RESUMO
BACKGROUND AND AIMS: Diagnostics to differentiate deep submucosal invasive (invasion depth ≥1000 µm [T1b]) colorectal cancer (CRC) from muscularis propria invasive (T2) CRC are limited. We aimed to establish and validate a scoring system that differentiates T1b from T2. METHODS: A multicenter retrospective cross-validation study was performed. Four hundred sixty-one consecutive pathologically confirmed T1b or T2 CRCs were divided into the development (T1b, 222; T2, 189) and internal validation (T1b, 31; T2, 19) cohorts. Eight potential endoscopic findings were evaluated using the development cohort: loss of lobulation, deep depression, demarcated depressed area, protuberance within the depression, expanding appearance, fold convergency, erosion or white plaque, and Borrmann type 2 or 3 tumor. A scoring system that differentiates T1b from T2 was developed, and diagnostic performance was tested using the internal validation cohort by 8 endoscopists. External validation was conducted using 50 CRC images by 4 endoscopists from other institutions, including outside of Japan. RESULTS: Multivariate analysis identified the following 5 independent predictive endoscopic findings of T2 CRC: deep depression (odds ratio [OR], 2.08; 95% confidence interval [CI], 1.07-4.04), demarcated depressed area (OR, 4.40; 95% CI, 1.39-13.9), 4-fold convergency or more (OR, 3.41; 95% CI, 1.90-6.11), erosion or white plaque (OR, 8.28; 95% CI, 2.77-24.7), and Borrmann type 2 or 3 tumor (OR, 8.76; 95% CI, 3.58-21.5). The area under the receiver-operating characteristic curve (AUROC) was .90 (95% CI, .87-.93) in the development cohort, .80 (95% CI, .76-.85) in the internal validation, and .76 (95% CI, .69-.83) in the external validation. CONCLUSIONS: We established and validated a new scoring system to differentiate T1b from T2 CRC using 5 simple endoscopic findings.