. J Clin Oncol 2019;37:1886–94. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Kumar A, Kennecke HF, Renouf DJ, Lim HJ, Gill S, Woods R, et al. Adjuvant chemotherapy use and outcomes of patients with high-risk versus low-risk stage II colon cancer. Cancer 2015;121:527–34. [DOI] [PubMed] [Google Scholar]
- 16.Kleppe A, Skrede O-J, De Raedt S, Hveem TS, Askautrud HA, Jacobsen JE, et al. A clinical decision support system optimising adjuvant chemotherapy for colorectal cancers by integrating deep learning and pathological staging markers: a development and validation study. Lancet Oncol 2022;23:1221–32. [DOI] [PubMed] [Google Scholar]
- 17.Tie J, Cohen JD, Lahouel K, Lo SN, Wang Y, Kosmider S, et al. Circulating Tumor DNA Analysis Guiding Adjuvant Therapy in Stage II Colon Cancer. N Engl J Med 2022;386:2261–72. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Alexandrou G, Mantikas K-T, Allsopp R, Yapeter CA, Jahin M, Melnick T, et al. The Evolution of Affordable Technologies in Liquid Biopsy Diagnostics: The Key to Clinical Implementation. Cancers (Basel) 2023;15:5434. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.O’Connell MJ, Lavery I, Yothers G, Paik S, Clark-Langone KM, Lopatin M, et al. Relationship between tumor gene expression and recurrence in four independent studies of patients with stage II/III colon cancer treated with surgery alone or surgery plus adjuvant fluorouracil plus leucovorin. J Clin Oncol 2010;28:3937–44. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Allar BG, Messaris E, Poylin VY, Schlechter BL, Cataldo TE. Oncotype DX testing does not affect clinical practice in stage IIa colon cancer. Med Oncol 2022;39:59. [DOI] [PubMed] [Google Scholar]
- 21.Mlecnik B, Lugli A, Bindea G, Marliot F, Bifulco C, Lee J-KJ, et al. Multicenter International Study of the Consensus Immunoscore for the Prediction of Relapse and Survival in Early-Stage Colon Cancer. Cancers (Basel) 2023;15:418. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Pagès F, Mlecnik B, Marliot F, Bindea G, Ou F-S, Bifulco C, et al. International validation of the consensus Immunoscore for the classification of colon cancer: a prognostic and accuracy study. Lancet 2018;391:2128–39. [DOI] [PubMed] [Google Scholar]
- 23.Mlecnik B, Bifulco C, Bindea G, Marliot F, Lugli A, Lee JJ, et al. Multicenter International Society for Immunotherapy of Cancer Study of the Consensus Immunoscore for the Prediction of Survival and Response to Chemotherapy in Stage III Colon Cancer. J Clin Oncol 2020;38:3638–3654. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Foersch S, Glasner C, Woerl A-C, Eckstein M, Wagner D-C, Schulz S, et al. Multistain deep learning for prediction of prognosis and therapy response in colorectal cancer. Nat Med 2023;29:430–9. [DOI] [PubMed] [Google Scholar]
- 25.Bell PD, Pai RK. Immune response in colorectal carcinoma: a review of its significance as a predictive and prognostic biomarker. Histopathology 2022;81:696–714. [DOI] [PubMed] [Google Scholar]
- 26.Pai RK, Banerjee I, Shivji S, Jain S, Hartman D, Buchanan DD, et al. Quantitative Pathologic Analysis of Digitized Images of Colorectal Carcinoma Improves Prediction of Recurrence-Free Survival. Gastroenterology 2022;163:1531–1546. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Pai RK, Hartman D, Schaeffer DF, Rosty C, Shivji S, Kirsch R, et al. Development and initial validation of a deep learning algorithm to quantify histological features in colorectal carcinoma including tumour budding/poorly differentiated clusters. Histopathology 2021;79:391–405. [DOI] [PubMed] [Google Scholar]
- 28.Lugli A, Kirsch R, Ajioka Y, Bosman F, Cathomas G, Dawson H, et al. Recommendations for reporting tumor budding in colorectal cancer based on the International Tumor Budding Consensus Conference (ITBCC) 2016. Mod Pathol 2017;30:1299–311. [DOI] [PubMed] [Google Scholar]
- 29.Collins GS, Reitsma JB, Altman DG, Moons KGM. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD statement. BMJ 2015;350:g7594. [DOI] [PubMed] [Google Scholar]
- 30.González IA, Bauer PS, Liu J, Chatterjee D. Intraepithelial tumour infiltrating lymphocytes are associated with absence of tumour budding and immature/myxoid desmoplastic reaction, and with better recurrence-free survival in stages I-III colorectal cancer. Histopathology 2020;78:252–64 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Ueno H, Kanemitsu Y, Sekine S, Ishiguro M, Ito E, Hashiguchi Y, et al. A Multicenter Study of the Prognostic Value of Desmoplastic Reaction Categorization in Stage II Colorectal Cancer. Am J Surg Pathol 2019;43:1015–22. [DOI] [PubMed] [Google Scholar]
