Clinicopathological Features Combined With Immune Infiltration Could Well Distinguish Outcomes in Stage II and Stage III Colorectal Cancer: A Retrospective Study

BACKGROUND: The Immunoscore predicts prognosis in patients with colorectal cancer (CRC). However, a few studies have incorporated the Immunoscore into the construction of comprehensive prognostic models in CRC, especially stage II CRC. We aimed to construct and validate multidimensional models integrating clinicopathological characteristics and the Immunoscore to predict the prognosis of patients with stage II–III CRC. METHODS: Patients (n = 254) diagnosed with stage II–III CRC from 2009 to 2016 were used to generate Cox models for predicting disease-free survival (DFS) and overall survival (OS). The variables included basic clinical indicators, blood inflammatory markers, preoperative tumor biomarkers, mismatch repair status, and the Immunoscore (CD3(+) and CD8(+) T-cell densities). Univariate and multivariate Cox proportional regressions were used to construct the prognostic models for DFS and OS. We validated the predictive accuracy and ability of the prognostic models in our cohort of 254 patients. RESULTS: We constructed two predictive prognostic models with C-index values of 0.6941 for DFS and 0.7138 for OS in patients with stage II–III CRC. The Immunoscore was the most informative predictor of DFS (11.92%), followed by pN stage, carcinoembryonic antigen (CEA), and vascular infiltration. For OS, the Immunoscore was the most informative predictor (8.59%), followed by pN stage, age, CA125, and CEA. Based on the prognostic models, nomograms were developed to predict the 3- and 5-year DFS and OS rates. Patients were divided into three risk groups (low, intermediate, and high) according to the risk scores obtained from the nomogram, and significant differences were observed in the recurrence and survival of the different risk groups (p < 0.0001). Calibration curve and time-dependent receiver operating characteristic (ROC) analysis showed good accuracy of our models. Furthermore, the decision curve analysis indicated that our nomograms had better net benefit than pathological TNM (pTNM) stage within a wide threshold probability. Especially, we developed a website based on our prognostic models to predict the risks of recurrence and death of patients with stage II–III CRC. CONCLUSIONS: Multidimensional models including the clinicopathological characteristics and the Immunoscore were constructed and validated, with good accuracy and convenience, to evaluate the risks of recurrence and death of stage II–III CRC patients.