Prediction of Adolescent Mfective Disorder: Effects of Prior

Parental Mfective Disorders and Child Psychopathology

WILLIAM R. BEARDSLEE, M.D., MARTIN B. KELLER, M.D., RONALD SEIFER, PH.D.,
PHILIP W. LAVORI, PH.D., JANET STALEY, M.A., DONNA PODOREFSKY, PH.D.,
AND DAVID SHERA, M.S.

ABSTRACT
Objective: To examine the role of major parental and child diagnostic factors in predicting episodes of serious affective
disorder in adolescents in a nonreferred sample. Method: The sample included 139 youngsters (average age 14 years
at enrollment) drawn from a health maintenance organization and evaluated at two points in time 4 years apart. Both
parents and adolescents were assessed using structured diagnostic instruments scored according to criterion systems.
Parent and child lifetime diagnoses identified in the first assessment were used to predict the onset of episodes of
serious affective disorder in the adolescents which occurred between the first and second assessment. Results:
Stepwise multiple regression analyses of the significant univariate factors showed that the most powerful predictors of
episodes of affective disorder were total number of diagnoses the adolescents received prior to first assessment, lifetime
duration of parental major depressive disorder, and total number of lifetime nonaffective disorders of the parents.
Conclusion: Broad risk factors from different domains best predict episodes of affective disorder in children and
adolescents. J. Am. Acad. Child Ado/esc. Psychiatry, 1996, 35(3):279-288. Key Words: parental affective disorders,
childhood psychopathology, predictor variables.

Numerous studies have consistently established that study the emergence of depression in adolescence be-
youngsters whose parents have experienced affective cause of the high rates of affective disorders in this
disorder are at substantially increased risk for a variety group. A combination of psychosocial (Rutter, 1990)
of impairments, including the diagnosis of depression and genetic influences (Gershon et al., 1985; Rutter,
during adolescence (Beardslee and Wheelock, 1994; 1990; Tsuang and Faraone, 1990) probably accounts
Downey and Coyne, 1990). The rate of diagnosis in for the higher rates of depression in offspring of parents
children of parents with affective disorders is several- with affective disorders.
fold higher than in comparison samples (Downey and Prospective longitudinal studies of episodes of disor-
Coyne, 1990). Investigation ofoffspring of parents with der in children of parents with affective disorder are
affective disorders provides an important opportunity to rare. Most studies have used retrospective and cross-
sectional designs (Beardslee and Wheelock, 1994;
Downey and Coyne, 1990). Prospective longitudinal
Accepted June 14, 1995.
s s
From theJudge Baker Children Centerand Children Hospital, Department research designs offer considerable advantages in that
ofPsychiatry, Boston; Brown University, Department ofPsychiatryand Human they identify risk factors and also may help describe
Behavior, Butler Hospital, Providence, RJ,. and the Veteram Administration
the relative contributions of different risk factors to
Cooperative Studies Program Coordination Center, Palo Alto, CA.
Supported by the William T. Grant Foundation, the NIMH through a outcomes (Rutter, 1989). The best available methods
grant titled "Children at Risk for Affective Disorder" (RO-1-MH34780-3) in for assessing diagnoses of current and past psychopa-
conjunction with the Boston Center ofCollaborativePsychobiology ofDepression
thology are standard interview protocols. Only two
Study (2-U02-MH25475-09), the Harris Trust through Harvard University,
the Overseas Shipholding Group, the George P. Harrington Trust, and a prospective studies of children of parents with serious
Faculty Scholar Award ofthe William T. Grant Foundation to Dr. Beardslee. affective disorder have used such protocols and focused
s
Reprint requests to Dr. Beardslee, Judge Baker Children Center, 295
on adolescent cohorts. Both studies used samples of
Longwood Avenue, Boston, MA 02115.
0890-8567/96/3503-0279$03.00/0©1996 by the American Academy clinically referred parents identified because of affective
of Child and Adolescent Psychiatry. disorder. Weissman et al. (1992) compared outcomes

J. AM. ACAD. CHILD ADOLESC. PSYCHIATRY, 35:3, MARCH 1996 279

BEARD SLEE ET AL.

