Comparative study on nomogram and machine learning algorithms for predicting dental caries in middle-aged and elderly people
Objective To compare the efficiency of nomogram and different machine learning algo-rithms for constructing the dental caries predictive models for middle-aged and elderly people.Methods The multi-stage stratified random sampling method was used to select 510 middle-aged and elderly people from Nanning City,Guigang City and Chongzuo City as the research subjects for conducting the questionnaire sur-vey and oral cavity examination.The univariate analysis and Lasso regression were used to screen the related variables,and the multivariate logistic regression analysis was used to determine the final independent influen-cing factors.Based on the salient features,the nomogram predictive model was established,and the seven ma-chine learning algorithms,including linear discriminant analysis (LDA),partial least squares (PLS),range Doppler algorithm (RDA),generalized linear models (GLM),random forest (RF),support vector machine (SVM) kernel function (SVM-Radial),and SVM linear kernel function (SVM-Linear),were used to construct the seven kinds of dental caries risk predictive models.The area under the receiver operating characteristic (ROC) curve (AUC) was adopted to evaluate the predictive performance of various models and the predictive performance of models constructed using different variable screening methods.Results The detection rate of dental caries in middle-aged and elderly people was 71.18％.After feature screening,the five predictive factors were ultimately retained,which were the age (OR=0.945,95％CI:0.917-0.973),brushing frequency (OR=0.688,95％CI:0.475-0.997),whether having teeth cleaning in the past one year (OR=0.303,95％CI:0.103-0.890),number of remaining teeth (OR=1.062,95％CI:1.038-1.087) and oral health assess-ment tool (OHAT) score (OR=1.363,95％CI:1.234-1.505).The results of comparison of various models showed that the predictive model constructed by the RF algorithm performed the best,the median of AUC was 0.747,followed by the nomogram,and the median of AUC was 0.733.The median of AUCs in the predic-tion model constructed by single factor+Lasso+multivariate logistic (Lasso+logistic) screening independent variables were higher than those constructed by RF algorithm screening independent variables.ConclusionBased on Lasso+logistic screening variables,RF provide more reliable predictive efficiency in predicting dental caries in middle-aged and elderly people than nomogram and the other machine learning algorithms.
middle-aged and elderly peopledental cariespredictionmachine learningcolumn diagram
万方数据
维普
