|
Abstract:
|
Marginal structural models (MSMs) are used to estimate the causal effect of an exposure history on an outcome when time dependent confounding is present in longitudinal, observational data. Missing data is common in longitudinal studies and here we explore the situation when missing data occurs in the exposure of interest. We present the results of simulations based on the Framingham Heart Study data set designed to compare four missing data methods for dealing with missing data, under seven possible missing data structures. The results suggest that two of the methods, a complete case analysis, in which all subjects with missing data are removed from the analysis, and an analysis that censors individuals at the first occasion of missing data, both provide little bias. We discuss the results and the which missing data structures provide the opportunity for bias using causal diagrams.
|