in youngsters of parents with affective disorders to Wheelock, 1994; Downey and Coyne, 1990); (2) dura-
outcomes in youngsters in homes where parents had tion and severity of these disorders (Keller et al., 1986);
no disorder during a 2-year follow-up period . Only (3) nonaffective psychiatric disorder (Weissman et al.,
children of parents with affective disorder experienced 1984); and (4) the occurrence of divorce and separation
episodes of affective disorder. Hammen and associates (Beardslee et al., 1993a).
(Hammen, 1991; Hammen et al., 1991) found much Predictors identified within the adolescent were (1)
higher rates of disorder in children of mothers hospital- both presence of and duration of prior affective disor-
ized at least once with a unipolar depression than in ders (Harrington et al., 1990; Kovacs and Goldston,
a comparison group of children whose parents had no 1991; Lewinsohn et al., 1994); (2) learning disabilities
disorder. By life-table estimates, more than 80% of (Rutter, 1983); and (3) other psychiatric diagnoses
children of parents with unipolar disorder were ex- (Keller et al., 1991). The relationship ofIQ and gender
pected to experience an episode of diagnosable major to onset of episodes were also examined.
psychiatric disorder by age 19 years (Hammen, 1991). Our second aim was to determine, using regression
This article examines the contribution of parental af- analyses, the comb ination of variables that best ex-
fective disorder and other parental and child variables plained the onset of serious affective disorder in the
in predicting the onset of adolescent affective disorder, adolescents. Based on these analyses, our third aim
using a prospective longitudinal design with a group was to develop a usable risk-prediction model for use
of families identified through random sampling from by clinicians.
a prepaid health plan. We sampled families who had The sample used for these analyses has been pre-
at least one adolescent child, without previous knowl- viously described in cross-sectional and retrospective
reports. These (Beardslee et al., 1988, 1993a) demon-
edge of the presence or absence of parental diagnoses.
strated that lifetime history of parental affectivedisorder
The use of a nonreferred, random sample provided
was strongly and significantly associated with lifetime
a necessary complement to the usual approach of
history of diagnosable disorder in offspring. In particu-
identifying parents with affective disorder through pre-
lar, episodes of childhood depression in two separate
sentation for clinical treatment. Rates of depression in
investigations were significantly correlated with parental
epidemiological or community studies (Kessler er al.,
disorder-one when the youngsters were an average
1994; Robins and Regier, 1991; Weissman and Myers,
age of 14 years and another when they were an average
1978) have consistently demonstrated that lifetime
age of 18 years (Beardslee et al., 1988, 1993a). The
rates of depression in the general population are far latter lifetime assessment of the youngsters (Beardslee
higher than rates that would be identified by measuring et al., 1993a) was conducted blind to any information
individuals who were seeking treatment for depression. gathered at age 14 years. The use of these two separate
Moreover, use of a random sample from the roster of investigations allowed for construction of independent
a prepaid health plan ensured that the sample was predictor and outcome variables.
representative of the larger group of parents with adoles-
cent children in the prepaid health plan.
The central hypothesis of the current analysis was METHOD
that it would be possible to predict episodes of depres-
Subjects
sion and other serious affective disorders in youngsters.
Our first specific aim was to examine the predictive Initial Contact. Families were selected using a table of random
numbers and without knowledge of the presence or absence of
power of single variables identified at initial assessment parental affective disorder. Families were eligible if (1) they were
on the rate of episodes of serious affective disorder of Caucasian race; (2) chey were enrolled in the health plan; (3)
occurring in the interval between first and second they were English-speaking; (4) the household included both the
biological mother and her children; and (5) they had at least one
assessment. These predictor variables were identified adolescent child. All children aged 6 to 19 years in the family
from prior analyses of this sample or through a review were assessed. Eighty percent of those contacted who were eligible
of the literature. The main domains of predictor vari- were enrolled.
SecondAssessment. All families interviewed at initial assessment
ables in parents were (1) presence of affective disorder (T ,), were asked to part icipate in the follow-up study (T 2) 4 years
and major depressive disorder (MDD) (Beardslee and later. One hundred thirty-nine (91%) of 153 youngsters who were

