Article
1136 Am J Psychiatry 165:9, September 2008ajp.psychiatryonline.org
This article is featured in this month’s AJP Audio and is discussed in editorials by Dr. Reiss (p. 1083) and Dr. Markowitz (p. 1086).
Children of Depressed Mothers 1 Year After the Initiation of
Maternal Treatment: Findings From the STAR*D-Child Study
Daniel J. Pilowsky, M.D., M.P.H.
Priya Wickramaratne, Ph.D.
Ardesheer Talati, Ph.D.
Min Tang, M.A.
Carroll W. Hughes, Ph.D.
Judy Garber, Ph.D.
Erin Malloy, M.D.
Cheryl King, Ph.D.
Gabrielle Cerda, M.D.
A. Bela Sood, M.D.
Jonathan E. Alpert, M.D., Ph.D.
Madhukar H. Trivedi, M.D.
Maurizio Fava, M.D.
A. John Rush, M.D.
Stephen Wisniewski, Ph.D.
Myrna M. Weissman, Ph.D.
Objective: Maternal depression is a con-
sistent and well-replicated risk factor for
child psychopathology. The authors ex-
amined the changes in psychiatric symp-
toms and global functioning in children
of depressed women 1 year following the
initiation of treatment for maternal major
depressive disorder.
Method: Par t ic ipants were 1 ) 151
women with maternal major depression
who were enrolled in the Sequenced
Treatment Alternatives to Relieve Depres-
sion (STAR*D) study and 2) their eligible
offspring who, along with the mother,
participated in the child STAR*D (STAR*D-
Child) study (mother-child pairs: N=151).
The STAR*D study was a multisite study
designed to determine the comparative
effectiveness and acceptability of various
treatment options for adult outpatients
with nonpsychotic major depressive dis-
order. The STAR*D-Child study examined
children of depressed women at baseline
and involved periodic follow-ups for 1
year after the initiation of treatment for
maternal major depressive disorder to as-
certain the following data: 1) whether
changes in children’s psychiatric symp-
toms were associated with changes in the
severity of maternal depression and 2)
whether outcomes differed among the
offspring of women who did and did not
remit (mother-child pairs with follow-up
data: N=123). Children’s psychiatric symp-
toms in the STAR*D-Child study were as-
sessed using the Schedule for Affective
Disorders and Schizophrenia for School-
Age Children—Present and Lifetime Ver-
sion (K-SADS-PL), and maternal depres-
sion severity in the STAR*D study was as-
sessed by an independent clinician, using
the 17-item Hamilton Depression Rating
Scale (HAM-D).
Results: During the year following the
initiation of treatment, maternal depres-
sion severity and children’s psychiatric
symptoms continued to decrease over
time. Decreases in the number of chil-
dren’s psychiatric symptoms were signifi-
cantly associated with decreases in mater-
nal depression severity. When children’s
outcomes were examined separately, a
statistically significant decrease in symp-
toms was evident in the offspring of
women who remitted early (i.e., within
the first 3 months after the initiation of
treatment for maternal depression) or
late (i.e., over the 1-year follow-up inter-
val) but not in the offspring of nonremit-
ting women.
Conclusions: Continued efforts to treat
maternal depression until remission is
achieved are associated with decreased
psychiatric symptoms and improved func-
tioning in the offspring.
(Am J Psychiatry 2008; 165:1136–1147)
Maternal depression is a risk factor for child psy-
chopathology that is consistent and well-replicated. The
disorders among children of women with maternal de-
pression vary according to the developmental stage of the
affected offspring, with the average age at onset of anxiety
and behavior disorders before puberty, the average age at
onset of major depression in early adolescence, and the
average age at onset of substance use disorders in late ad-
olescence or early adulthood (1–4). Furthermore, mater-
nal depression has been associated with less robust re-
sponse to treatment among depressed adolescent
offspring (5). Until recently, little has been known about
the effect of treatment for depressed mothers on their off-
spring (6–8). The Sequenced Treatment Alternatives to Re-
lieve Depression (STAR*D) study provided a unique op-
portunity to investigate whether successful treatment for
depressed women would lead to improvement in psychi-
atric disorders and symptoms among their offspring (9).
