变系数模型在医学纵向数据研究中的应用

The application of varying coefficient model in the study of medical longitudinal data

为了对医学纵向数据进行建模研究,利用变系数模型探索时变的检测指标与时变的肿瘤大小之间的映射关系,提出基于B样条估计与adaptive-LASSO惩罚最小二乘结合的两步迭代法进行系数估计与变量选择,该方法在系数估计的同时分离变系数、常系数和零系数。将变量选择方法应用于医学肿瘤数据中,对各个变量进行系数估计,研究不同时刻的症状对肿瘤大小进展的影响,并最终对肿瘤大小进行预测。平均相对误差的计算结果表明该模型模拟效果较好。

In order to study the medical longitudinal data modeling, the varying coefficient was used to explore the rela- tionship 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 da- ta, 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.