Abstract
BACKGROUND: Growing evidence indicates that perineural invasion (PNI) is associated with local recurrence, distant metastasis, and unfavorable prognosis in patients with colorectal cancer (CRC). However, reliable tools for the diagnosis and outcome assessment of PNI-positive colorectal cancer remain insufficiently studied. In this study, diagnostic and prognostic models were constructed and validated for patients with PNI-positive colorectal cancer. METHODS: Patients included in this study were selected from two independent datasets within the Surveillance, Epidemiology, and End Results (SEER) database. Univariate and multivariate logistic regression analyses were performed to identify independent risk factors for perineural invasion in colorectal cancer. Univariate and multivariate Cox proportional hazards regression analyses were also conducted to identify independent prognostic factors associated with outcomes in patients with PNI positive colorectal cancer. Two nomogram models were then developed to estimate the risk and prognosis of PNI positive colorectal cancer. The area under the curve(AUC) and calibration curves were applied to evaluate the predictive performance of the nomograms. Decision curve analysis(DCA) and Kaplan Meier survival curves were further used to assess their potential clinical usefulness. RESULTS: Multivariate logistic regression analysis was performed to determine independent risk factors of colorectal cancer with perineural invasion (CRCPNI), including N stage, T stage, tumor deposits, grade, primary site, and carcinoembryonic antigen(CEA) level. Multivariate Cox regression analysis further identified age, T stage, CEA level, N stage, chemotherapy, and tumor deposits as factors independently associated with prognosis in CRCPNI patients. The results of these analyses were presented using nomogram models. Receiver operating characteristic(ROC) analysis, calibration plots, and DCA were applied to evaluate the performance of the risk and prognostic models. CONCLUSION: After validation, the predictive model was shown to be reliable and can provide supportive information for individualized clinical decision making in future practice.