Development Of Colorectal Cancer Risk Prediction Model And Its Validation In Colorectal Cancer Screening Population

Background:Colorectal cancer(CRC)is one of the most common type of malignant tumor and one of the most severe public health problems worldwide.The incidence and mortality rates of CRC rank the 3rd and 5th among all types of cancers in China,with increasing trends in recent years,respectively.Due to the lack of CRC screening nationwide,most patients with CRC have progressed to the middle and advanced stages at the time of diagnosis,which has led to a 5-year survival rate of less than 50%.If these patients can be diagnosed and treated at an early stage,the 5-year survival rates can be greatly improved to approximately 90%.Therefore,the early detection and diagnosis of CRC is very important.Domestic and foreign scholars have tried to establish a variety of CRC risk prediction models for identifying risk factors,screening high-risk groups and assessing the risk of onset,to provide personalized screening programs for different risk groups.However,the current established models are mostly based on a local sample,and whether they are applicable to the external population has yet to be verified.Therefore,building an efficient and easy-to-use CRC risk prediction model and finding the true high risk population has significant clinical and public health importance in precisely implementing CRC screening and reducing the burden of CRC.Objective:To build an efficient and easy-to-use model for predicting CRC risk based on a large-scale population-based cohort study in China and prospectively validated in a population-based CRC screening cohort,and to optimize the population-based CRC screening program in combination with existing mature screening technologies.Methods:1.Construction and internal validation of the CRC risk prediction model(1)Based on a biennial survey and health checkup of the Kailuan Cohort(2006-2015),data were collected including demographic,lifestyle,height,weight,waist circumference and other measurable variables at baseline and outcomes of CRC during follow-up.In addition,data linkage was used to identify missing new cases of CRC during follow-up.(2)The data set was randomly divided into a training set and a validation set(36 403 participants each)according to a ratio of 1:1.The candidate factors were included in a multivariate logistic regression model for the training set,and stepwise regression was used(P<0.20 was used as the selection criterion)to identify the independent predictors of CRC.By multiplying the ? coefficient in the regression equation by 3 and rounding to an integer,the corresponding score was.calculated and then a CRC risk prediction model was established.The model goodness of fit was tested by the Hosmer-Lemeshow method and validated by nonparametric bootstrapping and K-fold cross validation.(3)A receiver operating characteristic(ROC)analysis was performed to calculate the area under the curve(AUC)and 95%confidence interval(95%CI)according to the CRC risk score for each subject.The optimal cut-off value was established according to the ROC curve,and then the subjects were categorized as being at low-risk or the high-risk according to the risk of CRC.The sensitivity,specificity and Youden index of the CRC risk prediction model were calculated.2.External validation of the CRC risk prediction model and optimization of the CRC screening program(1)Based on the baseline data from a multicenter CRC screening cohort(the Target-C Study),a risk score was calculated for each subject according to the CRC risk prediction model,and the colonoscopy examination and fecal immunochemical test(FIT)were performed for each subject as well.A multivariate logistic regression model was used to evaluate the factors affecting the participation rate of colonoscopy screening,and the odds ratio(OR)and 95%CI of the corresponding factors were calculated.(2)ROC analysis was performed to calculate the AUCs and 95%CIs for both the CRC risk prediction model and FIT by using CRC,advanced adenoma,non-advanced adenoma,advanced neoplasm,and any neoplasm as individual endpoint,respectively.The diagnostic efficacy of the two tests alone and combined was further evaluated.The diagnostic efficacy was evaluated by the sensitivity,specificity,Youden index,etc.Results:1.Construction and internal validation of the CRC risk prediction model(1)During the 9-year follow-up,378 cases of CRC occurred in the Kailuan Cohort.After variable screening and model optimization,the CRC risk prediction model established in the training set consisted of three variables:age,alcohol consumption,and history of diabetes.The scores ranged from 0 to 7 and the predictions were consistent(P=0.860).The distinguishing ability is moderate(AUC=0.625,95%CI=0.586-0.665),but it is significantly better than other similar prediction models.Taking a score of 3 points as the cutoff value,the subjects in the high-risk group had a 2.04-fold(95%CI=1.52-2.74)higher risk of CRC than those in the low-risk group.The sensitivity,specificity,and Youden index were 56.04%,61.59%,and 0.18,respectively.