Adaptive analyses are presented of blood lead levels for children at 0, 1, 4, and 6 weeks using exponential regression with the natural log link function. The choice of the number of folds is addressed as well as the choice of the directly specified correlation structure. Results are compared for partially modified generalized estimating equations (GEE), fully modified GEE, and extended linear mixed modeling (ELMM). Linearity of the log of the means in week with constant dispersions is addressed as well as a comparison to standard GEE modeling and the dependence of means and dispersions on week. Adaptive additive and adaptive moderation models are generated for week and the indicator for being on the chelating agent succimer. Direct variance modeling of the blood lead levels is addressed. Models based on directly specified correlation structures are compared to models based on random effects/coefficients. A summary of the analysis results is provided. SAS code for generating these analyses is described along with output generated by that code.

错误:搜索内容不能为空,请输入英文关键词
错误:关键词超出字数限制,请精简
高级检索

Example Analyses of the Blood Lead Level Data

  • George J. Knafl

摘要

Adaptive analyses are presented of blood lead levels for children at 0, 1, 4, and 6 weeks using exponential regression with the natural log link function. The choice of the number of folds is addressed as well as the choice of the directly specified correlation structure. Results are compared for partially modified generalized estimating equations (GEE), fully modified GEE, and extended linear mixed modeling (ELMM). Linearity of the log of the means in week with constant dispersions is addressed as well as a comparison to standard GEE modeling and the dependence of means and dispersions on week. Adaptive additive and adaptive moderation models are generated for week and the indicator for being on the chelating agent succimer. Direct variance modeling of the blood lead levels is addressed. Models based on directly specified correlation structures are compared to models based on random effects/coefficients. A summary of the analysis results is provided. SAS code for generating these analyses is described along with output generated by that code.