| Objectives: 1.To understand the prevalence and trend of dyslipidemia(DLP)in the medical examination population in a hospital in Guilin from2011 to 2017.2.Establishing logistic regression,classification tree and radial basis function(RBF)neural network risk prediction model s and analyzing the predictive value of different models in dyslipidemia to provide effective tools for the early diagnosis of dyslipidemia.Methods: 1.Analyze the sex,age and blood lipid profile of 194045 adults who went to the physical examination center of a hospital in Guilin from 2011 to 2017,and evaluate the prevalence trend of dyslipidemia in 7 years.2.A cross-sectional study was conducted on 10345 peop le who had physical examination in a hospital physical examination center in Guilin from July to October 2017.The data of gender,age,height,weight,blood pressure and blood lipid were collected.Through the simple random sampling method,about 70% of the population is selected as the training set,with gender,age,blood pressure,body mass index(BMI)as the independent variable,and dyslipidemia as the dependent variable,and establish logistic regression,classification tree and RBF neural network pr ediction model,then the corresponding nomogram of multivariate logistic regression model was drawn.The remaining 30 % of the population was used as the test set to evaluate the prediction effect of the model.3.Establish the receiver operating characteri stics(ROC)curve,and evaluate the prediction effect of the model according to the ROC curve.Results: 1.From 2011 to 2017,the prevalence of dyslipidemia was 46.7%,22.2% and34.7% in men,women and standardized subjects,respectively.After standardization,the prevalence of dyslipidemia increased from 201 1 to2015,and decreased and stabilized from 2016 to 2017.The peak prevalence of dyslipidemia was 50-69 years old in the physical examination population.The prevalence of dyslipidemia in men increase d with age,reached the peak in 40-59 years old,and then decreased with age.The prevalence of dyslipidemia in women increased with age.2.Logistic regression,classification tree and RBF neural network risk prediction model showed that gender(male),hy pertension,age and BMI were risk factors of dyslipidemia.3.The prediction effect s of the three models are medium.In the training set,the sensitivity,specificity,area under the ROC curve and Youden index of logistic regression model were 71.29%,65.32%,0.740 and 36.6% respectively.According to the sex,age,blood pressure and BMI of the subjects,the specific scores could be read on the corresponding nomogram,so as to get the specific probability of dyslipidemia of the subjects.The sensitivity,sp ecificity,AUC and Youden index of the classification tree model were 80.73%,56.34%,0.744 and 37.1%,respectively.The curve sensitivity,specificity,AUC and Youden index of RBF neural network model were 70.78%,65.86%,0.736 and 36.6%,respectively.In the test set,the sensitivity,specificity,AUC and Youden index of logistic regression model were 77.41%,54.01%,0.704 and 31.42% respectively;the sensitivity,specificity,AUC and Youden index of classification tree model were 80.47%,56.99%,0.700 and 37.46% respectively;the sensitivity,specificity,AUC and Youden index of RBF neural network model were 78.67%,55.40%,0.704 and 34.07% respectively.Conclusions: 1.The prevalence of dyslipidemia was high in Medical Examination Population in a hospital in Guilin from2011 to 2017;Among them,the prevalence of dyslipidemia was highest in50-69 years old,and the prevalence of dyslipidemia in men was higher than that in women.2.Logistic regression,classification tree and RBF neural network prediction models used low-cost,easy-collected,non-invasive objective indicators(gender,age,blood pressure and BMI)to predict the risk of dyslipidemia,The diagnostic value were all medium and could be used in clinical.In particular,the nomogram correspondin g to logistic regression model is simple,easy to understand and fast,which has great practical value. |
PDF Full Text Request