Predictive Role of Tumor Budding in T1 Colorectal Cancer Lymph Node Metastasis

We read with interest the study by Kudo et al,1Kudo S.E. et al.Gastroenterology. 2021; 160: 1075-1084Abstract Full Text Full Text PDF PubMed Scopus (22) Google Scholar in which authors developed an artificial intelligence (AI) system to predict lymph node metastasis in T1 colorectal cancer. It provided a novel AI system that could help to prevent unnecessary extend surgeries. We noticed that the system was developed and validated by separate cohorts according to TRIPOD statement,2Collins G.S. et al.Br J Cancer. 2015; 112: 251-259Crossref PubMed Scopus (46) Google Scholar which would ensure the validity of artificial neural network (ANN) model. However, there are a few concerns that merit further exploration. First, tumor budding was defined as an important risk factor of lymph node metastasis in T1 colorectal cancer in many recent studies and guidelines.3Backes Y. et al.as. Gastroenterology. 2018; 154: 1647-1659Abstract Full Text Full Text PDF PubMed Scopus (29) Google Scholar,4NCCN Clinical Practice Guidelines in Oncology (NCCN Guidelines®) (Version 1.2019).www.nccn.org/professionals/physician_gls/f_guidelines.aspDate: 2019Google Scholar In this study, however, Kudo et al1Kudo S.E. et al.Gastroenterology. 2021; 160: 1075-1084Abstract Full Text Full Text PDF PubMed Scopus (22) Google Scholar developed the ANN model without tumor budding and did not analyze this feature in univariate or multivariate logistic regression. This practice is in contrast with their previous study,5Ichimasa K. et al.Endoscopy. 2018; 50: 230-240Crossref PubMed Scopus (47) Google Scholar in which tumor budding was determined to be one of the most significant risk factor in a machine learning model. The authors claimed that they collected pathologic factors, including depth of invasion and tumor budding according to the Japanese guidelines. However, these factors were not included in ANN model, possibly owing to the low agreement as stated in their Discussion. In their validation analysis, 939 patients were used to compare the accuracy of the ANN model, US guidelines, and Japanese guidelines. The predictive power of US guideline outperformed the Japanese guideline, although the main difference between these two was the inclusion of tumor budding and depth of infiltration in the latter. We are concerned that this result might mislead readers into thinking that tumor budding and submucosal invasion depth were not associated with lymph node metastasis, which was in contrast with most recent research including their previous study.5Ichimasa K. et al.Endoscopy. 2018; 50: 230-240Crossref PubMed Scopus (47) Google Scholar To address this issue, another machine learning model that includes tumor budding and submucosal invasion depth is necessary to clarify the predictive value of these factors for lymph node metastasis. Moreover, external validation was performed in this study, and the outcome defined as pathologic proved lymph node metastasis. However, the external validation results only showed predict accuracy of total patients (n = 939) and initial endoscopic resection (n = 517) in the validation cohort. The predicted outcomes of patients who underwent endoscopy resection alone were not clearly shown. Although these patients did not receive lymph node dissection, long-term local recurrence or survival outcome could be used as an endpoint to assess the total accuracy of AI system. These outcomes may determine whether the AI system could assist in surgical decision-making for the most “appropriate” patients. Overall, the study is very well-conducted. AI technology is promising in clinical healthcare and we look forward to the advent of practical predictive models like ANN to better aid the therapeutic decision-making in patients with colorectal cancer. Artificial Intelligence System to Determine Risk of T1 Colorectal Cancer Metastasis to Lymph NodeGastroenterologyVol. 160Issue 4PreviewIn accordance with guidelines, most patients with T1 colorectal cancers (CRC) undergo surgical resection with lymph node dissection, despite the low incidence (∼10%) of metastasis to lymph nodes. To reduce unnecessary surgical resections, we used artificial intelligence to build a model to identify T1 colorectal tumors at risk for metastasis to lymph node and validated the model in a separate set of patients. Full-Text PDF ReplyGastroenterologyVol. 161Issue 2PreviewWe thank Huang et al for their comments on our article and appreciate the opportunity to discuss the following 2 points1,2: (1) validation of the artificial intelligence (AI) system with the cohort who underwent endoscopic resection of T1 colorectal cancer but received no adjuvant surgery and (2) the development of an AI model which incorporates another 2 pathologic factors, namely, tumor budding and depth of submucosal invasion. Both points are considered clinically relevant and thus we are happy to provide additional data. Full-Text PDF