Statistical analysis
Multinomial logistic regression models were used to estimate
the association between both paternal and maternal
prenatal depression and very and moderately preterm birth.
Model 1 was adjusted for calendar year of birth and maternal
age as a continuous variable, with quadratic terms of
maternal age, parity, paternal education, and paternal age
(45 years). In model 2 we further adjusted for
maternal smoking, BMI category, and chronological order
of parental depression.
Stratified analyses were performed to compare the
depression exposures measured (from a diagnosis of depression
or from a filled prescription of an antidepressant drug
only), and to compare the effects of paternal depression on preterm birth in families where mothers cohabitated with
the father against those who did not. Sensitivity analyses
were performed, as follows: (1) on births of primiparous
women, to exclude residual confounding from previous
adverse birth outcomes; (2) on mothers who had never had
a previous miscarriage or an ectopic pregnancy; (3) on the
effects of paternal depression on mothers who had never
experienced depression (to exclude residual confounding
from maternal depression); and (4) on paternal depression
only observed until 12 weeks of gestation.
We used Huber–White sandwich estimates of variance to
account for correlations between maternal siblings. The
same method was used in the sensitivity analysis of primiparous
women to account for correlations between paternal
siblings.
All statistical analyses were performed using STATA/
MP 12.0 for WINDOWS (Stata Corporation, College Station,
TX, USA).