- 32.Shivji S, Conner JR, Barresi V, Kirsch R. Poorly differentiated clusters in colorectal cancer: a current review and implications for future practice. Histopathology 2020;77:351–68. [DOI] [PubMed] [Google Scholar]
- 33.Sinicrope FA, Graham RP. Tumor-Infiltrating Lymphocytes for Prognostic Stratification in Nonmetastatic Colon Cancer-Are We There Yet? JAMA Oncol 2021;7:969–70. [DOI] [PubMed] [Google Scholar]
- 34.Lee H, Sha D, Foster NR, Shi Q, Alberts SR, Smyrk TC, et al. Analysis of tumor microenvironmental features to refine prognosis by T, N risk group in patients with stage III colon cancer (NCCTG N0147) (Alliance). Ann Oncol 2020;31:487–94. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Rogers AC, Winter DC, Heeney A, Gibbons D, Lugli A, Puppa G, et al. Systematic review and meta-analysis of the impact of tumour budding in colorectal cancer. Br J Cancer 2016;115:831–40. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36.van Pelt GW, Sandberg TP, Morreau H, Gelderblom H, van Krieken JHJM, Tollenaar RAEM, et al. The tumour-stroma ratio in colon cancer: the biological role and its prognostic impact. Histopathology 2018;73:197–206. [DOI] [PubMed] [Google Scholar]
- 37.Landis JR, Koch GG. The measurement of observer agreement for categorical data. Biometrics 1977;33:159–74. [PubMed] [Google Scholar]
- 38.Cancer of the Colon and Rectum - Cancer Stat Facts [Internet]. SEER. [cited 2023 Oct 15]. Available from: https://seer.cancer.gov/statfacts/html/colorect.html
- 39.Ammendola S, Turri G, Marconi I, Burato G, Pecori S, Tomezzoli A, et al. The presence of poorly differentiated clusters predicts survival in stage II colorectal cancer. Virchows Arch 2021;478:241–8. [DOI] [PubMed] [Google Scholar]
- 40.Trinh A, Lädrach C, Dawson HE, Ten Hoorn S, Kuppen PJK, Reimers MS, et al. Tumour budding is associated with the mesenchymal colon cancer subtype and RAS/RAF mutations: a study of 1320 colorectal cancers with Consensus Molecular Subgroup (CMS) data. Br J Cancer 2018;119:1244–51. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 41.Ueno H, Kanemitsu Y, Sekine S, Ishiguro M, Ito E, Hashiguchi Y, et al. Desmoplastic Pattern at the Tumor Front Defines Poor-prognosis Subtypes of Colorectal Cancer. Am J Surg Pathol 2017;41:1506–12. [DOI] [PubMed] [Google Scholar]
- 42.Ueno H, Shinto E, Shimazaki H, Kajiwara Y, Sueyama T, Yamamoto J, et al. Histologic categorization of desmoplastic reaction: its relevance to the colorectal cancer microenvironment and prognosis. Ann Surg Oncol 2015;22:1504–12. [DOI] [PubMed] [Google Scholar]
- 43.Väyrynen JP, Lau MC, Haruki K, Väyrynen SA, Dias Costa A, Borowsky J, et al. Prognostic Significance of Immune Cell Populations Identified by Machine Learning in Colorectal Cancer Using Routine Hematoxylin and Eosin-Stained Sections. Clin Cancer Res 2020;26:4326–38. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 44.Tie J, Wang Y, Tomasetti C, Li L, Springer S, Kinde I, et al. Circulating tumor DNA analysis detects minimal residual disease and predicts recurrence in patients with stage II colon cancer. Sci Transl Med 2016;8:346ra92. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 45.Kotani D, Oki E, Nakamura Y, Yukami H, Mishima S, Bando H, et al. Molecular residual disease and efficacy of adjuvant chemotherapy in patients with colorectal cancer. Nat Med 2023;29:127–34. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 46.Dawson H, Kirsch R, Messenger D, Driman D. A Review of Current Challenges in Colorectal Cancer Reporting. Arch Pathol Lab Med 2019;143:869–82. [DOI] [PubMed] [Google Scholar]
- 47.Bokhorst JM, Blank A, Lugli A, Zlobec I, Dawson H, Vieth M, et al. Assessment of individual tumor buds using keratin immunohistochemistry: moderate interobserver agreement suggests a role for machine learning. Mod Pathol 2020;33:825–33. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 48.Harris EI, Lewin DN, Wang HL, Lauwers GY, Srivastava A, Shyr Y, et al. Lymphovascular invasion in colorectal cancer: an interobserver variability study. Am J Surg Pathol 2008;32:1816–21. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 49.Kirsch R, Messenger DE, Riddell RH, Pollett A, Cook M, Al-Haddad S, et al. Venous invasion in colorectal cancer: impact of an elastin stain on detection and interobserver agreement among gastrointestinal and nongastrointestinal pathologists. Am J Surg Pathol 2013;37:200–10. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The data generated in this study are not publicly available due to information that could compromise patient privacy but are available upon reasonable request from the corresponding author.