The models adjusted for such potential
confounding variables as gender, age, race/
ethnicity, household income, education (< 12
years, high school graduate or general
equivalency diploma, or any college), marital
status (currently married or living as married,
widowed, divorced, separated, or never
married), physical illness (cardiovascular
illnesses, gastrointestinal illnesses, and arthritis),
residence (metropolitan central city,
other metropolitan, or nonmetropolitan), region
(Northeast, Southwest, West, or South).
Cardiovascular illnesses comprised hardening of
arteries or arteriosclerosis, high blood pressure
or hypertension, chest pain or angina
pectoris, heart attack or myocardial infarction,
and other forms of heart disease. Gastrointestinal
illnesses comprised gastritis, stomach ulcer,
cirrhosis of the liver, and other liver
diseases. The arthritis category had arthritis
only. Participants were asked whether they had
each condition in the past 12 months and
whether the diagnosis was confirmed by
a doctor or other health professional.
In the second stage of the analyses, we
conducted binary logistic regression analyses in
which we assessed the association of smoking
with the onset of any mood or anxiety disorders
and with new onset of individual disorders.
We categorized smoking status into 5
levels: 4 smoker groups and 1 nonsmoker
group. We further divided the nonsmoker
group into ex-smokers and lifetime nonsmokers.
The analyses also adjusted for the
same variables as in the first stage. We
excluded participants with a lifetime history
of each target disorder from the analyses for
that disorder. For example, we excluded
participants with lifetime history of panic
disorder at baseline from the analyses of
new-onset panic disorder. The analyses adjusted
for other lifetime disorders.
In the third stage of the analyses, we assessed
whether the association of smoking with newonset
mental disorders was consistent across
groups defined by gender, age, race/ethnicity,
household income, education, marital status, and
past history of mental disorders (as a dichotomous
variable, any vs none at baseline). To this end, we
entered the interaction terms of smoking status
with these characteristics 1 at a time into a logistic
regression model and tested their association with
the new onset of any mood or anxiety disorders.
If an interaction term was statistically
significant, we conducted further stratified
regression analyses for any mood or anxiety
disorders at aggregate level and for individual
disorders. In the interaction analyses, we
adjusted for the same sociodemographic,
physical, and mental health variables as in the
main effect analyses.