Methods
Data
Data are obtained from European Longitudinal Cohort Study of Pregnancy
and Childhood–Czech Republic (ELSPAC-CZ). The study was initiated by
World Health Organization to investigate factors in maternal and child
health in several European countries. The Czech part was approved for
adherence to ethical guidelines by the Scientific Council of Masaryk
University. The Czech cohort targeted the entire population of births in
two cities between March 1, 1991 and June 30, 199227.
Brno, a large metropolis and Znojmo, a smaller nearby town, were chosen
to represent urban and rural populations. Medical documentation was
available for 7,589 births (96% of all eligible births). To assess
prenatal characteristics, the study partnered with local
obstetrics/gynecology practices to distribute survey questionnaires to
gravidas at 20 weeks (henceforth mid-pregnancy). This effort yielded
4,811 responses for the baseline survey sample. All participating women
provided written informed consent. When the focal child reached age six
months, eighteen months, three years, five years, and seven years, women
were mailed follow-up surveys on child’s health. At child age seven
years, 3,312 mother-child pairs remained in the cohort for 67% of the
baseline sample. Of these, 1,849 mother-child pairs had complete data
and are included in the analytical sample.
Measures
The main outcome, child’s wheeze/wheeze with whistle (henceforth
“wheeze”), was collected at age six months, eighteen months, three
years, five years, and seven years. At each time point, mothers were
asked, “Were there any periods [between observation time points]
when there was wheezing or wheezing with whistling on your child’s chest
when breathing?” (yes/no). Following Tucson Children’s Respiratory
Study (TCRS)28,29, wheeze was coded as four
longitudinal phenotypes, including “never” (no wheezing at any time
point), “early-onset transient” (EOT) for onset before age three years
with resolution by age seven years, “early-onset persistent” (EOP) for
onset before age three years persisting at age seven years, and
“late-onset” (LO) for onset between age three years and seven years.
While TCRS used age six years as a cut-off, we use seven years as the
closest available observation point.
Prenatal and postnatal life events, main predictors, were
assessed using Social Readjustment Scale30 adapted for
Avon Longitudinal Study of Pregnancy and Childhood. The questionnaire,
tailored specifically for pregnancy and postpartum and used in prior
population-based research31–33 lists 42 types of life
events (Figure A1 and A2 in SI). Events range from rare, e.g., death of
a spouse, to more common, e.g., problems at work. Respondents report
whether each event occurred during a specified period. Prenatal stress
was assessed in mid-pregnancy using events that occurred since
conception. Postnatal stress was assessed at six months postpartum using
events that occurred since the delivery. For each observation period,
stress exposure was represented by the total number of reported events
categorized as low, medium, and high using terciles.
Mediators. To represent no/short breastfeeding as a risk factor,
we used maternal reports of breastfeeding duration collected at eighteen
months; for non-responding mothers, breastfeeding duration was extracted
from paediatric records. The duration was dichotomized using the cut-off
of four months, the approximate mean. Mothers reporting no breastfeeding
or breastfeeding <4 months were coded as 1 for no/short
breastfeeding; others were coded as 0.
LRTI and allergy diagnoses for age 0-18 months were extracted from
paediatric documentation. Both variables were coded 1 for children with
the diagnosis and 0 for others.
Current maternal smoking (yes/no) was self-reported in mid-pregnancy.
Child’s SSE was mother-reported at eighteen months, coded as hours per
week, and treated as continuous.
Covariates . Demographic background includes maternal education
(years), marital status (married vs. non-married), age at delivery
(years), child sex (male vs. female), and residential region (Brno vs.
Znojmo). Additional covariates, selected using literature on paediatric
wheeze/asthma, include child allergy extracted from medical records at
age eighteen months, maternal histories of asthma and
allergy34 self-reported in mid-pregnancy, and low
birth weight (<2,500 grams)35 extracted from
obstetric documentation. Parity and singleton birth are used as proxies
for siblings in the household11.
Statistical Analysis
Analysis was conducted using Stata 14.0 statistical software.
Descriptive statistics were obtained including frequencies, percentages,
means, and standard deviations. Bivariate relationships between life
events and wheeze phenotypes were assessed using simple multinomial
logistic regression models. Next, full multivariate models were fitted
including all hypothesized predictors, mediators (prenatal life events,
postnatal life events, LRTI, allergy, smoking in pregnancy, SSE,
breastfeeding), demographic characteristics (child sex, maternal age and
its square, education, region, singleton birth), and all remaining
covariates. Final models testing H1 and H2 retained hypothesized
predictors, mediators, and demographic characteristics; other covariates
were trimmed for parsimony to those with p<0.10 in the full
model. These included parity and maternal asthma history. Predicted
probabilities for each wheeze phenotype were obtained using marginal
effects. Finally, mediation analysis was conducted to test H3 and H4.
Wheeze was dichotomized as no wheeze vs. any wheeze in mediation
analysis since methodology for mediation with dichotomous outcomes is
better developed compared to multi-category
outcomes36. Decomposition of total effects into direct
and indirect was performed using “ldecomp” command37with 1,000 bootstrap replications to produce 95% bias-corrected
confidence intervals.