280 J. AM . ACAD . CHILD ADOLESC. PSYCHIATRY. 35 :3 . MAR CH 1996

 PREDICTION OF EPISODES

originally assessed were interviewed in person at follow-up, as were Outcome Variable Assessment
the majority of their parents (Beardslee et al., 1993a).
For this sample, all those youngsters who were interviewed at 1;, Assessment ofAdolescents. The interview of the children at T 2
follow-up are included. The mean age of the parents was 40.7 at was done blind to information obtained at T 1 and covered present
initial assessment and 44.6 years at follow-up. There were 73 and past lifetime psychopathology and symptomatology using the
families assessed. The average interval between the assessments was Schedule for Affective Disorders and Schizophrenia-Epidemiologic
4.09 years. For the adolescents, the average age at initial interview Revised Version (K-SADS-ER) (Orvaschel et al., 1982). Diagnoses
was 13.9 years and 18.5 years at follow-up. were made according to RDC criteria, with a few additional
One hundred four of the 139 youngsters lived in families in DSM-III-R categories (Beardslee et al., 1993a). The interview was
which the biological parents were still married. Five had parents conducted by an assessor who had no prior contact with the
who had remarried. Marital separation had occurred in 7 families subject. The K-SADS-ER, rather than the SADS, was used even
and divorce in 21 families. In one family a parent was widowed. with youngsters older than 19 because their history of childhood
disorders was of interest as well as their current diagnostic status.
Both mother and youngster were interviewed separately about the
Parent Assessment youngster's diagnosis at T 2 when the youngster was younger than
Initial Assessment ofParents. Past and present psychiatric disorder 19. When there were conflicting reports, the interviewer made a
in both biological parents was assessed using the Schedule for best-estimate diagnosis based on clearly specified principles devel-
Affective Disorders and Schizophrenia-Lifetime Version (SADS-L) oped in the previous study and consistent with the use of the
(Andreasen et al., 1982; Endicott and Spitzer, 1978), and diagnoses K-SADS-ER by other investigators (Orvaschel et al., 1988). Mothers
were made according to Research Diagnostic Criteria (RDC) were not interviewed when subjects were 19 years or older.
(Spitzer et al., 1978). When fathers were unavailable, mothers Outcome Variables. The following diagnoses in the interval
provided diagnostic information about them. In a few cases, fathers between the first assessment (T 1 ) and the second assessment (T 2 )
became available for lifetime assessment subsequent to T 1 and these were defined as episodes of serious affective disorder in the adoles-
interview data were included for the period prior to T 1 assessment. cents: (1) an episode of MDD; (2) minor depression of at least 1
Detailed descriptions of instruments and assessment techniques year's duration; (3) schizoaffective disorder; or (4) two or more
have been presented elsewhere (Beardslee et al., 1988, 1993a). The episodes of hypomania.
Personal History of Depressive Disorders (Larkin and Hirshfeld,
1977) was used to obtain marital status and other demographic Statistical Analyses
information.
The first stage of development of the prediction model used
measures of association (Kendall's r) between each univariate pre-
Child Assessment dictor and the dichotomous outcome variable of serious affective
disorder in the interval (Conover, 1980).
T, Parental Report about the Child. The Diagnostic Interview
The second stage of analysis used logistic regression of variables
for Children and Adolescents for Parents (Herjanic and Reich,
that proved significant in the univariate comparison. Construction
1982; Reich et al., 1982) was administered to the mother to obtain
of the model involved standard stepwise procedures of eliminating
reports of her child's past and present psychopathology. Adaptive
the least significant variable. The model selected was that which
functioning was assessed using a revised version of the Rochester
best predicted the outcome with the fewest variables (Nelder and
Adaptive Behavior Inventory (jones, 1977; Keller et al., 1986).
Wedderburn, 1972).
All child assessments were conducted by experienced raters who
were blind to any knowledge of parental diagnostic category.
~ DirectAssessment of Child. Past and present psychopathology
and symptomatology were assessed in the children using the Diag- RESULTS
nostic Interview for Children and Adolescents (Herjanic and Reich,
1982; Reich et al., 1982), scored according to DSM-III criteria.
Sample Description
Selected portions of an interview designed to characterize current
adaptive functioning from Project Competence in Minnesota (Fin-
kleman, 1983) were also included. The IQ was obtained using the
High rates oflifetime parental affective disorder were
WISC-R (Wechsler, 1974) or WAIS (Wechsler, 1958), depending found at the initial assessment. The most common
on the age of the child. Diagnostic data from parent and child parental diagnostic category was MDD, with 40 cases
were combined to yield a single diagnosis using standard procedures followed by 20 cases of phobic disorder, 18 cases of
(Beardslee et al., 1988).
Overall adaptive functioning was based on a review of all intermittent depressive disorder, 19 cases of alcohol
information collected at the T 1 interview using a 9-point ordinal abuse and drug use, 13 cases of minor depression, 13
scale developed by Dr. Lyman Wynne (Beardslee et aI., 1988; cases of generalized anxiety disorder and 16 cases of
Jones, 1977). The adaptive functioning rating was made for the
period of 3 months prior to interview. Similar to a children's
other psychiatric disorder (RDC category which often
global assessment scale score (Shaffer et al., 1983), the adaptive connotes personality disorder), and a smaller number
functioning rating accounts for psychosocial function and diagnostic of other diagnoses. In this sample, 88 children had
information. Measurement of learning disabilities was rated on a
parents who had at some point within their lives
1~-p~i?-~ standard scale, with 0 representing no evidence of learning
disabilities, and 8 and above representing the presence of serious experienced affective disorder, 21 children lived in
disability. homes with nonaffective lifetime parental disorder,

]. AM. ACAD. CHILD ADOLESC. PSYCHIATRY, 35:3, MARCH 1996 281

BEARDSLEE ET AL.

and 30 children lived III homes with no history of Child Predictors

parental disorder. In these youngsters at initial assessment, 16 cases
oflifetime major depression, sometimes in combination
ParenVFamily Predictors
with dysthymia, and an additional 10 cases of lifetime
Composite family variableswere constructed for each dysthymia were identified in the sample at T l- Young-
child by summing the variable from both parents. sters frequently received multiple diagnoses (Beardslee
Duration of episodes of affectivedisorder in the parents et al., 1988).
was also measured by summing the total duration The duration of child affective disorder was exam-
during the lifetime for both parents for the diagnosis ined in addition to the presence or absence of diagnoses.
under consideration. There was wide variation within For the 16 youngsters who had major depression the
the sample (e.g., for MDD the duration ranged from range was 2 to 104 weeks (total number of weeks ill
1 to 246 weeks, with a mean of 29.24 weeks, and for during lifetime until time of first assessment) with a
overall affective disorder from 1 to 492 weeks) (Table mean of 24 weeks, whereas for those with major
1). The total number of nonaffective disorders was depression and/or dysthymia the range was 2 to 629
rated on a 0- to 4-point scale. weeks with a mean duration of 131 weeks (Table 1).

