Prognostic Model Of Colorectal Cancer Based On Clinicopathological Indicators

Background: Inflammatory factors,tumor markers,serum enzymes and pathological features are widely available clinicopathological indicators with tumor prognostic potentials,however,practical prognostic models based on these indicators has been merely reported.The purpose of this study was to develop practical prognostic models with good performance for resectable colorectal cancer through the combination of clinicopathological indicators,and to validate the prognostic models by using dynamic data of clinicpathologcial indicators.Methods: This study collected patients with pathologically diagnosed primary colorectal cancer from January 2013 to December 2017 in the Union Hospital of Tongji Medical College of Huazhong University of Science and Technology and Hubei Cancer Hospital.All participants were followed until August 2018.Using the hospital’s electronic medical record system,we collected the patient’s age,gender,tumor location,TNM stage,tumor size,differentiation,circumferential margin,vascular tumor thrombus,nerve invasion,radiotherapy and chemotherapy and other pathological information,and we also collected the detection results of clinical indicators such as tumor markers,serum enzymes,and inflammatory factors before the surgery,and one week,one month and three months after the surgery.Epidata 3.1 was used to establish the database during the data collection.Patients who met the inclusion exclusion criteria and had complete clinical and follow-up information were included in the study.Prognostic outcomes included Progression-free survival(PFS)and Overall survival(OS).Receiver operating characteristic(ROC)curves were applied to transform the continuous variables into dichotomized variables by using inflexion points as cut-offs.Kaplan-Meier survival curves and log-rank test were used to compare the survival difference between groups classified by dichotomized clinical indicators.Cox proportional hazard regression analysis was used to detect the prognostic value of individual clinical indicators and comprehensive prediction models constructed based on clinical pathological indicators.Then the time-dependent ROC curve was used to obtain the area under the curve of each prognostic models to select the model with the best predictive performance.Based on the best prognostic model,the corresponding nomogram was constructed to evaluate the one-year and three-year progression-free survival and overall survival of patients,and the calibration curve was used to reflect the predictive accuracy of the models through the consistency between the observed and predicted events.Finally,the SAS Proc traj process was used to construct dynamic trajectories to verify the predictive performance of prognostic models.Statistical analyses were performed using SAS 9.4 and R 3.5.1.All analyses were two-sided,and P values <0.05 were considered statistically significant.Results: A total of 2165 patients with primary colorectal cancer were included in this study.The mean age of the patients was 58.300 ± 12.000 years.There were 1302(60.140%)male patients and 863(39.860%)female patients.Mortality and progression rates were 13.160% and 19.860%,respectively,within a median follow-up time of 28.830 months.Among pathological factors,there were differences in the distribution of TNM stage,differentiation,vascular tumor thrombus,nerve invasion,radiotherapy and chemotherapy(P<0.05)in the two groups with or without tumor progression.There were differences in the distribution of TNM stage,tumor location,differentiation,and circumferential margin(P<0.05)in the two groups of patients with or without death.Cox proportional hazards regression analysis showed that Carcinoembryonic antigen(CEA)(HR = 2.072,95% CI: 1.709-2.512),Carbohydrate antigen 19-9(CA19-9)(HR = 2.340,95% CI: 1.930-2.839),Monocyte / Lymphocyte(MLR)(HR = 1.875,95% CI: 1.549-2.271),Gamma-glutamyl transpeptidase(GGT)(HR = 1.720,95% CI: 1.412-2.094),TNM stage(HR = 3.914,95% CI: 3.028-5.058),differentiation(HR = 0.784,95% CI: 0.640-0.960),vascular tumor thrombus(HR = 1.510,95% CI: 1.224-1.863),nerve invasion(HR = 1.386,95% CI: 1.115-1.723),chemotherapy(HR = 1.596,95% CI: 1.299-1.960)and radiotherapy(HR = 1.849,95% CI: 1.313-2.606)were independent prognostic factors for progression-free survival.While CEA(HR = 2.584,95% CI: 2.015-3.313),CA19-9(HR = 2.247,95% CI: 1.766-2.860),MLR(HR = 2.057,95% CI: 1.624-2.606),GGT(HR = 1.978,95% CI: 1.568-2.497),TNM stage(HR = 3.543,95% CI: 2.575-4.874),circumferential margin(HR = 2.601,95% CI: 1.335-5.065),differentiation(HR = 0.748,95% CI: 0.587-0.952)and tumor location(HR = 0.858,95% CI: 0.678-1.087)were independent prognostic factors for overall survival.The corresponding prognostic models based on survival related clinicopathological characteristics presented more profound predictive values than individual indicators,and the area under the ROC curve for one-year progression-free survival and overall survival were 0.841 and 0.844,respectively.The C-indices of the predictive models were 0.800(95% CI: 0.779-0.821)and 0.792(95% CI: 0.765-0.818),and the Brier scores of the calibration curves were 0.123 and 0.107 for progression-free survival and overall survival,respectively.By constructing dynamic trajectories on clinicphathological indicators,the patients were dichotomized into high-risk group and low-risk group.We found that the high-risk group had 5.258-fold(95% CI: 3.162-8.742)risk to progress,and 5.342-fold(95% CI: 2.700-10.570)risk to die.Therefore,the prediction performance of the predictive models were good.Conclusions: The results of this study showed that the prognostic models based on tumor markers,serum enzymes,inflammatory factors,and clinicopathological features presented good performance for survival prediction in resectable colorectal cancer patients.Therefore,we have reason to believe that the predictive models constructed in this study can be widely used in clinical practice serve as economical and effective prognostic tools and provide a basis for the selection of clinical treatment methods.