Abstract:
|
Longitudinal data are often collected in modern medical studies. With the improvements of technology, researchers are able to collect information on an increasing number of predictors which presents the statistical challenge of variable selection. We propose a three-stage, model-based method to select informative factors and two-way interactions which is crucial for subgroup identification in longitudinal clinical trials. At the first step, we use marginal score tests to select variables associated with the longitudinal outcome. At the second stage, we use covariance-insured screening to identify variables associated with those selected during the first step. In the third step, we apply a penalized LASSO method using the variables in Steps 1 and 2 to obtain all informative variables and their interactions. Simulation studies are conducted to evaluate the performance of the proposed method. A longitudinal clinical trial study is used to illustrate the proposed method.
|