TABLE 1
Correlation of Predictor Variables with Affective Disorder Outcome
Correlation"
Parent Predictors with Adolescent
Affective
Parental Variables % No. Mean SD Disorder

Affective disorder
Presence 53.2 74 .215**
Duration (wk) 79.88 116.2 .196**
MDD
Presence 36.0 50 .206**
Duration 29.2 40.1 .217**
Nonaffective disorder 58.3 1.16 1.20 .259**
Total No. 81
Marital discord/divorce/separation 25.2 35 -.220**

Correlation"
Child Predictors with Adolescent
Affective
Child Variables % No. Mean SD Disorder

MD
Presence ll.5 16 .192*
Duration 23.6 29.4 .195*
MD/dysthymia
Presence 18.7 26 .271***
Duration 130.5 145.7 .259***
Nonaffective disorder: Total No. of
episodes in lifetime 0.86 1.12 .188*
Total No. of disorders 55.4 77 1.13 1.12 .229**
Adaptive functioning 3.04 1.62 .229**
Degree of learning disorder 2.37 2.72 .225**
Female 46.0 64 .082
FIQ 112.5 14.1 .040

Note: Child sample: N = 139. MDD = major depressive disorder; MD = major depression; FIQ = Full Scale IQ.
a Kendall's 't correlations.
* p :::; .05; ** P s .01; *** P :::; .001.

282 J. AM. ACAD. CHILD ADOLESC. PSYCHIATRY, 35:3, MARCH 1996

 PREDICTION OF EPISODES

The number of any diagnosed nonaffective disorder affective disorder. Relative risk of developing affective
(range 0 to 4) and the total number of diagnoses illness was nearly double when several of these main
received by the youngster (range 0 to 5) in his or her risk factors were present. Of the 50 youngsters whose
lifetime until first assessment were also examined. Data parents had experienced lifetime MDD, the rate of
on these diagnoses have been previously presented adolescent affective disorder was 34%, whereas for
(Beardslee et al., 1988). IQ ranged from 64 to 153, the remainder of the sample it was 18%. When any
with a mean of 112 and a standard deviation of 14. childhood disorder was diagnosed at T\, the rate of
serious affective disorder in interval increased from
Outcome Variables 16% to 30%.
Thirty-three of the 139 children reported serious Multivariate Comparisons
affective disorder in the interval between T\ and T 2•
Twenty-seven youngsters received a diagnosis ofMDD All variables that proved significant in the univariate
in the interval, several in combination with hypomania. comparison were entered into a stepwise logistic regres-
sion procedure (Conover, 1980; Nelder and Wedder-
Of the remaining six youngsters, three had longstanding
burn, 1972). Gender was additionally considered
minor depression, two had hypomania, and one had
because of the considerable literature indicating this
schizoaffective disorder. Single episodes of disorder
as a risk factor. The least significant predictor variable
were diagnosed in 20 of these youngsters while multiple
was removed from the model one step at a time until
episodes of disorder were reported by 7 youngsters.
all predictor variables were significant at p < .05.
The average duration of the MDD was 62.7 weeks.
The resulting model included three predictor variables:
Affective diagnoses often occurred in combination with
number of child diagnoses, duration of family MDD,
nonaffective diagnoses (Beardslee et al., 1993a).
and number of parental nonaffective disorders. This
Single Variable Comparisons model was highly significant (X 2 = 22.841, P < .0001).
Given that some variables were highly skewed, such
All variables significantly predicted subsequent epi- as parental MDD duration, several other analyses on
sodes of adolescent affective disorder (p < .05) except transformed alternatives, such as log or square root of
for gender and IQ (Table 1). Particularly strong correla- parental MDD; were run and gave similar results to
tions were found for parental MDD duration, number the model presented. Second-order interaction effects
of nonaffective diagnoses, total duration of affective of the predictor variables were not significant.
disorder, and marital discord. For the child predictors, From the clinical perspective, the most interesting
duration of child affective disorder and total number alternative model was derived by using a broader defini-
of diagnoses were strongly correlated with onset of tion of "statistical significance," p < .15 instead of p

TABLE 2
Estimated Probability of Interval Affective Disorder at Selected Points
Parent MDD Sample with Duration
No. of Parent No. of Child 75% 90% 95% 99%
Nonaffective Disorders Disorders (7) (47) (60) (I20)

o o 0.09 0.10 0.22 0.27 0.59

o 2 0.15 0.17 0.34 0.41 0.73
o 5 0.30 0.34 0.56 0.63 0.87
2 o 0.17 0.19 0.37 0.44 0.76
2 2 0.27 0.30 0.52 0.59 0.85
2 5 0.48 0.52 0.73 0.78 0.93
4 o 0.30 0.33 0.55 0.62 0.87
4 2 0.44 0.48 0.69 0.75 0.92
4 5 0.66 0.69 0.85 0.88 0.97

Note: MDD = major depressive disorder.

a Percentile.
b Months.

J. AM. ACAD. CHILD ADOLESC. PSYCHIATRY. 35:3. MARCH 1996 283

BEARDSLEE ET AL.

