| Objective:This study aimed to adopt the data mining method to construct lung cancer prediction models based on plasma P53-related miRNAs,so as to provide reference for the auxiliary diagnosis of lung cancer.Materials and Methods:1.Research objects:The 106 patients with primary lung cancer who attended the First Affiliated Hospital of Zhengzhou University,Henan Cancer Hospital and Henan Chest Hospital from June 2016 to September 2019 were selected as the case group.A total of 106 people who attended a physical examination at a Henan Center for Disease Control and Prevention during this period were selected as controls.After informed consent,the peripheral blood of the subjects was collected and related information was collected.2.Plasma P53-related miRNAs screening and expression level detection:By searching the PubMed database to screen P53-related miRNAs,7 types of miRNAs(miR-34a,miR-29a,miR-125b,miR-145,miR-192,miR-215,miR-194)were included in the study.Fluorescence quantitative PCR method was used to detect the expression levels of 7 plasma miRNAs.3.Statistical analysis:Excel 2016 establishes the database.SPSS 27.0 carries on the statistical analysis.Quantitative data conforms to the normal distribution using the two independent sample t test,and does not conform to the normal distribution using the Mann-Whitney U test.The qualitative data uses the chi-square test.Logistic regression screens the risk factors related to lung cancer.Based on the results of logistic regression screening,Clementine 20.0 software was used to build decision tree C5.0,support vector machine,artificial neural network,Bayesian method and Fisher discriminant analysis lung cancer prediction model.Model evaluation indicators include sensitivity,specificity,positive predictive value,negative predictive value,accuracy,and area under the receiver operating characteristic curve(AUC).Significance level α=0.05.Results:1.Plasma P53-related miRNAs expression level detection results:The expression levels of miR-34a,miR-29a,miR-125b,miR-192,miR-215,and miR-194 in the plasma of lung cancer patients are lower than those of the control group.The expression level of miR-145 is higher than that of the control group,and the differences were statistically significant(P<0.05).2.Results of analysis of factors affecting lung cancer:The results of univariate analysis showed that the differences in the distribution of smoking,drinking,fever,chest pain,cough,sputum,and hemoptysis between the lung cancer group and the control group were statistically significant(P<0.05).The differences in the expression levels of 7 miRNAs in plasma were also statistically significant(P<0.05).Logistic regression analysis showed that smoking(OR=3,303,95%CI=1.145～9.528),fever(OR=11.523,95%CI=1.781-74.553),cough(OR=7.770,95%CI=2.529～23.869),miR-29a(OR=0.478,95%CI=0.310～0.738),miR-125b(OR=0.699,95%CI=0.545～0.897),miR-145(OR=1.883,95%CI=1.304-2.719),miR-194(OR=0.768,95%CI=0.601～0.981)are related factors of lung cancer.3.Evaluation results of the data mining model:According to the logistic regression screening results,smoking,fever,cough,miR-29a,miR-125b,miR-145,miR-194 were included in the data mining model.The prediction accuracy of the Fisher discriminant analysis model was 77.19%,the sensitivity and specificity were 78.57%and 68.96%,respectively,and the AUC was 0.738(0.604,0.845).The prediction accuracy of the decision tree C5.0 model is 94.74%,the sensitivity and specificity are 96.42%and 93.1%,respectively,and the AUC is 0.948(0.854,0.989).The prediction accuracy of the support vector machine model is 91.23%,the sensitivity and specificity are 85.71%and 96.22%,respectively,and the AUC is 0.911(0.806,0.970).The prediction accuracy of the artificial neural network model is 87.72%,the sensitivity and specificity are 92.85%and 82.75%,respectively,and the AUC is 0.878(0.764,0.950).The prediction accuracy of the Bayesian model is 84.21%,the sensitivity and specificity are 78.57%and 89.65%,respectively,and the AUC is 0.841(0.720,0.924).Conclusion:1.The expression levels of P53-related miRNAs(miR-34a,miR-29a,miR-125b,miR-145,miR-192,miR-215,miR-194)in human plasma may be related to lung cancer.2.Decision tree C5.0,support vector machine,Bayesian,artificial neural network lung cancer prediction model and Fisher discriminant analysis model were established,based on smoking,fever,cough,and the expression levels of plasma miR-29a,miR125b,miR-145,and miR-194.Among them,the decision tree C5.0 model has the best effect on lung cancer prediction. |
