Regency EF
Modeling Longitudinal Change in Biomarkers in the Presence of Disease Treatment with Application to the Atherosclerosis Risk in Communities (ARIC) Study (304078)
*Nicole Butera, The George Washington University
Jianwen Cai, The University of North Carolina at Chapel Hill
Gerardo Heiss, The University of North Carolina at Chapel Hill
Donglin Zeng, The University of North Carolina at Chapel Hill
Keywords: Biomarkers, doubly robust, inverse probability weighting, longitudinal, treatment
There is often interest in modeling natural course (i.e., untreated) longitudinal biomarker change and its association with other factors. In observational studies, since participants may start treatment over the course of the study, observed post-treatment biomarker values may differ from values that would be observed if untreated. Excluding treated participants may bias effect estimates if treated participants differ systematically from untreated participants. We considered post-treatment biomarker values as missing data, and applied missing data methods (inverse probability weighting and doubly robust estimation). Our approach uses data commonly collected between visits in longitudinal cohort studies on treatment initiation, by modeling the treatment mechanism using survival analysis methods. We will present the statistical properties for this approach, and illustrate finite sample performance via simulation studies. We will illustrate this approach by modeling the association between body mass index and blood pressure change in the Atherosclerosis Risk in Communities Study. Lastly, we will provide guidance for implementing these approaches in statistical practice.