< .05, in the stepwise procedure. Using this broadened TABLE 4

definition, the best model included five predictor vari- Single Indicator Predictors and Combination
ables, the three from the first model with the addition Will Become
of marital status and gender. DisorderNalue No. III (%)

Parent Nonaffective diagnosis

Multiple Risk Factors in Interaction None 58 12
Any 81 32
Table 2 provides a description of outcome rates at Parent MDD
selected values of the three predictors based on the None 89 18
predicted probabilities. For example, with no parental Any 50 34
Child diagnosis
affective disorder, no history of child disorder, and no None 62 16
parental MOD, only 9% of the youngsters are predicted Any 77 30
to experience an episode of disorder in the interval Combination (sum of the three indicators)
between assessments. The presence of severalrisk factors None 27 7
1 38 18
dramatically increases the risk to the child: when a 2 52 25
total of four parental nonaffective disorders and five 3 22 50
previous child disorders are present, the chance of an
Note: MDD = major depressive disorder.
episode of affective disorder developing over the 4-year
period was predicted to be between 66% and 97%,
depending on the duration of MOD in the parent. were simply rated as present or absent. If none was
present, only 7% of the children became ' ill in the
Concentration of Risk within a Few Individuals
interval. If all three were present, 50% of the children
In this analysis the risk was not equally distributed became ill in the interval (Kendall's 't = P < .001).
across all the youngsters but was concentrated in a few
Receiver Operating Characteristic Analysis
with many risk factors. One hundred six youngsters
had risk levels of 24% or less. The overall rate of Regression analyses provided a range of predicted
disorder in this group was 16 of 106, or 15%. A much outcomes; there is no single index of the degree of
larger degree of risk was concentrated in the remaining correspondence between predictor and outcome. Figure
33 youngsters. The rate of disorder was high (17 cases 1 presents the model results graphically using Receiver
in 33 youngsters, or 52%) and increased as the risk Operating Characteristic analysis (Swets, 1986a,b). As
increased within this group (Table 3). is evident, the obtained curve is far above the diagonal
line, which would indicate no effect. Hence, at all
Clinical Model
points it provides better predictive power than chance.
A clinical model was constructed on the basis of The odds ratio is different at each point on the curve,
the three predictor variables (Table 4). These variables with the lowest odds ratio being 5.65 (confidence
interval, 2.02 to 15.7), and is considerably higher at
TABLE 3
other points. For comparison, the four-level clinical
Estimate of Risk in Selected Groups model is also presented in Figure 1.
Likelihood of No. of Children % Who
Becoming Ina (%) in Subsample . Became III
DISCUSSION
0-9.5 31 6
9.5-13 28 11 Parental and child diagnostic risk factors and related
13-16 6 17 variables strongly predicted the onset of episodes of
16-24 41 22 adolescent affective disorder in the interval between
24-36 14 43
36-51 10 50 first and second assessment, confirming the main hy-
51-71 6 67 pothesis of this study. Dimensional predictors based
71-100 3 100 on duration of disorder or total number of diagnoses
a Groupings have been taken from the ReceiverOperating Char-
were better predictors than presence or absence of the
acteristic curve of Figure 1. factor alone, although use of presence or absence alone

284 J. AM . ACAD. CHILD ADOLESC . PSYCHIATRY. 35:3. MARCH 1996

 PREDI CTION OF EPI SOD ES

does yield a viable clinical model. When parental MDD Our data indicate that the same approach should
is present without other factors and is of short duration, be applied in this nonreferred sample to the emergence
it has little predictive effect; in combination with other of adolescent affective disorder. Parental affective disor-
factors it has a powerful effect. der is an identifier of a constellation of risk factors
Large-scale epidemiological studies, using nonre- that together operate to determine poor outcome.
ferred or randomly sampled population-based surveys Knowledge of the component risk factors, i.e., duration
(Offord et al., 1992; Rutter, 1986), demonstrate the of affective disorder, nonaffective disorder, and child's
influence of a constellation of nonspecific risk factors history of disorder, is essential in understanding who
in predicting childhood disorders. These studies have will or will not become ill in a nonreferred sample.
often included a parental mental illness factor. Rutter Parental affective disorder in this sample was highly
(1986) identified six major risk factors: parental crimi- confounded with other risk variables. These variables
nality, maternal psychopathology, overcrowding or hav- need to be assessed and included in order to predict
ing a large family size, presence of severe marital the onset of affectivedisorder. Moreover, in this nonre-
discord, social disadvantage, and admission into the ferred sample, risk was not randomly or equally distrib-
care of local authorities (placement in a foster home). uted among the group. Risk clustered in a relatively
The presence of only one risk factor did not heighten small group in which the variables occurred together
the likelihood that a youngster would experience diffi- and was associated with higher rates of affective disor-
culty, but two or more dramatically increased the der, whereas most of the sample had very few' cases
likelihood. In a pioneering study of younger children of affective disorder.
followed through age 4 years, including a group of This is the first nonreferred sample in which parental
children of mothers with depression, the same finding diagnostic variableshave been examined in considerable
emerged (Sameroff et al., 1987), i.e., the number of detail, as lifetime assessments of parental diagnoses are
risk factors predicted poor outcome rather than any not common in epidemiological studies of children
one in particular. Offord and colleagues, in a large- because of the costs associated with such assessments.
scale epidemiological study in Ontario, obtained similar These data compellingly argue that consideration of
findings (Offord et al., 1987). these variables is needed.
A number of these variables, e.g., parental nonaf-
fective disorder or number of child diagnoses, are not

0.9
'"'" a ...
specific to parental affective disorder. These most likely
indicate overall impairment in either parental function-
'"'" ......
......