The study of outcomes in these offspring is referred to as
the child STAR*D (STAR*D-Child) study (10, 11). The main
YYL
加亮
YYL
加亮
YYL
加亮
YYL
加亮
Am J Psychiatry 165:9, September 2008 1137
PILOWSKY, WICKRAMARATNE, TALATI, ET AL.
ajp.psychiatryonline.org
objective of the STAR*D study (12–14) was to determine
which treatment or sequence of treatments would be ef-
fective in adult patients with a diagnosis of major depres-
sion who did not remit after first-line treatment with the
antidepressant citalopram, a selective serotonin reuptake
inhibitor. Remission was defined as a score of ≤7 on the
17-item Hamilton Depression Rating Scale (HAM-D [15]).
At the initiation of maternal treatment, approximately
one-third of the offspring (aged 7–17 years) of women in
the STAR*D study had a current psychiatric disorder (10).
Eligible offspring were independently assessed in the
STAR*D-Child study by a team that was not involved with
the mothers’ treatment. Three months after the initiation
of treatment for maternal depression, we reported that re-
mission among the mothers was associated with a signifi-
cant decrease in the rates of children’s diagnoses as well as
decreases in internalizing, externalizing, and total symp-
toms (11). Interestingly, a ≥50% maternal response to treat-
ment was required to detect improvement in the offspring.
In the STAR*D-Child study, mothers and their offspring
were followed every 3 months for 1 year after the initiation
of treatment for maternal depression. The objective of the
present study was to assess both maternal and child out-
comes as reported in the STAR*D-Child study. We investi-
gated two important issues that concern adult and child
clinicians. First, we assessed whether those children of
mothers who remitted by the 3-month follow-up re-
mained well throughout the remainder of the 1-year fol-
low-up interval. Second, we assessed whether those chil-
dren of mothers who remitted after the 3-month follow-up
similarly benefitted from maternal remission. In the
present study, we refer to depressed mothers who remit-
ted by the 3-month follow-up as early remitting subjects
and to those who remitted after the 3-month follow-up as
late remitting subjects. Most of the mothers (92%) who re-
mitted by the 3-month follow-up were treated with citalo-
pram (11). Late remitting subjects received two or more
treatments as provided by the STAR*D protocol (12).
We hypothesized that the following outcomes would oc-
cur during the 1 year after the initiation of treatment for
maternal depression: 1) mothers would show a significant
decrease in HAM-D scores, with children showing a signif-
icant decrease in offspring psychiatric symptoms, and
changes in child outcomes would be associated with
changes in maternal HAM-D scores; 2) children of early
remitting subjects, who were shown to benefit from remis-
sion of maternal depression 3 months after the initiation
of treatment (11), would continue to demonstrate a lower
prevalence of psychiatric symptoms relative to children of
nonremitting subjects, providing that their mothers re-
mained in remission; and 3) children of late remitting sub-
jects would demonstrate intermediate outcomes.
Method
The STAR*D Study Design
STAR*D (http://www.star-d.org) was a multisite study de-
signed to determine the comparative effectiveness and accept-
ability of different treatment options for a broadly representative
group of outpatients with nonpsychotic major depressive disor-
der (12, 14). STAR*D study participants were adults (age range:
18–75 years) with nonpsychotic major depressive disorder who
did not have a lifetime diagnosis of bipolar, schizophrenia, or
schizoaffective disorders. Initially, all study participants were
treated with citalopram. Those subjects who did not remit with
citalopram treatment or who were intolerant of citalopram were
offered other treatments, including antidepressants, cognitive
behavioral therapy, or a combination of treatments, using an
equipoise-randomized design as described elsewhere (16, 17).
The STAR*D-Child Study Design
Among the 14 STAR*D regional centers, seven participated in
the STAR*D-Child study. The STAR*D study recruited 824 women
(age range: 25–60 years) at these seven regional centers. Among
the women recruited, 808 (98%) were screened to ascertain
whether they had at least one child between the ages of 7 and 17
years. Of these women, 177 (22%) had children in this age range.
Among the 177 women with at least one child aged 7–17 years,
174 (98%) met all eligibility criteria, and 151 (87%) entered the
child study.
