Colorectal Neoplasia Risk Prediction Model Based On Nomogram

Objective:The incidence and mortality of colorectal cancer(CRC)in China are increasing year by year,which is a threat to the health of Chinese residents.Early prevention and detection are essential to reduce the incidence and mortality of CRC.Due to the large population base and limited medical resources,risk stratification has become a cost-effective screening strategy in China.The aim of this study is to analyze the related risk factors of colorectal neoplasia and advanced colorectal neoplasia,and to establish effective prediction models.Methods:A total of 429 participants who underwent colonoscopy at the Affiliated Hospital of Qingdao University from May 2017 to June 2022 were included in this study.Clinical data were collected.Univariate analysis and multivariate logistic regression were used to screen the related factors of colorectal neoplasia and advanced colorectal neoplasia.According to the risk factors,nomogram visualization models were constructed,and the receiver operating characteristic(ROC)curve was used to test the discrimination of models.The calibration curve and Hosmer-Lemeshow goodness-of-fit test were used to evaluate the fit of the model.Decision curve analysis(DCA)was used to evaluate the clinical benefit.The Bootstrap method was used for internal validation(1000 repeated samples).The risk stratification was performed for colorectal neoplasia and advanced colorectal neoplasia model respectively.The risk prediction model of advanced colorectal neoplasia was compared with the Asia-Pacific Colorectal Screening(APCS)score in terms of discrimination,fit and clinical benefit.Results:1.For colorectal neoplasia,univariate analysis showed that gender,age,intake frequency of fruit,yogurt and milk,smoking,alcohol consumption,hypertension,regular use of angiotensin Ⅱ receptor blocker(ARB)and aspirin,mean corpuscular hemoglobin(MCH),mean corpuscular hemoglobin concentration(MCHC),neutrophil-to-lymphocyte ratio(NLR),lymphocyte-to-monocyte ratio(LMR),total protein(TP),globulin(GLB),total cholesterol(TC)and direct bilirubin(DBIL)were associated with colorectal neoplasia.Multivariate logistic regression analysis showed that gender,age,intake frequency of fruit,NLR and DBIL were independent factors for colorectal neoplasia.Gender,age,intake frequency of fruit,NLR,DBIL and family history of CRC were included in the model,which the visualization model of nomogram was established.The AUC of the model was0.73(95%CI 0.68-0.78),and the AUC of bootstrap sampling was 0.72.The calibration curve and Hosmer-Lemeshow test(P=0.251)showed that the model fitted well.DCA showed that the model had good clinical benefit.Nomogram model score of colorectal neoplasia higher than 103.5 points was defined as high risk,while lower than 103.5 was defined as low risk.The detection rate of colorectal neoplasia in the high-risk group was66.84%,which was significantly higher than that in the low-risk group(26.78%),indicating that this risk stratification could effectively detect colorectal neoplasia in the screening population.2.For advanced colorectal neoplasia,univariate analysis showed that gender,age,intake frequency of fruit and yogurt,regular use of calcium channel blocker(CCB),family history of CRC,family history of CRC in a first-degree relative,carcinoembryonic antigen(CEA),red blood cell volume distribution width standard deviation(RDWSD),mean platelet volume(MPV),systemic immune inflammation index(SII),NLR,LMR,TP,GLB,DBIL and creatinine(Cr)were associated with advanced colorectal neoplasia.Multivariate logistic regression analysis showed that gender,age,DBIL and Cr were independent factors for advanced colorectal neoplasia.Gender,age,DBIL,Cr,and family history of CRC were included in the model,and a nomogram visualization model was established.The AUC of the model was 0.79(95%CI 0.74-0.84),and the AUC of bootstrap sampling was 0.78.The calibration curve and Hosmer-Lemeshow test(P=0.807)showed that the model fitted well.DCA showed that the model had good clinical benefit.Nomogram model score of advanced colorectal neoplasia higher than 122.6 points was defined as high risk,while lower than122.6 was defined as low risk.The detection rate of advanced colorectal neoplasia in the high-risk group was 46.30%,which was significantly higher than that in the low-risk group(9.36%),indicating that this risk stratification could effectively detect advanced colorectal neoplasia in the screening population.Compared with APCS,the nomogram model showed better discrimination(0.73 vs 0.79,P=0.004)and clinical benefit(same clinical decision threshold probability).There was no significant difference in the detection rate of advanced colorectal neoplasia between nomogram model and APCS high-risk group(46.30% vs38.38%,P=0.211).The comparison of non-advanced colorectal neoplasia rate between nomogram model and APCS low-risk group(90.64% vs 91.00%,P=0.915)wasn’t significant.Conclusions:1.The nomogram model based on gender,age,intake frequency of fruit,family history of CRC,NLR and DBIL could be used as a practical model to evaluate the risk of colorectal neoplasia,which had good discrimination,fit degree and clinical benefit.The model could effectively stratify the risk of colorectal neoplasia.2.The nomogram model based on gender,age,family history of CRC,DBIL and Cr could be used as a practical model to evaluate the risk of advanced colorectal neoplasia,which had good discrimination,fit degree and clinical benefit.The model could effectively stratify the risk of advanced colorectal neoplasia.