.
0.8 ........,4
ing or child functioning prior to adolescence. Further-
a
~
~ '"'" "''q,.
" ,,
more, the presence of comorbidity or multiple diagnoses
within an individual parent (Weissman et al., 1987,

.
0.7

j
.
J: 0.6 '"'" " ,,,,, 1993) or child (Keller et al., 1991) is more impairing
~
!
o.s '"'" " ,,, than a single diagnosis. Thus the interaction of these
nonspecific variables with the specific stressor of paren-
l> 0.4
is
:2
'"'"-, '" ,,, tal MDD may have led to the onset of affective
disorders, but this cannot be resolved from the data.
't)
.l: 0.3 A number of important variables described elsewhere
.t..
_ _ Logistic __ -<> __ Four ... Parem
·ft Model Level Non·Aff in the literature exerted no influence in the model.
0.2
<ll The first of these is social class. Poverty (Bruce et al.,
Child No
0.1
• Parent
MOD • Diagnosis InConnation 1991) has been shown to be a powerful predictor of
depressive disorder in adults; however, as yet the data
o ~ ~ ~ U U MUM M
are unequivocal or unclear in children. Social class
Spectnclty=ProbBblUt, Corre, t Wbe. Subject wm Not Get In was not examined because this sample was relatively
homogeneous in social class, i.e., all had health insur-
Fig. 1 Receiver Operatin g Characteristic analysis for identification of
high-risk patients. Non-Aff = nonaffective disorder; MDD = major de- ance. None of the subjects were living in poverty and
pressive disorder. prior analyses had demonstrated that social class had

] . AM. ACAD . CHILD ADOLESC . PSYCHIATRY, 35:3, MARCH 19 96 285

BEARDSLEE ET AL.

no effect on outcome (Beardslee et al., 1988), possibly derived from extensive interviews with parents and
due to the homogeneity of social class. Two other children. However, most of the diagnoses made in
variables identified in our alternative model, gender both parents and children were retrospective, i.e., not
and marital status, were not predictive of outcome. current at the point of assessment. Biases in recall have
Gender has been well described as a predictor for been described in studies of adults (Aneshensel et aI.,
MOD. It has been well documented that in adulthood 1987; Bromet et aI., 1986; Roberts, 1991) and hence
the rates of disorder in women are twice as high as in may have operated in our duration variables (i.e., recall
men (Robins and Regier, 1991; Weissman and Myers, of duration of past episodes may not be as accurate
1978). A developmental perspective may provide the as data gathered at the time an episode is active). In
best understanding of the influence of female gender addition, the number of outcome events is small relative
on onset of affective illness in adolescence. The cohort to the size of the sample, and the sample is relatively
examined in this sample was passing through the age small. Larger samples followed over longer time inter-
at which there is a reversalin the sex ratio for depression, vals would provide more precise results. Furthermore,
from a higher rate in males to a higher rate in females. shorter intervals of measurement could give more pre-
Finally, marital status, while a significant predictor in cise information about timing and duration of episodes.
the univariate analyses, did not emerge as a predictor Finally, although the number of parents who received
in the multivariate analyses. This is most likely because diagnoses in this nonreferred sample is high, the num-
it was completely confounded with the parental af- ber of individuals with illnesses of severity similar to
fective and nonaffective disorder. Examination of other those who present for clinical treatment is small, i.e.,
samples where this confounding was not complete very few of the parents were hospitalized 'for MOD
would elucidate this overlap. (the most common affective disorder experienced by
This sample was too small to examine some variables the parent). It will be important to explore these
within adolescent subjects that have been found to variableswithin a clinically referred sample, for example
predict the onset ofdepression. On the basisof extensive one with a group of individuals who had all been
cross-sectional and follow-up assessments of more than hospitalized for MOD. Moreover, this was a racially
1,700 adolescents, Lewinsohn and colleagues (1994) homogeneous sample. The effect of these and other
identified a number of variables within adolescents that variablesneeds to be explored in non-Caucasian samples
predicted future depressive episodes. These included as well as in different cultural groups.
past suicide attempts, past depressive disorders, low In these analyses we defined risk using broad defini-
energy levels, internalizing problems, and lifetime num- tions, i.e., the occurrence of any affective disorder in
bers of symptoms. All of these reflect aspects of past either parent at any point by lifetime retrospective
child affective diagnoses and past other psychiatric assessment led to inclusion in the category of parental
diagnoses, which were entered in our model and were affectivedisorder. The most common parental diagnosis
significant. Lewinsohn and colleagues (1993) also devel- by far was affective disorder. The rates of MOD
oped a model of summed risk variables, in which the described in the parents are similar to those found in
presence ofa number of these variables greatly increased epidemiological studies using the same instrument, the
the likelihood that the child would become depressed, SAOS-L, by Weissman and Myers (1978). Moreover,
that is similar to our clinical model. our estimates of lifetime MOD are similar to those
All of our findings should be considered in the found recently by Kessleret al. (1994), although higher
context of the study design. The interval for outcome than those obtained in the Epidemiologic Catchment
measurement was relatively short, i.e., 4 years (average Area study (Robins and Regier, 1991). Similarly, a
age, 14 to 18 years). The greatest likelihood of onset relatively high rate of diagnoses in youngsters was
of affective disorders is between 18 and 35 years, found using extensive retrospective methods. This rate
i.e., early adulthood (Institute of Medicine, 1994). is similar to that found by Lewinsohn et al. (1994) in
Therefore, more precise predictions could be made a nonreferred sample for MOD.
with a longer interval of follow-up. This is particularly A stringent test of the hypothesis was examined in
true since childhood depression, like adult depression, this study because only the onset of serious affective
is an episodic disorder. The T 1 predictor variables were disorder in a relatively brief time frame was examined.