The STAR*D-Child assessors were 1) not involved with the
mothers’ treatment, 2) blind to their remission status, and 3) in-
dependent of the team that treated these mothers. Since the as-
sessors were involved with multiple assessments of children over
time, they were not blind to child status from prior assessments.
The mothers’ clinical assessments were conducted by STAR*D
staff who were not involved with the children’s assessments. Base-
line assessments (Figure 1) were conducted before or within 2
weeks following the initiation of treatment for maternal depres-
sion and every 3 months thereafter (3, 6, 9, and 12 months).
Subjects
Mothers (age range: 25–60 years) were adult outpatients with
nonpsychotic major depressive disorder who were enrolled as
participants in the STAR*D study. Mothers of children aged 7–17
FIGURE 1. STAR*D-Child Study Designa
a Data are based on the year following the initiation of treatment for
maternal depression.
Baseline
Assessment
Treatment of maternal
depression
Remission
Non-
Remission
Assessed every
3 months
Follow-up ends a year after
remission
Assessed every
3 months
Follow-up ends two years after
baseline assessment
Baseline
assessment
Treatment of maternal
depression
Remission Non-remission
Assessed every
3 months
Follow-up ends 1 year after
remission
Assessed every
3 months
Follow-up ends 2 years after
baseline assessment
Depressed
Mothers
Depressed
mothers
1138 Am J Psychiatry 165:9, September 2008
CHILDREN OF DEPRESSED MOTHERS
ajp.psychiatryonline.org
years were invited to participate in the STAR*D-Child study with
one eligible child. These mothers provided separate written in-
formed consent. If a mother had more than one offspring in the
age range of eligibility, one child was randomly selected. Addi-
tional data pertaining to the STAR*D-Child study sample are
available elsewhere (10).
Maternal Assessments
Maternal assessments were conducted as part of a comprehen-
sive battery of tests for all STAR*D participants (17). A diagnosis of
current nonpsychotic major depressive disorder was established
by a clinical interview and confirmed using a symptom checklist
based on DSM-IV criteria (17). The HAM-D was used to assess the
severity of depressive symptoms (15). The HAM-D has good reli-
ability, and scores are highly correlated with the results from
other observer-rated instruments (18). In the STAR*D study, re-
mission of maternal depression was defined as a HAM-D score
≤7, and relapse was defined as a HAM-D score ≥14 any time after
remission. Mothers who were in remission during and after the
first 3 months of follow-up were considered early and late remit-
ting subjects, respectively, provided that they remained in remis-
sion for the entire 1-year follow-up interval.
Remission and treatment response were assessed using HAM-
D scores that were obtained from the STAR*D study. In cases in
which mothers discontinued their participation in the adult study
but remained in the child study, HAM-D scores were not avail-
able. In these cases and whenever HAM-D scores were unavail-
able, we converted the mothers' Quick Inventory of Depressive
Symptomatology—Self-Report scores to HAM-D scores. Quick
Inventory of Depressive Symptomatology—Self-Report scores
were obtained via interactive voice response (13) through the use
of a method based on item response theory analysis of the rela-
tionship between this self-report and HAM-D, as employed by the
STAR*D study (9, 19, 20). The following HAM-D scores were un-
available for assessment: 0 at baseline and 51/121, 9/87, 4/61, and
22/79 at 3, 6, 9, and 12 months, respectively.
Child Assessments
Children were assessed with the Schedule for Affective Disor-
ders and Schizophrenia for School-Age Children—Present and
Lifetime Version (K-SADS-PL). The K-SADS-PL is a semistructured
diagnostic interview used to assess children at baseline and at 3-
month intervals (21). Among the K-SADS diagnostic measures, K-
SADS-PL, a downward extension of the adult SADS, is the most
widely used version currently available (22, 23). It is a valid and re-
liable diagnostic instrument, with test-retest reliability kappa co-
efficients in the good to excellent range (0.63–1.00) for present and
lifetime diagnoses (21). K-SADS-PL includes a screening section
and supplements for children who screen positive for one or more
disorders. In addition, all sections are available online
(www.wpic.pitt.edu/ksads/default.htm). Furthermore, the
mother (or parent) and child are interviewed separately, and
symptoms are scored separately (as absent or present) at the
threshold or subthreshold level. Because STAR*D was an effective-
ness study and mimicked clinical practice, we were required to
minimize the number of assessments. Thus, we selected the fol-
lowing sections of the K-SADS-PL that focus on disorders that are
highly prevalent among children of depressed parents: affective,
anxiety, and disruptive behavior disorders (2, 3). The qualifica-
tions, training of interviewers, and reliability of diagnostic assess-
ments of children have been reported elsewhere (10).
