| In clinical medicine and epidemiological research,individuals are usually tracked for a period of time to obtain longitudinal indicators of individuals and relevant survival data.In early studies,longitudinal and survival data were often modeled separately for such data,which ignored the correlation between the two parts.Therefore,joint modeling method of longitudinal and survival data was proposed.The longitudinal measurement of multiple variables and the data of event time need to consider the problem of variable selection.Traditional variable selection methods have many defects in practical application,such as low computational efficiency.In comparison,Lasso and other penalty likelihood methods are effective.In this article,we develop penalized likelihood methods to select variables in the joint model of longitudinal and survival data.Our model consists of a linear mixed effects model for the longitudinal outcome and a Cox proportional risk model for survival submodel,linked together by shared random effects.We used Lasso penalty method to select fixed and random effects of the joint models,the variances of parameterized random effects were decomposed by Cholsky and variable selection was achieved by a two-stage method.Finally,we selected AIDS data to built a joint model of shared random effects,and used Lasso method to select variables.Compared with maximum likelihood estimation,the Lasso method selects significant variables by compressing coefficients,which is more efficient and faster than the method using P-values.In model construction,Lasso method has better fitting effect than SCAD method,and the model obtained is more concise. |