286 J. AM. ACAD. CHILD ADOLESC. PSYCHIATRY. 35:3. MARCH 1996

 PREDICTION OF EPISODES

Rigorous random sampling applied to a group represen- associated with parental affective disorder and examin-
tative of a much larger population and inclusion of all ing their combined effects is the most useful approach.
members sampled for analyses indicates that these
factors are likely to operate in other similar samples
in health maintenance organizations or other similar REFERENCES
health care organizations and, hence, the findings are
Andreasen NC, McDonald-Scott P, Grove WM (1982), Assessment of
generalizable. A high rate (91%) of follow-up of the reliability in multicenter collaborative research with a videotape ap-
sample 4 years later, as well as the blind, independent proach. Am J Psychiatry 139:876-882
Aneshensel CS, EsrradaAL, Hansell MJ, Clark VA (1987), Social psycholog-
assessments of the outcome variables, strengthen the ical aspects of reporting behavior: lifetime depressive episode reports.
evidence for the importance of these variables. J Health Soc Behav28:232-246
BeardsleeWR, Keller MB, Lavori PW, Klerman GK, Dorer DJ, Samuelson
H (1988), Psychiatric disorder in adolescent offspring of parents with
Clinical Implications affective disorders in a non-referred sample. J Affict Disord15:313-322
Beardslee WR, Keller MB, Lavori PW, Staley J, Sacks N (l993a), The
The three factors identified in this study-parental impact of parental affective disorder on depression in offspring: a
longitudinal follow-up in a nonreferred sample. JAm Acad ChildAdolesc
MDD, parental nonaffective disorder, and history of Psychiatry 32:723-730
child diagnoses-are all easily recognizable in clinical BeardsleeWR, Salt P, Porterfield K et al. (l993b), Comparison ofpreventive
practice and can be obtained in a brief history with interventions for families with parental affective disorder. JAm Acad
ChildAdolesc Psychiatry 32:254-263
parents in a clinical setting. Youngsters with a clustering Beardslee WR, Wheelock I (1994), Children of parents with affecrive
of these risk factors are at much greater risk for devel- disorders: empirical findings and clinical implications. In: Handbook of
Depression in Children and Adolescents, Reynolds WM, Johnston HF,
oping affective disorder. This information provides an eds. New York: Plenum Publishers, pp 463-479
opportunity for early recognition and treatment should Bromet EJ, Dunn LO, Connell MM, Dew MA, Schulberg HC (1986),
Long term reliability of diagnosing lifetime major depression in a
these youngsters begin to develop difficulties. Height- community sample. Arch Gen Psychiatry 43:435-440
ened attention to this group in terms of preventive Bruce ML, Takeuchi DT, Leaf PJ (1991), Poverty and psychiatric status:
longitudinal evidence from the New Haven Epidemiologic Catchment
intervention (Beardslee et al., 1993b) should also be Area Study. Arch Gen Psychiatry 48:470-474
considered. Conover WJ (1980), Practical Nonparametric Statistics. New York: John
Wiley
Downey G, Coyne JC (1990), Children of depressed parents: an integrative
Conclusion review. Psychol Buill 08:50-76
Endicott J, Spitzer RL (1978), A diagnostic interview: the Schedule for
Serious parental affective disorder in combination Affective Disorders and Schizophrenia. Arch GenPsychiatry 35:837-844
with several important broad indicators of impairment, Finkleman DG (1983), Interview used in Project Competence Studies. In:
The Relationships of Children's Attributes to Levels of Competence
i.e., nonaffective disorder in parents and previous child and Familial Stress. Dissertation, University of Minnesota, Minneapolis
diagnoses in families identified in a random survey, (used with permission of the author), pp 234-302
Gershon ES, McKnew DH, Cytryn L et aI. (1985), Diagnoses in school-
are powerful predictors of episodes of affective disorder age children of bipolar affective disorder patients with normal controls.
in adolescence. This finding compellingly argues for J Affict Disord8:283-291
Hammen C (1991), Depression Runsin Families. New York: Springer-Verlag
greater attention to these variables in longitudinal risk Hammen C, Burge D, Adrian C (1991), Timing of mother and child
studies. Mfective disorder in parents serves as an iden- depression in a longitudinal study of children at risk. J Consult Clin
PsychoI59:341-345
tifier of a constellation of risk factors, which together Harrington RC, Fudge H, Rutter M, Pickles A, Hill J (1990), Adult
lead to a poor outcome in youngsters. These findings outcomes of childhood and adolescent depression. Arch Gen Psychia-
are similar to findings on nonspecific risk factors in try 47:465-473
Herjanic B, Reich W (1982), Development of a structured psychiatric
epidemiological studies. Both clinicians and researchers interview for children. Part 1: agreement between child and parent on
should more fully describe the constellation of risk individual symptoms. J Abnorm Child Psychol10:307-324
Institute of Medicine (1994), Reducing Risks fOr Mental Disorders: Frontiers
factors associated with parental affective disorder rather fOr Preventive Intervention Research. Washington, DC: National Acad-
than assessing only the presence of the disorder. This emy Press
Jones F (1977), The Rochester Adaptive Behavior Inventory: a parallel
study provides an intersection between the emerging series of instruments for assessing social competence during early and
literature on the children of parents with affective middle childhood and adolescence. In: The Origins and Course of
Psychopathology, Strauss J, Babisian H, Roff M, eds. New York: Ple-
disorder in clinical samples with the larger epidemiolog- num Press
ical surveys of common nonspecific risk variables for Keller MB, Beardslee WR, Dorer DJ, Lavori PW, Samuelson H, Klerman
GL (1986), Impact of severity and chronicity of parental affective illness
childhood psychiatric disorder in general. The findings on adaptive functioning and psychopathology in children. Arch Gen
suggest that identifying a constellation of risk factors Psychiatry 43:930-937

