METHODS
Study population and data
collection
This study is based on the Norwegian Mother, Father and Child Cohort
Study (MoBa),10,11 linked to records in the Medical
Birth Registry of Norway (MBRN), 12 the Norwegian
Prescription Database (NorPD),13 and the Norwegian
Patient Registry (NPR)14 via the maternal personal
identification number and pregnancy sequence. MoBa is a nation-wide,
prospective population-based pregnancy study conducted by the Norwegian
Institute of Public Health.10,11 Participants were
recruited in 1999-2008 through a postal invitation in connection with a
publicly offered routine ultrasound at 17-18 weeks of gestation.
Prenatal data were gathered via two self-administered questionnaires at
week 17 (Q1) and 30 (Q3). Postnatal follow-up questionnaires on maternal
and child health were sent to mothers from child age 6 months to
adolescence. Follow-up of children started in 1999 and is still ongoing.
Prospective fathers also completed one prenatal questionnaire at week
17. The current study is based on version 9 of the quality-assured data
files released for research. The cohort now includes 114500 children,
95200 mothers and 77300 fathers.10 The participation
rate for all invited pregnancies was 41%.11 This
study followed the STROBE reporting guideline for cohort studies.
The MBRN is a nationwide registry based on compulsory notification of
all live births, stillbirths and induced abortions.12The NorPD collects data on all prescribed medications dispensed from
community pharmacies irrespective of reimbursement since 2004. The NPR
contains records on admission to hospitals and specialist healthcare
since 2008. The data include date of admission and discharge, primary
and secondary diagnosis, and cover all government-owned hospitals and
outpatient clinics, and all private health clinics that receive
governmental reimbursement. Diagnostic codes in the NPR follow the
International Classification of Diseases, version 10 (ICD-10). Figure S1
outlines the exclusion criteria to achieve the i) final ADHD diagnosis
sample, with complete registry-based outcome data for all MoBa children,
and the ii) final ADHD symptom sample, including MoBa children with
maternal-reported data at age 5 years.
Self-reported clinical depression and
anxiety
We included pregnancies within women reporting depression and/or anxiety
during gestation.15,16 In MoBa Q1 and Q3 women were
presented with a list of concurrent illnesses, and could report whether
they were having “depression” or “anxiety” or “other mental
disorders” (hereafter, clinical depression/anxiety) in pregnancy, and
likewise in the time prior to pregnancy. To further tease apart the role
of underlying maternal psychiatric disease from that of drug treatment
in pregnancy, we additionally included women (discontinuers) with no
self-reported clinical depression/anxiety in pregnancy, but who reported
using antidepressant solely in the six month period prior to pregnancy.
The study measured severity of maternal symptoms of depression and
anxiety at week 17 and 30 via the short versions of The Hopkins Symptom
Checklist-25, i.e. the 5-item (SCL-5) scale.17,18 More
information is outlined in the Supplement.
SSRI and SNRI exposure
In MoBa Q1 and Q3 women reported the name of the medication taken and
timing of use in four-week intervals according to indication (Q1 for
week 0-13+ and also 6 months before pregnancy; Q3 for week
13-29+).19 Drug classification was based on the
Anatomical Therapeutic Chemical (ATC) Classification
System.20
In a sub-sample of women enrolled in MoBa since 2004, NorPD was used as
complementary source of exposure data. Available data in NorPD include
ATC codes of individual antidepressants dispensed, dispensing dates, and
the amount dispensed. We measured any antidepressant prescriptions
filled within the period from pregnancy start to delivery, in accordance
to prior research.21 More detail about exposure
definition in NorPD is given in the Supplement.
Gestational exposure to each individual antidepressant was defined as
exposure to a drug belonging to the ATC group N06A. SSRIs (sertraline,
fluoxetine, paroxetine, citalopram, escitalopram, fluvoxamine) and SNRIs
(venlafaxine, duloxetine) were grouped together (SSRI/SNRI) because of
their pharmacological properties.
Because symptoms of depression/anxiety were measured at week 17 and 30,
and reflected disease severity in the prior two weeks, we defined the
following points of exposure in the timing analysis, as described in
prior work:8,22 early (weeks 1-16), mid (weeks 17-28)
and late (> week 29) pregnancy. Length of SSRI/SNRI use was
defined according to how many 4-week intervals, out of the eight
possible throughout pregnancy, were checked. These were grouped into
“1-8 weeks”, “9-20 weeks” or “> 20 weeks”. In
addition, we defined an ever exposure group during gestation. Women were
classified as exposed if they reported use of SSRI/SNRI during these
periods. Two mutually exclusive comparison groups were defined: i)
Non-medicated: women with self-reported clinical depression/anxiety in
pregnancy but non-medicated; ii) Discontinuers: women who reported use
of any antidepressant only in the six months period prior to pregnancy,
who did not report depression/anxiety in pregnancy.
