Modeling Biomarkers in the Presence of Medication Use Endogenous to Pre-Treatment Biomarker Values

• Activity Number: 634
• Type: Contributed
• Date/Time: Thursday, August 4, 2011 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Robyn McClelland*+
• Companies: University of Washington
• Keywords: medication use ; blood pressure ; cholesterol ; censoring ; simultaneous equation

Abstract

Suppose we are interested in understanding the predictors of a biomarker whose value may be altered by medication use. Examples include cholesterol and blood pressure, which are of biological importance and are frequently lowered by medications. Traditional methods produce biased estimates, and do not incorporate the fact that medication use is endogenous. That is, the decision to place a participant on medication will depend on their risk factors and underlying pre-treatment value for the biomarker in question. In an epidemiologic study with many participants on treatment at baseline, that pre-treatment value is unobserved. I propose likelihood-based estimates that incorporate important features of the model: endogeneity of medication use with respect to risk factors and the latent pre-treatment biomarker value; and right-censoring of the observed biomarker value by medication use. Simulation studies will demonstrate that the model produces unbiased estimates and good coverage for key parameters of interest. Data from NHANES on blood pressure trajectories will be presented that yield new insights compared with traditional models.