Predictive Model For Estimating Survival In Colorectal Cancer Liver Metastases Including Traditional Chinese Medicine Factors

BackgroundIt is estimated that over 1.9 million cases of colorectal cancer have occurred in 2020 globally,with more than 935,000 deaths,accounting for 10%of cancer cases and deaths.Incidence rate of colorectal cancer ranks third in all malignancies,and the mortality rate ranks second.Liver metastases is the most common distant metastases of colorectal cancer.29%-40%of colorectal cancer patients have liver metastases.The treatment of liver metastases of colorectal cancer includes surgery,chemotherapy,radiofrequency ablation and radiotherapy.A large number of clinical prediction models for estimating the survival of colorectal cancer with liver metastasis have been developed,but less than 1/4 of them have been validated externally,and the extrapolation of the results needs to be improved.The prognosis of colorectal cancer patients with liver metastases were correlated with the number of liver metastases,the level of tumor markers(CEA,CA19-9),the primary lymph node metastases,the size of liver metastases,the existence of extrahepatic metastases,the location of the primary tumor,the grade of tumor differentiation,the margin of hepatectomy,the treatment of liver metastases,and tumor stage.There are also many factors related to the prognosis of liver metastases,such as,KRAS mutation,transaminase level,distribution of liver metastases,age,gender,marital status,albumin level,invasion of serosa,time-point of liver metastases and response to chemotherapy.TCM related factors affecting the prognosis of advanced colorectal cancer include TCM syndrome type,TCM treatment principle,TCM treatment time.At present,there is a lack of externally validated clinical prediction model based on large sample data of Asian population.At present,there is no clinical prediction model for survival prognosis of colorectal cancer with liver metastasis that includes clinical factors of traditional Chinese medicine,which directly affects the judgment of clinical prognosis.The purpose of this study is to solve the following three problems:whether we can establish a large sample of Asian colorectal cancer liver metastasis survival and prognosis model based on SEER database;what is the predictive ability of the prediction model applied to colorectal cancer liver metastasis patients in our hospital;whether we can establish a colorectal cancer liver metastasis survival and prognosis model including clinical factors of traditional Chinese medicine based on the medical record data of traditional Chinese medicine hospital?ObjectiveThe purpose of this study is to develop a large sample of Asian colorectal cancer liver metastasis survival predictive model based on SEER database and external validation;to evaluate the prediction ability of the prediction model for colorectal cancer liver metastasis patients hospitalized in traditional Chinese medicine hospital;to try to develop a colorectal cancer liver metastasis survival predictive model of better prediction ability based on the medical record data of traditional Chinese medicine hospital and incorporating clinical factors of traditional Chinese medicine.Methods and MaterialStudy 1:A retrospective study was conducted to screen the Asian colorectal cancer patients with liver metastasis registered in SEER database from 2010 to 2015.The patients were randomly divided into training set and internal validation set according to the ratio of 7:3.Patients with liver metastasis from colorectal cancer admitted to the Department of oncology,Guang’anmen Hospital,China Academy of Chinese Medical Sciences from January 1,2010 to December 31,2015 were selected as the external validation set.Extraction of patient information:①demographic data:gender,age at diagnosis;②tumor situation:location of primary tumor,degree of differentiation,primary lymph node metastasis,TNM stage(AJCC 6th Edition);③ treatment information:treatment of primary tumor;④ tumor marker:CEA;⑤follow-up information:survival time,survival status.SPSS 25.0 software was used for univariate survival analysis of related factors.The important factors affecting survival prognosis and other clinically significant factors were integrated to select the influencing factors included in the development of prediction model.R software was used to conduct multivariate Cox regression for the influencing factors included in the model development,and nomogram clinical prediction model was established according to the results.The training set is used to evaluate the discrimination and calibration of the prediction model,and the internal validation set and external validation set are used to verify the discrimination and calibration of the prediction model.R language and shinyapps platform were used to develop an external validated nomogram prediction model for prognosis of crlm patients into an online nomogram model.Study 2:In this study,patients with liver metastasis from colorectal cancer admitted to the Department of oncology,Guang’anmen Hospital,China Academy of Chinese Medical Sciences from January 1,2010 to December 31,2015 were selected as the training set.Extraction of patient information:①relevant data involved in the nomogram model;②TCM four diagnostic data:tongue and pulse;③TCM syndromes;④TCM treatment related information;⑤follow-up information:survival time and survival status.Univariate survival analysis was carried out for the relevant factors and TCM factors included in the study 1 model,and the factors that have significant impact on the prognosis and other clinically significant factors were selected for comprehensive inclusion prediction model development.R software was used to carry out Cox regression for the influencing factors of the included model development,and a nomogram clinical prediction model was established according to the results.If necessary,optimize the model.The training set was used to evaluate the discrimination and calibration of the prediction model,because the number of patients in this study was small,no internal verification was carried out;because the relevant data of traditional Chinese medicine was difficult to obtain in the public database,no external verification was carried out.OutcomesStudy 1:988 Asian colorectal cancer patients with liver metastasis registered in SEER database from 2010 to 2015 were enrolled in this study.The patients were randomly divided into training set and validation set according to the proportion of 7:3,training set 692 and internal validation set of 296.From 2010 to 2015,71 patients in Oncology department of Guang’anmen Hospital of Chinese Academy of Chinese Medical Sciences were included as external validation set.The results of log-rank analysis showed that age,sex,location of primary disease,T stage,N stage,surgery for primary tumor,extrahepatic metastases and differentiation grade were important factors influencing prognosis(P<0.05).The results showed that age,location of primary tumor,T stage,surgery for primary tumor,extrahepatic metastases and differentiation