The variable child symptoms consisted of a simple count of the
child-reported symptoms from the screening interview that were
present at the threshold (definite symptoms) or subthreshold
level. We created a similar variable based on the mother’s clinical
interview about the child (mother-reported child symptoms) and
TABLE 1. Baseline Demographic and Clinical Characteristics of Depressed Mothers and Their Offspring (N=123)
Characteristic and Assessment Measure
Mother-Child Pair Assessment Before (or within) 2 Weeks After Initiation of Treatment for
Maternal Depression
Maternal Mean SD Range Median
Age (years) 37.7 6.6 24.0–52.0 36.0
HAM-D score 24.6 5.2 14.0–42.0 25.0
N % Range Median
Race
African American 45 36.6
White 52 42.3
Hispanic 22 17.9
Other 4 3.3
Education
90
indicate superior functioning, and scores <70 indicate impaired
global functioning.
Data Analysis
Changes in maternal and child outcomes. Changes in ma-
ternal HAM-D scores and in child C-GAS scores, which are con-
tinuous variables, were modeled using longitudinal mixed-effect
regression models with random coefficients. The changes in
HAM-D scores of each individual mother over time were first
modeled using a polynomial regression model, with the individ-
ual HAM-D score at each specific time as the dependent variable
and the number of months after baseline assessment as the inde-
pendent variable. In addition, household income was included as
a potential confounding variable. The polynomial regression
model posits that the individual treatment response of each
mother was a function of the number of months since the base-
line assessment. This model was used to estimate 1) the average
pattern of change (e.g., linear, quadratic) in HAM-D scores over
time and 2) the extent of variation around regression parameters.
Initially, each regression parameter was treated as a random coef-
ficient to determine whether these parameters varied between
mothers. If there was no significant variation, the coefficients
were treated as fixed. A similar approach was used to model child
C-GAS scores, with children’s age, gender, and family household
income included as potential confounding variables. All longitu-
dinal mixed-models were performed using the PROC MIXED pro-
cedure of the SAS system (SAS, Cary, N.C.).
Changes in child symptoms were modeled using Poisson re-
gression analysis, which is appropriate for modeling count data.
The logarithm of a child’s symptom count at each specific time
was considered the dependent variable, and the number of weeks
after the baseline assessment was treated as an independent con-
tinuous variable. Changes in child diagnoses were similarly mod-
eled using logistic regression analyses, which are appropriate for
modeling dichotomous outcomes. Both Poisson and logistic re-
gression analyses were conducted within the framework of the
generalized estimating equation approach to adjust for correla-
tions between repeated observations over time using an ex-
changeable correlation matrix, as suggested by Asarnow et al.
(25), and were performed using the PROC GENMOD procedure of
the SAS system (SAS, Cary, N.C.). For all models, the potential
confounding variables (children’s age, gender, and household in-
come) were included in the analysis.
For exploratory analysis of changes in symptoms over time in
children with at least one definite symptom at baseline, we used
1) a log (x+1) transformation to normalize the data and 2) relevant
random regression analyses as described previously.
Associations between maternal HAM-D scores and child
outcomes. The series of analyses conducted allowed us to de-
termine whether mother and child outcomes improved over the
mother’s treatment/follow-up period. However, if there was a sig-
nificant change over time in the mother’s HAM-D score and in the
child’s outcome variables, we sought to determine whether
changes in child outcomes over time were related to changes in
maternal HAM-D scores over time. This was initially accom-
plished by fitting 1) a Poisson regression model to child symp-
toms; 2) a logistic regression model to child diagnoses; and 3) a
random regression model to child functioning, as described pre-
viously except for the inclusion of the mother’s HAM-D score at
current assessment as a time-dependent covariate. We were able
to infer that changes in child outcomes were at least partially re-
lated to changes in the mother’s current HAM-D score if 1) the es-
timate of the coefficient corresponding to time (i.e., the nu