Abstract:

In order to study the medical longitudinal data modeling,the varying coefficient was used to explore the relationship between time-varying detection index and tumor size, and two-step iterative procedure based on basis expansion and adaptive-LASSO penalty least square method was proposed. This method could estimate the coefficient and meanwhile separate the variable coefficient, constant coefficient and zero coefficient. This method was applied to study the cancer data, to estimate the cofficient of each variable, to study the effect of symptom to the progression of tumor size in different time, and to predict the tumor size. The average relative error of model showed that the simulation effect was well.

Key words: adaptive-LASSO penalty, coefficient estimation, varying coefficient model, variable selection, longitudinal data, B-spline