Jian Deng, Qiang Gao, Jie Jiao, Baochuan Yang, Xunying Zhang, Jiayong Wang, Jinbo Jiang, Zhi Liu
In this multicenter study, a reduced pathology-inflammation model improved postoperative recurrence prediction beyond conventional clinicopathologic assessment in stage I-III CRC and remained stable after adjustment for newly incorporated treatment, surgical approach, and MSI/MMR information. Because postoperative inflammatory markers may partly reflect operative and early postoperative factors, and because the study remains retrospective, the model should be interpreted as a risk-stratification aid requiring prospective validation before routine clinical implementation.
BACKGROUND: Recurrence risk after curative surgery for non-metastatic colorectal cancer (CRC) remains heterogeneous within the same TNM stage. Pathological invasiveness and perioperative systemic inflammation may provide complementary prognostic information, but their combined value and the role of early postoperative inflammatory dynamics remain insufficiently defined. We aimed to develop and externally validate a multicenter prognostic model integrating invasive pathological features with perioperative inflammatory biomarkers to improve recurrence-free survival (RFS) prediction.
PATIENTS AND METHODS: We retrospectively included 1,260 patients with stage I-III colorectal adenocarcinoma who underwent curative-intent resection between January 2017 and March 2023. The development cohort comprised 1,008 consecutive patients from Qilu Hospital of Shandong University, and the external validation cohort comprised 252 patients from three regional hospitals. Data extraction followed a prespecified case-report framework with harmonized definitions for pathology, perioperative laboratory timing, adjuvant treatment, surgical approach, MSI/MMR status, and follow-up. The final Cox model was evaluated against prespecified comparator models, with additional sensitivity analyses adjusting for adjuvant chemotherapy, surgical approach, and MSI status.
RESULTS: The reduced final pathology-inflammation model incorporated eight predictors: histological differentiation, pN stage, extramural venous invasion (EMVI), tumor budding (BD), tumor deposits (TD), preoperative CEA, preoperative systemic immune-inflammation index (preSII), and postoperative day-1 C-reactive protein-to-albumin ratio (postCAR). The final model used 9 effective parameters and 437 recurrence events in the development cohort, corresponding to an events-per-variable ratio of 48.6. Its apparent Harrell C-index was 0.786 in the development cohort and 0.793 in the external validation cohort; bootstrap-based internal validation showed minimal optimism (mean optimism 0.003; optimism-corrected C-index 0.783). Adding adjuvant chemotherapy, surgical approach, and MSI status did not materially change discrimination in either cohort. The development-derived cutoff retained clinically meaningful separation in the external validation cohort, with 3-year sensitivity of 78.0% and specificity of 72.5%.
CONCLUSIONS: In this multicenter study, a reduced pathology-inflammation model improved postoperative recurrence prediction beyond conventional clinicopathologic assessment in stage I-III CRC and remained stable after adjustment for newly incorporated treatment, surgical approach, and MSI/MMR information. Because postoperative inflammatory markers may partly reflect operative and early postoperative factors, and because the study remains retrospective, the model should be interpreted as a risk-stratification aid requiring prospective validation before routine clinical implementation.