CT Staging of Colon Cancer: Reproducibility and Pathology Association in a Multicenter Study.
To assess interobserver agreement of computed tomography (CT) features in colon cancer, their association with pathological staging, and the influence of baseline characteristics on clinical-pathological agreement.
CT scans of patients with locally advanced colon cancer treated between 2011 and 2020 at two Dutch hospitals were analyzed. Eleven radiologists independently reviewed the scans in pairs using a structured template. Interobserver agreement was evaluated using Krippendorff's alpha (α) and intraclass correlation coefficients (ICC). Associations between CT features and pathological stages, and the impact of baseline characteristics on clinical-pathological agreement, were assessed using mixed-effects logistic regression.
Interobserver agreement was α = 0.55 (95% CI: 0.51-0.60) for T stage, α = 0.57 (95% CI: 0.52-0.61) for N stage, α = 0.44 (95% CI: 0.36-0.51) for retroperitoneal surgical margin, α = 0.59 (95% CI: 0.52-0.65) for bowel obstruction, α = 0.27 (95% CI: 0.22-0.33) for extramural vascular invasion, α = 0.22 (95% CI: 0.14-0.31) for tumor deposits, ICC = 0.72 (95% CI: 0.70-0.75) for tumor length, and ICC = 0.62 (95% CI: 0.58-0.65) for largest node diameter. Significant associations (P < 0.05) were observed between clinical cT, cN, tumor length, and cEMVI with pT, and between cN, node diameter, and node heterogeneity with pN. Age, tumor location, and differentiation grade significantly influenced agreement.
Several CT features were significantly associated with pathological stage, but their inconsistent interpretation across observers, indicates limited reliability for individualized treatment decisions. Interpretation should therefore focus on features with proven reproducibility, namely tumor length and largest node diameter, applied within standardized protocols, and integrated with the broader clinical and pathological context.
CT scans of patients with locally advanced colon cancer treated between 2011 and 2020 at two Dutch hospitals were analyzed. Eleven radiologists independently reviewed the scans in pairs using a structured template. Interobserver agreement was evaluated using Krippendorff's alpha (α) and intraclass correlation coefficients (ICC). Associations between CT features and pathological stages, and the impact of baseline characteristics on clinical-pathological agreement, were assessed using mixed-effects logistic regression.
Interobserver agreement was α = 0.55 (95% CI: 0.51-0.60) for T stage, α = 0.57 (95% CI: 0.52-0.61) for N stage, α = 0.44 (95% CI: 0.36-0.51) for retroperitoneal surgical margin, α = 0.59 (95% CI: 0.52-0.65) for bowel obstruction, α = 0.27 (95% CI: 0.22-0.33) for extramural vascular invasion, α = 0.22 (95% CI: 0.14-0.31) for tumor deposits, ICC = 0.72 (95% CI: 0.70-0.75) for tumor length, and ICC = 0.62 (95% CI: 0.58-0.65) for largest node diameter. Significant associations (P < 0.05) were observed between clinical cT, cN, tumor length, and cEMVI with pT, and between cN, node diameter, and node heterogeneity with pN. Age, tumor location, and differentiation grade significantly influenced agreement.
Several CT features were significantly associated with pathological stage, but their inconsistent interpretation across observers, indicates limited reliability for individualized treatment decisions. Interpretation should therefore focus on features with proven reproducibility, namely tumor length and largest node diameter, applied within standardized protocols, and integrated with the broader clinical and pathological context.
Authors
Shkurti Shkurti, Groot Lipman Groot Lipman, Tissier Tissier, Chupetlovska Chupetlovska, Pooch Pooch, van Dijk-de Haan van Dijk-de Haan, Pijl Pijl, Taveira Taveira, Huige Huige, Gielens Gielens, Tonino Tonino, van Rooij van Rooij, Berenschot Berenschot, Coolen Coolen, Beets-Tan Beets-Tan, Nederend Nederend
View on Pubmed