Our data show that other behavioral factors, considered
as potential confounders, do not attenuate the
association between depression and stroke incidence.
For example, although persons with a high level of
depressive symptoms in our study were more likely to
have low levels of nonrecreational physical activity,
adjusting for this factor as well as other risk factors did
not significantly change the observed association.
However, these covariates may not have been sufficiently
modeled. In particular, alcohol consumption in
the past year was modeled dichotomously (none vs.
any use) and thus may lack the sensitivity to fully
detect variation in stroke incidence. Similarly, adjusting
for preexisting heart disease or diabetes did not
substantially alter the depression-stroke association.
These conditions were generally significant covariates
in our models. However, these baseline measures were
self-reported and thus may not be as accurate as other
measures, such as medical records from healthcare
facilities. Furthermore, although the GWB-D is well
validated, a score indicating a high level of depressive
symptoms is not synonymous with a clinical diagnosis
of depression. The prospective association between a
clinical diagnosis of depression and stroke also needs
to be studied. Confounding by variables not measured
cannot be excluded.