Construction And Application Of Tumor-related Death Prediction Model In Elderly Patients With Colorectal Cancer(?60 Years Old) After Surgery

PurposeThe main purpose of this study is to construct a prediction model of tumor-related death in elderly patients(?60 years old)with colorectal cancer after surgery using competing-risk approach,and to visualize the model in the form of a nomogram,and to evaluate the performance of the model by internal and external validation,so as to help clinicians to make a more reasonable clinical decision.MethodsThe data were extracted from Surveillance,Epidemiology,and End Results(SEER)database to include patients with colorectal cancer who had undergone surgical treatment from 2010 to 2015.The data were from "SEER 18 Regs Custom Data(with additional treatment field),Nov 2017 Sub(1973-2015 varying)".Using competing-risk methodology,death from other cause was regarded as a competitive event,the cumulative incidence function(CIF)of tumor-related death was calculated as univariate analysis.After excluding the covariates with no statistical difference in univariate analysis,the residual variables were included in multivariate analysis.In multivariate analysis a proportional subdistribution hazard model was constructed.After excluding the variables with no statistical difference in multivariate analysis,the remaining variables are included in the construction of the nomogram.Based on this model,a competing-risk nomogram was implemented to predict the probability of tumor-related death at 1-year,3-year and 5-year.In internal validation,we used concordance(C)-indexes to evaluate discrimination,and calibration was evaluated using a calibration plot.Telephone follow-up data of elderly patients with colorectal cancer in gastrointestinal center from 2012 to 2016 in Northern Jiangsu People's Hospital,Clinical Medical School,Affiliated Hospital to Yangzhou University,including postoperative survival,chemotherapy,combined with other model-related indicators.Finally,669 cases of effective follow-up data were obtained and used for external verification.The external validation is also carried out by calculating C-index and drawing calibration plot.ResultsDataset of 20608 patients who met the inclusion criteria were eventually selected for analysis.The median follow-up was 21 months.The follow-up results of 7218 patients were death,including 5171 cancer-related deaths and 2047 non-cancer-related deaths.Predictors in this study included marital status,age,sex,race,tumor site,tumor size,pathological grade,AJCC TNM stage,lymph node ratio(LNR),CEA,perineural invasion and chemotherapy.5-year CIF of tumor-related death was 29.285%(95%confidence interval[Cl]29.281%-29.289%)and 13.650%(95%Cl 13.648%-13.653%)for other causes.Constructing a predictive model after transforming continuous variables into categorical variables.Univariate analysis showed that there was no significant difference in sex,race,and tumor site in tumor-related deaths(P>0.05).After excluding the three indicators,the remaining indicators were included in multivariate analysis.Multivanate analysis showed that the remaining indicators were related to tumor-related death.After adjusting the variable grouping method,the remaining seven covariates were finally included in the construction of the nomogram,and the CIF were shown in the nomogram for 1 year,3 years,and 5 years.The internal verification showed that the C-index of the nomogram was 0.818(95%Cl 0.812-0.824)and the calibration curve performed well.External verification showed a C-index of 0.764(95%Cl 0.728-0.801)and the calibration curve performed equally well.The nomogram was shown to have good discrimination and calibration after internal and external validation.ConclusionUsing the large sample database and competing-risk analysis,a postoperative prediction model for elderly patients was established.The performance of the model was good.It realizes the individual prediction of the elderly colorectal cancer population and helps clinicians to make clinical decisions.