|
Abstract:
|
Research on air pollution’s effect on dementia-related outcomes has proliferated. Many studies have leveraged large established cohorts or administrative databases. Large sample sizes can reduce random error in estimating the effects of air pollution on these outcomes, but potential for systematic error remains. Selection bias is of particular concern in these studies, which typically impose an advanced baseline age (e.g., 65+ years) for inclusion. This means that, over follow-up, participants are at high risk of attrition due to death or illness that makes it difficult to engage in the study protocol. Less appreciated is that these attrition processes also operate prior to the age of first observation, yielding a sample that is already highly selected at baseline. If these selection processes are jointly related to air pollution exposure and incipient dementia, both of which affect morbidity and mortality, the estimated effect of air pollution can be biased. We demonstrate how to apply inverse probability weighting to mitigate bias in these estimates from differential post-baseline attrition, and outline an approach for quantifying bias from differential pre-baseline attrition.
|
