Risk stratification scores for lymph node metastases in <scp>T1</scp> colorectal cancer—A systematic review

Abstract Aim Local resection, by an endoscopic or surgical approach, has the potential to completely treat T1 colorectal cancers (CRCs). However, T1 CRCs have a 10% risk of lymph node metastasis (LNM), requiring colonic resection and lymph node dissection. Several high‐risk clinico‐histopathological features are associated with LNMs and multiple risk stratification scores have been developed. In this systematic review, we aimed to analyse these scores to identify which is most accurate and clinically useful. Method A search of MEDLINE, Cochrane Database of Systematic Review and EMBASE for T1 CRC risk assessment scores was performed following PRISMA guidelines. Results Of 323 studies, 22 full texts and three abstracts met the inclusion criteria. Twelve studies developed clinicopathological scores presented as nomograms or algorithms. They used an average of 4.8 (SD ±1.72) parameters, the most utilized being tumour grade, lymphovascular invasion and tumour budding. Two studies incorporated preoperative CT results in their risk score. Artificial intelligence (AI) machine learning models were used for 10 studies, with pathologist‐dependent parameters and pathologist‐independent whole‐slide imaging. The area under the curve (AUC) of the scores ranged from 0.57 to 0.99. Only two scores were externally validated, including a nomogram with an AUC of 0.77 and an AI model with an AUC of 0.83. The generalizability of several scores is limited by using special histopathology tests and AI programming/equipment. Conclusion There are several promising risk stratification scores for predicting LNM, particularly with the advent of AI. However, no score adequately stratifies the independent risks of rectal and colonic malignant polyps. Further studies are required to address the heterogeneity and lack of external validation within these nonrandomized trials to provide a more accurate risk stratification of LNMs.