(2)Based on the CRC risk prediction model established in the training set,a CRC risk score was calculated for each subject in the validation set.The prediction consistency was good(P=0.571),but the distinguishing ability was reduced(AUC=0.578,95%CI=0.539-0.618).Taking a score of 3 points as the cut-off value for screening,the subjects in the high-risk group had a 1.60-fold(95%CI=1.21-2.11)higher risk of CRC than those in the low-risk group.The sensitivity,specificity,and Youden index were 51.02%,60.52%,and 0.12,respectively.2.External validation of the CRC risk prediction model and optimization of the CRC screening program(1)In total,3 723 people met the inclusion criteria and a questionnaire survey was conducted.Among them,1 665 people completed the colonoscopy examination and the participation rate was 44.72%.Multivariate logistic regression analysis indicated that age,education,drinking status,and history of hypertension were associated with the participation rate of colonoscopy screening.(2)Overall,there were 9 cases of CRC,85 cases of advanced adenoma,312 cases of non-advanced adenoma,94 cases of advanced neoplasm,and 406 cases of any neoplasm.The yielded detection rates for CRC,advanced adenoma,non-advanced adenoma,advanced neoplasm,and any neoplasm were 0.54%,5.11%,18.74%,5.65%,and 24.38%,respectively.The detection rate in men is higher than that in women.In addition to CRC,the detection rates of advanced adenoma,non-advanced adenoma,advanced neoplasm,and any neoplasm showed increasing trends with age.(3)According to the CRC risk prediction model established in the training set,the risk score of colorectal advanced neoplasm was calculated for each study subject in the external validation set,with an AUC of 0.582.Taking a score of 3 points as the cut-off value for screening,the subjects in the high-risk group had a 1.70-fold(95%CI=1.12-2.57)higher risk of advanced neoplasm than those in the low-risk group.The sensitivity,specificity,and Youden index were 50.00%,62.89%,and 0.13,respectively.The number of colonoscopies needed to detect one advanced neoplasm was 13.40.If only colonoscopy screening is performed in high-risk populations,50%of cases of advanced neoplasm can be detected through screening,with a 62%reduction in the number of colonoscopy examinations performed.(4)When the CRC risk prediction model was combined with the FIT(CRC risk score>3 or FIT value?100?g Hb/g was identified to be at high risk for colorectal lesions),the subjects in the high-risk group had a 2.01-fold(95%CI=1.29-3.13)higher risk of advanced neoplasm than those in the low-risk group.Compared with the FIT alone,the sensitivity to advanced neoplasms after the parallel connection was increased from 9.52%to 54.76%.The number of colonoscopies needed to detect one advanced neoplasm was 12.61.If only colonoscopy screening is performed in a high-risk population,55%of cases of advanced neoplasm can be etected through screening,with a 61%reduction in the number of colonoscopy examinations performed.Conclusion:This study explored the potential risk factors for CRC based on a prospective cohort study,established a CRC risk prediction model,and validated it in a screening population.The CRC risk prediction model based on age,alcohol consumption and history of diabetes is useful for determining the risk of advanced neoplasms in asymptomatic people.The CRC risk prediction model combined with the FIT has higher accuracy in predicting the risk of colorectal lesions in asymptomatic people than the two tests alone,which can greatly reduce the number of colonoscopies performed.It is of great significance for efficiently conducting CRC screening and reducing the disease burden of CRC.Affected by the population mobilization and screening participation rate,the effectiveness and accuracy of the combined CRC risk prediction model and FIT screening should be further verified in larger population screening projects in the future.