| ObjectivesTo construct the risk prediction model of colorectal cancer,early detect high-risk groups of colorectal cancer,and achieve accurate screening and accurate behavior intervention for high-risk groups.Methods1.Big data research:through the front page data of the medical records of patients with colorectal cancer in Sichuan Province to study the characteristics of the disease.2.Literature research:relevant literatures were searched in Chinese and English databases,and the risk factors of colorectal cancer were determined by meta analysis.3.Questionnaire survey:referring to the risk factors of colorectal cancer suggested/determined by big data research and meta analysis,we developed a questionnaire to assess the risk factors of colorectal cancer.4.Case control study:the cases were from 5 hospitals in Sichuan Province,and the controls were from the physical examination center of Sichuan Cancer Hospital and two communities.5.Machine learning:compared with the traditional logistic regression,a variety of data mining methods(including k-nearest neighbor,decision tree,support vector machine,neural network)were used to build the risk prediction model.Results1.Risk factors of colorectal cancer(1)Characteristics of diseaseThe average 5-year prevalence of colorectal cancer in Sichuan Province was 24.5/100000 in 2015-2019,14.6/100000,24.1/100000,36.1/100000,41.1/100000,and 46.2/100000 respectively in Sichuan Province in 2015-2019;The proportion of patients in age group≥60,male,Han nationality,farmer and married were the highest.The highest 5-year prevalence rate in Chengdu is 39.4/100000,the lowest in Ganzi Prefecture is 6.0/100000,and the prevalence rate in surrounding areas of Chengdu has increased year by year in the past five years.(2)Meta analysis of risk factors of colorectal cancerThe OR values of≥60 years old,male,black,BMI≥25kg/m2,WHR≥0.85,WHtR≥0.5,family history,diabetes,hypertension,0-16 teeth,smoking,menarche age≥15 years old,premenopausal status,pregnancy,multiple pregnancies were greater than 1;The OR value of college education or above,better economic status,colonoscopy,statins,NSAIDs,aspirin,Mediterranean diet and healthy lifestyle,hormone replacement therapy,fertility,hysterectomy were less than 1.2.risk prediction model of colorectal cancer(1)Total populationThe AUC of risk models using Logistic regression,KNN,DT,SVM and NNs were 0.86,0.70,0.76,0.78,0.73 and the accuracy were 0.78,0.70,0.76,0.78,0.73 in order.(2)MaleThe AUC of risk models using Logistic regression,KNN,DT,SVM and NNs were 0.75,0.69,0.74,0.70,0.63,and the accuracy were 0.73,0.71,0.79,0.73,0.64 in order.(3)WomenThe AUC of risk models using Logistic regression,KNN,DT,SVM and NNs were 0.72,0.75,0.74,0.77,0.68,and the accuracy were 0.56,0.80,0.72,0.78,0.70 in order.Conclusions①Logistic regression method was the best method to establish the risk prediction model of colorectal cancer in the general population.The final results included personal income,hyperlipidemia,intestinal polyps,NSAIDs,heavy fat food,spicy food,milk,drinking,intensity of exercise,duration of exercise and physical strength at work.②The decision tree method was the best method to establish the risk prediction model of male colorectal cancer,which included physical strength,intestinal polyps,duration of exercise and annual income.③SVM was the best method to establish the risk prediction model of female colorectal cancer.Marriage status,personal income,family history,hyperlipidemia,intestinal polyps,number of teeth,lipid-lowering drugs,NSAIDs,spicy food,edible oil,red meat,milk,nuts,folic acid,calcium,sugary drinks,exercise and intensity of exercise,duration of exercise,physical strength at work,age of first pregnancy,number of pregnancies,abortion,number of abortions,menopause type were included.④In conclusion,the risk assessment tool of colorectal cancer by gender was initially constructed. |
PDF Full Text Request