ADHD diagnosis
A diagnosis of ADHD in the offspring (hereafter, ADHD) was defined as i)
at least one primary or secondary diagnosis in the NPR based on the
ICD-10 codes F90 (hyperkinetic disorder), in the period 2008-2015; or
ii) one or more dispensed ADHD medication licensed in Norway (i.e.,
methylphenidate, atomoxetine, racemic amphetamine, dexamphetamine, and
lisdexamphetamine) in NorPD between 2004-2016.23 The
ICD-10 codes diagnosis of hyperkinetic disorder requires the combination
of both inattentive and hyperactive symptoms.24 The
majority of MoBa children were born in 2004 or later and thus outcome
data since birth were available for most of the children in this study
(Figure S2).
ADHD symptoms
Child ADHD symptoms by age 5 years were mother-reported via completion
of the widely-used, validated Conners Parent Rating Scale-Revised
(CPRS-R).25 MoBa included 12 selected CPRS-R items
measuring the ‘inattention’ and ‘hyperactivity/impulsivity’ domains.
Mothers were asked to rate whether each item reflected their child’s
behavior in the last six months. The CPRS-R items and related scoring
have been previously published.22 Mean CPRS-R score
was calculated and standardized, and higher z-scores indicated greater
ADHD symptoms. In the current study the internal CPRS-R consistency was
0.90.
Measured confounders and other postnatal
factors
We identified a sufficient set of confounders with the aid of directed
acyclic graphs.26 These were pre-pregnancy maternal
Body Mass Index (BMI), parity, education and gross yearly income,
marital status, folic acid, smoking and alcohol use in early pregnancy,
paternal age, and an obstetric comorbidity index,27 as
described in detail in the Supplement; co-medication in early pregnancy
with opioid analgesics, paracetamol, nonsteroidal anti-inflammatory
drugs (NSAIDs), benzodiazepine/z-hypnotics, and antipsychotics; severity
of maternal depressive and anxiety symptoms in pregnancy via the SCL-5,
and Life Time History of Major Depression (LTH of MD), as measured in Q1
via five key depressive symptoms closely corresponding to the DSM-III-R
criteria for lifetime major depression.28 We included
maternal and paternal filled prescriptions for ADHD medication as proxy
of familial risk of ADHD. In separate models, we included other maternal
psychiatric, paternal, child and postnatal factors (see Supplement and
Table S1). More details on covariates are given in the Supplement.
Data analysis
To estimate associations with ever SSRI/SNRI exposure as ever in
gestation and by duration, we fit crude and weighted analyses using
inverse probability of treatment weighting (IPTW), based on the
propensity score.29 Logistic regression models were
first fit to estimate the probability of ‘SSRI/SNRI exposure’ as ever
and in the duration windows (1-8, 9-20, more than 20 weeks), relative to
non-medicated or discontinuers, given the set of sufficient confounders.
To estimate associations by timing of exposure, we fit marginal
structural models (MSM)30 with two time points to
account for i) time-varying SSRI/SNRI exposure; ii) time-varying
confounders (i.e, SCL-5 in pregnancy and co-medications) which are
affected by prior SSRI/SNRI treatment, as illustrated in prior
work.8,22 We estimated the probability of SSRI/SNRI
treatment using a pooled logistic regression in which the outcome was
current treatment with an SSRI/SNRI in mid or late pregnancy, and
covariates were maternal baseline, time-varying and time-fixed
confounders, and SSRI/SNRI in early pregnancy. We then derived
stabilized IPTW for each pregnancy at each time point.
To estimate standardized mean differences in symptoms and hazard ratio
(HR) for ADHD, we respectively fit crude and weighted generalized linear
and Cox regression models with robust standard errors. In the Cox
regressions, we used child age as time scale and a quadratic term for
year of birth to address left-truncation for children born before 2004;
the follow-up period for all live-born children started at birth and
ended on the date of ADHD diagnosis, date of first drug prescription for
ADHD, or 31 December 2016, whichever came first. The current study did
not have information about dates of potential emigration or death, but
only whether these events had occurred (91 children (1.4%) had
emigrated). Because the proportionality hazard assumption was not met,
we split the follow-up time at child age 7 and 9 years (Figures S3-S4),
estimating period-specific HRs. All statistical analyses were performed
by using Stata MP 16. Data are presented crude and weighted hazard
ratios (wHR), and as standardized means scores with 95% CI. Power
analysis is outlined in Table S2.