Variable Selection Of Complex Data Joint Model Based On Improved Lasso Method

Complex data such as longitudinal and time-to-event data often appear in the fields of medical research and psychology research.The joint model of linear mixed effect model and Cox regression model established for this kind of data has been widely used.However,as the number of data observation variables increases,the problem of variable selection in the joint model is particularly important.When the traditional variable selection method can not meet the conditions for constructing the ideal model,the coefficient compression method begins to show its head.In this paper,we construct a joint model of linear mixed-effects model and Cox regression model with frailty based on shared random effects.Based on Lasso penalty method and its improved method,Elastic Net penalty method,the constructed model is constructed.The variables are selected and a sparse estimate of the regression coefficients is obtained.Through numerical simulation research and case analysis,the following conclusions are obtained for the model constructed in this paper:(1)In the process of variable selection,the Lasso and Elastic Net methods automatically compress the coefficients with smaller absolute values to 0 to simultaneously achieve the target of variable selection and estimation of the parameters to be estimated.(2)For the longitudinal and time-to-event data with Multicollinearity problems,both the Lasso method and the Elastic Net method can achieve the selection of significant variables and the estimation of parameters.(3)The model constructed for this paper: In the streamlined aspect of the model,the model obtained by the Lasso method is more concise than the Elastic Net method,that is,the number of variables selected by the Lasso method is less;the Elastic Net method is used in the fitting accuracy of the selected model.May be better than the Lasso method.However,the gap between the two is very small,both in terms of the streamlined aspect of the model and the model fitting accuracy of the selected model.(4)For the data with group effect,the Elastic Net method can make up for the deficiency of the Lasso method,and select the group effect data that has a significant influence on the response variable.