were the important factors influencing prognosis(P<0.05).The independent factors influencing survival and prognosis of the patients(P<0.05).The independent risk factors and N-stage related to the high frequency lymph node metastasis were selected.The clinical predictive model of 1-year survival,2-year survival and 3-year survival rate of CRLM patients was established by R software based on training set.After the model was established,N stage was found to account for a very low proportion in this model,and is removed from the prediction model by over fitting.The predictive model finally included six factors,including age,location of primary tumor,T stage,surgery for primary tumor,extrahepatic metastases and differentiation grade.It suggested that T2 stage,old than 70 years old,no surgery for primary tumor,undifferentiated tumor,right colon and present extrahepatic metastases had poor prognosis.The constructed prediction model is evaluated by the discrimination and calibration.Discrimination refers to the ability of the prediction model to distinguish high-risk individuals from low-risk individuals,and calibration is an indicator to measure the consistency between the prediction value of the prediction model and the actual survival situation of patients.Among them,c-index is 0.649,AUC of ROC curve for predicting 1-year survival is 0.692,AUC of ROC curve for 2-year survival is 0.695,AUC of ROC curve for 3-year survival is 0.754.Combined with calibration chart,it shows that the prediction model has nearly medium prediction division and calibration degree.The developed prediction model is applied to the internal validation set and external validation set,and the prediction model is validated by the division and calibration.Among them,AUC of ROC curve of internal validation and prediction of 1-year survival is 0.676,ROC curve AUC=0.642 for 2-year survival,AUC=0.721 for 3-year survival,and analysis based on calibration plots,which shows that the prediction model has a prediction partition of moderate in internal verification,but the calibration of prediction of 3-year survival is slightly lower.The external validation cohort is used to verify the prediction of 1-year survival and 2-year survival.Among them,the ROC curve AUC of 1 year survival is 0.671,and the ROC curve AUC=0.68 for the prediction of 2-year survival is analyzed.The analysis combined shows that the prediction model has a prediction index of moderate in internal verification,and the calibration of prediction of 1-year and 2-year survival is slightly lower.The results show that the model has good extrapolation.Access online Nomogram at https://drwang2021.shinyapps.io/dynnomapp/Study 2:In this study,71 patients with crlm admitted to the Department of oncology,Guang’anmen Hospital,China Academy of Chinese Medical Sciences from 2010 to 2015 were enrolled retrospectively.Log rank analysis showed that age,primary tumor operation,extrahepatic metastasis,tumor differentiation,CEA and TCM treatment time were important factors affecting the prognosis(P<0.05),while primary tumor location,T stage,TCM syndrome type,tongue color,tongue coating and pulse condition had no significant effect on the prognosis(P>0.05).Age,primary location,primary surgery,extrahepatic metastasis,tumor differentiation,T stage,CEA and TCM treatment time were included in Cox regression analysis.According to the principle that each variable corresponds to more than 10 events,only up to 7 variables can be included in this study.According to the results of multivariate regression,the primary lesion location of the variable with the largest P value is removed.The nomogram model of 1-year survival rate and 2-year survival rate was constructed by using R software according to 7 variables including age,primary tumor operation,extrahepatic metastasis,degree of differentiation,T stage,CEA and TCM treatment time.The c-index and ROC curve were used to evaluate the discrimination of the model,and the calibration chart was used to evaluate the calibration degree of the model.The c-index was 0.80,the 1-year survival rate predicted AUC=0.848,and the 2-year survival rate predicted AUC=0.906.Combined with the calibration chart analysis,this model has the possibility of over fitting.Because the sample size of this study is small,the prediction model is optimized by reducing the variables.The predictive factors of extrahepatic metastasis were removed from the predictive model,and a clinical predictive model was established,which was composed of six variables:age,primary surgery,degree of differentiation,T stage,CEA,and TCM treatment time.The c-index and ROC curve were used to evaluate the discrimination of the model,and the calibration chart was used to evaluate the calibration degree of the model.The c-index was 0.80,the 1-year survival rate predicted AUC=0.814,and the 2-year survival rate predicted AUC=0.848.The calibration plots showed that the predictive ability of the prediction model for 1-year and 2-year survival rate was better.Compared with the prediction model established in Study 1,the prediction model established in this study can better predict the survival prognosis of patients with CRLM included in this study.However,this prediction model has not been verified by internal verification set and external verification set,and its external applicability needs to be tested.Access online Nomogram(Version.TCM)at https://drwang2020.shinyapps.io/dynnomap/Conclusion1.Based on SEER database,a nomogram prediction model of survival and prognosis of colorectal cancer patients with liver metastasis was established,including age,primary location,T stage,primary surgery,extrahepatic metastasis and degree of differentiation.It is suggested that the prognosis of patients with liver metastasis of right colon cancer with T2 stage,over 80 years old,without primary surgery,undifferentiated or unknown tumor differentiation,and extrahepatic metastasis is poor.After internal and external validation,the prediction model has good accuracy and external applicability;based on this predictive model,an online nomogram is developed.2.Based on the medical records of colorectal cancer patients with liver metastasis in the oncology department of Guang’anmen Hospital of China Academy of Chinese Medical Sciences,a nomogram predictive model was developed,which was composed of six variables:age,surgery of primary tumor,grade of differentiation,T stage,CEA and duration of TCM treatment.It is suggested that the prognosis of CRLM patients younger than 50 years old,undifferentiated carcinoma,without surgery of primary tumor,T3 stage,taking traditional Chinese medicine for less than 6 months,CEA positive is poor.The clinical prediction model including TCM related factors has moderate prediction ability for the survival prognosis of CRLM patients,and has better prediction ability than the prediction model based on SEER database with large samples,but its external applicability needs to be tested.The extrapolation of the clinical prediction model incorporating TCM factors needs further large sample prospective study to verify.