| Over the years,the health examination has accumulated a large number of health index measurement data,forming longitudinal physical examination cohort data.Cohort data are usually composed of survival data and longitudinal data,that is,there is a close relationship between longitudinal observation process and survival time.When studying longitudinal cohort data,the information in the data will be lost if only a single longitudinal model or survival model is established.In order to make the prediction results closer to the data itself and more accurate,the longitudinal process and survival results should be included in the model at the same time.In this study,10 consecutive years of physical examination data were collected to establish a longitudinal physical examination cohort,and chronic disease risk assessment methods were studied based on cross-sectional and longitudinal data respectively.In the dynamic analysis of chronic disease,it was found that the prevalence of fatty liver increased continuously.In order to evaluate the application effect of the chronic disease risk assessment method in this paper,the research method was applied to the prediction of fatty liver disease.This study mainly did the following work:(1)The physical examination data for 10 consecutive years were collected,the health examination cohort was established,and the detection of chronic diseases was dynamically analyzed.According to the data of different variables,different ways are used to clean the data to improve the quality of the data.Manual error value modification and outlier capping method were used to deal with outliers.Last observation advance method and algorithm were used to deal with missing values.(2)Based on cross-sectional data,random forest algorithm,Lasso Logistic regression model and index calculation method were adopted to conduct variable screening,and the screening effects were compared.Taking fatty liver as an example,the risk factors were screened out mainly including body mass index,hypertension,etc.,which provided an important basis for the research of chronic disease risk assessment methods based on cross-sectional data.The risk grade index of fatty liver was calculated using the Harvard Cancer Index method,and the population was divided into seven grades from "very high risk" to "very low risk".The higher the risk grade,the higher the prevalence was(x~2=1174.020,P<0.001).The predictive model of fatty liver was established by synthetic analysis method,and its predictive performance was evaluated by sensitivity,specificity,Youden index and other indexes.(3)Study chronic disease risk assessment methods based on longitudinal data.Taking fatty liver as an example,six machine learning algorithms,including decision tree,support vector machine,XGboost,Bagging,random forest and artificial neural network,were adopted to establish a prediction model,and the prediction performance was explored through simulation experiments.It was verified that the area under the ROC curve of the fatty liver prediction model established by XGBoost algorithm was 0.958,the recall rate was0.790,the accuracy rate was 0.761,and the accuracy rate was 0.898,which were all higher than other machine learning algorithms.The Joint Joint model was established by combining the XGBoost longitudinal submodel with the time-dependent Cox survival submodel.The stability and prediction effect of the Joint Joint model under different sample sizes were explored through simulation experiments and case analysis.The results show that the fitting effect and prediction performance of the Joint model established by XGBoost longitudinal submodel and Cox survival submodel are better than that of the single model,and also better than that of the Joint model established by other machine learning methods and survival function. |
