|
Abstract:
|
In this study we want to investigate the association between loosing a partner and cognitive decline, using data from a large population based study where participants are followed over 15 years. One complication is that longitudinal data of older cohorts are prone to dropout due to illness, e.g. dementia, or death. Hence, the non-response may be directly related to the cognitive outcome of interest in which case it is not ignorable. An additional consequence of the existing dropout, is that, not only the outcome, but information regarding family status may also be missing, and in which case exposure is not observed. In the presence of such incomplete data scenarios, conventional statistical methods are invalid and lead to biased estimates. We therefore develop a modeling strategy within a Bayesian framework to deal with these issues, and provide inference on the effect of a (time-varying) binary exposure on a longitudinal outcome in the presence of dropout and death.
|
