Interobserver Agreement and Radiological-Pathological Concordance of CT Staging in Localized Colon Cancer

Accurate preoperative staging is essential for risk stratification and treatment planning in localized colon cancer (lCC), particularly in the context of emerging neoadjuvant strategies. However, the reproducibility of CT-based classification and its concordance with pathology remain incompletely defined.

We analyzed 175 patients with lCC who underwent baseline contrast-enhanced CT followed by curative-intent surgery. Two independent radiologists assessed clinical tumor (cT) and nodal (cN) stages, with discrepancies resolved by consensus. Interobserver agreement was evaluated using Cohen's kappa. Radiological-pathological (R-P) concordance was assessed for T, N, and overall stage. Multivariable logistic regression analyses were performed to identify predictors of concordance.

Interobserver agreement was good for cT classification (κ = 0.77) and excellent for cN classification (κ = 0.94), with overall concordance of 82.9%. R-P concordance was achieved in 78.3% for T staging and 73.7% for N staging, while overall concordance was 57.1%. Most discrepancies reflected adjacent misclassification (29.3%), whereas gross errors were rare (2.3%). CT showed moderate diagnostic performance for T stage (sensitivity 75.7%, specificity 81.6%) and lower sensitivity for nodal involvement (sensitivity 52.3%, specificity 87.3%). Performance varied by tumor location, with higher accuracy in left-sided tumors. Lymphovascular invasion was associated with lower interobserver agreement (odds ratios [OR] 0.34; P = .014), while advanced cT stage predicted reduced R-P concordance for T staging (OR 0.04; P = .004).

CT classification demonstrates high interobserver reproducibility but only moderate concordance with pathology in lCC. Misclassification may impact clinical decision-making, underscoring the need for improved staging strategies and multidisciplinary integration.