Development And Validation Of A Survival Prediction Model For Patients With Colorectal Cancer Lung Metastasis

Background and propose:Colorectal cancer(CRC)is one of the most common malignancies worldwide.The incidence rate of CRC in China is increasing over years.Lung is the second most common distant metastatic site of CRC.Lung is also a common distant metastatic site of other cancers.Therefore,lung metastasis(LM)is a crucial issue in the field of cancer.The survival of CRC-LM patients is correlated with various clinicopathological features,different treatment regimens and so on,so it is complicated and hard to predict the survival of CRC-LM patients.Clinicians have an urgent need for tools to predict the survival of CRC-LM patients.This study aimed to develop and validate models to predict individualized survival for patients with CRC-LM,which helps doctors to predict the survival of CRC-LM patients more accurately.Method:The data of CRC-LM patients was downloaded from the surveillance,epidemiology,and end result program(SEER)database and data from our institute.SEER database of the National Cancer Institute in the United States.The 7408 CRC-LM patients in SEER from2010 to 2016 were included in the development model cohort.In the development cohort,the survival status of patients was followed up to November 2018.Furthermore,7183 cases of CRC-LM patients who didn’t undergo LM resection and 225 cases of CRC-LM patients who underwent LM resection were selected as the development cohorts to establish two new survival prediction models according to the status of LM resection.The candidate variables were the same in all development cohorts,and the final variables were selected using the LASSO regression.Cox proportional hazards regression was used to establish a survival prediction model.The internal validation of the CRC-LM survival prediction model is based on different registration centers in the development cohort to the performance of the model.The internal validation of the model conducted by 1000 bootstrap.The C-index and Brier score were used to evaluate the model discrimination and calibration,and the receiver operating characteristic(ROC)curve and calibration curve were drawn.Meanwhile,we retrospectively collected 37 pathologically diagnosed CRC patients with lung metastasis from 2009 to 2020 in the First Affiliated Hospital of Guangzhou Medical University as an external validation cohort.These patients were followed up until April 1,2021.The inclusion and exclusion criteria of the two cohorts should be as consistent as possible.All patients underwent resection of lung metastases for CRC and had only lung metastasis at the time of diagnosis.The C-index and Brier score are used to evaluate the discrimination and calibration of the new models.All data were processed,analyzed and plotted using R(v4.0.2)in this study.Results:1.Survival prediction model for CRC-LM patientsThe overall survival(OS)of study patients: The 1-year,3-year,and 5-year OS in the development cohort were 52%,11%,and 2%,respectively.The final model had good calibration and discrimination(c-index=0.685,Brier score=0.190).Calibration and discrimination of internal validation were as follows : c index at 1-year =0.704,Brier score at 1-year =0.153,c-index at 3-year =0.656,Brier score at 3-year =0.228,the c-index at 5-year =0.650,the Brier score at 5-year =0.153.The final variables in CRC-LM survival prediction model included age at diagnosis,T stage,N stage,primary site resection status,perineural invasion,other metastases(liver,brain,bone),and radiotherapy.2.Survival prediction model for CRC-LM population without LM resectionThe OS of study patients: In the development cohort,there were 7183 CRC-LM patients’ data available according to the inclusion criteria.The 1-,3-,and 5-year OS of CRC-LM patients who did not undergo LM resection were 47%,9%,and 2%,respectively.The final variables in survival prediction model of CRC-LM patients without LM reaction included: diagnosis age,primary tumor site,T stage,CEA level before treatment,perineural invasion,liver metastasis.The calibration and discrimination of the final model were acceptable(c index=0.543,Brier score=0.185),internal validation used 1000-bootstrap validation(c index at1-year=0.515,Brier score at 1-year =0.117,c index at 3-year =0.513,Brier score at 3-year=0.252,c-index at 5-year =0.548,Brier score at 5-year =0.187)3.Survival prediction model for CRC-LM patients who underwent LM resection.The OS of study patients: In the development cohort,data of 225 CRC-LM patients were available according to the inclusion criteria.In the development cohort,the 1-,3-,and 5-year OS were 71%,33%,and 12%,respectively.In the external validation cohort,the 1-,3-,and 5-year OS were 89%,68%,and 49%,respectively.The final variables in survival prediction model for CRC-LM patients who underwent LM resection included primary tumor location,T stage,N stage,CEA level before treatment,and primary tumor resection status.The calibration and discrimination of the final model were moderate(c index=0.677,Brier score=0.127)Validation of survival prediction model for CRC-LM patients who underwent LM resection:Internal validation: The internal validation used 1000-bootstrap,and the calibration and discrimination were moderate(c index at 1-year=0.624,Brier score at 1-year =0.139,c-index at 3-year =0.605,Brier score at 3-year =0.259,c-index at 5-year =0.728,Brier score at 5-year =0.181).External validation: The external validation calibration and discrimination were moderate(external validated c index=0.604,external validated Brier score=0.478).4.Visualization of the model: Based on the above models,nomograms were developed to predict the survival probability of patients at 1-,3-and 5-year,and online tools based on the nomograms were provided as follows: Online survival prediction tool for CRC-LM patients,CRC-LM patients who did not underwent LM resection and CRC-LM patients who underwent LM resection was https://drxie.shinyapps.io/dynnomapp/,https://drxie.shinyapps.io/dynnomapp1/ and https://drxie.shinyapps.io/dynnomapp2/,respectively.Conclusion:We have developed the survival prediction models for patients with colorectal cancer lung metastasis.The three survival prediction models would provide clinicians with tools to accurately predict the prognosis for CRC-LM population,CRC-LM patients who did not undergo LM and CRC-LM patients who underwent LM resection.