]. AM. ACAD. CHILD ADOLESC. PSYCHIATRY, 35:3. MARCH 1996 287

BEARDSLEE ET AL.

Keller MB , Lavori PW, Beardslee WR, W under J, Ryan N (1991), Rut ter M, ed (198 3), Developmental Neuropsychiatry. New York: Guil-
D epression in children: and ado lescents: new dara on " undertreatrnent" ford Press
and a literature review on the efficacy of available treatments. ] Affect Rutter M (1986), M eyerian psychob iology, personal ity developmenr and
Disord 21:163-1 71 the role of life experiences. Am ] Psychiatry 143:1077-108 7
Kessler RC , Mcgonagle KA, Zhao S er al. (1994), Lifetime and 12-monrh Ru tter M (1989), Pathways from childhood to ad ult life. ] Child Psycho!
prevalence of DSM-III-R psychiatri c disorders in the Un ited States. Psychiatry 30 :23-5 1
Arch Gen Psychiatry 51:8- 19 Rutter M (1990), Commenrary: some focus and pro cess considerations
Kovacs M, Goldston D (1991), Cognitive and social cognitive develop menr regarding effe ct s of par ental depressio n o n children. Deu Psycho!
of de pressed child ren and ado lescents . ] Am Acad Child Adolesc Psychia- 26:60-67
try 30:388-3 92 Sameroff AJ, Seifer R, Zax M , Barocas R (1987), Early indica tors of
Larkin BH , H irshfeld RM (1977), The Personal History of Depressive develop me nr al risk: Rochester Lon gitudin al Study. Schizophr Bull
Disorders. Rockville, M D: N ation al Institute of M ental Health, Clinical 13:383-394
Research Branch Shaffer D , Gould MS, BrasicJ et al, (198 3), A children's global assessment
Lewinsohn PM , Roberts RE, Rohde P, Seeley JR , Gotlib IH, H ops H scale (C GAS). Arch Gen Psychiatry 40:1228- 1231
(1994), Adolescenr psychopath ology: II. Psychosocial risk factors for Spit zer RL, Endicott J, Robins E (1978) , Research Diagnostic Criteria:
depression. ] Abnorm Psychol 103:302-315 rationale and reliability. Arch Gen Psychiatry 35:773-782
Lewinsohn PM, Rohde P, Seeley JR (1993), Psychosocial characreri stics Swets JA (l986a), Indices of discrimin at ion or diagnostic accuracy : their
of adolescenrs with a history of suicide attempt. ] Am Acad Child ROCs and implied models. Psychol Bull 99: I 00-117
Adolesc Psychiatry 32:60-68 Swets JA (1986b), Form ofemp irical ROCs in discrimination and d iagnostic
Nelder JA, Wedderburn RW (1972 ), Gen eralized linear models. ] R Stat tasks: implications for theory and measurement of performance. Psychol
Soc 135:761- 768 Bull 99:18 1- 198
Offord DR, Boyle MH, Racine YA et al. (1992), Outcome, prognosis, T suang MT, Faraone SV (1990), Summary and conclusions. In : Genetics of
and risk in a long itudinal follow-up stu dy. ] Am Acad Child Adolesc Mood Disorders. Baltimore: Jo hns H opkins Un iversity Press, pp 166-1 76
Psychiatry 3 1:9 16-923 Wechsler D (1958), The Measurement and Appraisal of Adult Intelligence.
Offord DR , Boyle M H, Szatmari P et al. (1987), Ontario Child Health Baltimore: Will iams & Wi lkins
Study. 11. Six-mon th p revalence of disorder and rates of service ut iliza- Wechsler D (1974), Manualfor the Wechsler Intelligence Scalefor Children-
tion. Arch Gen Psychiatry 44:832-836 Revised. New York: Psychological Co rpo ration
O rvaschel H , Pu ig-Anr ich J, C ham bers W , Tabrizi MA, Johnson R (1982), W eissman MM, Fendrich M , W arner V, Wi ckramarame P (1992) , Inci-
Retro spective assessmenr of prepu berta l major depression with the dence of psychiatric disor der in offspring at high and low risk for
Kiddi e-SADS-E. ] Am Acad Child Psychiatry 21:392-397 depression . ] Am Acad Child Adolesc Psychiatry 31 :640-648
O rvaschel H , Wal sh-Allis G , Ye W (1988), Psychopathology in children W eissman MM, Gammon GD, John K et al. (1987), Children of depressed
of paren ts with recurrenr dep ression . ] Abnorm Child PsychoI16:1 7- 28 paren ts: increased psychopathology and early onset of major dep ression .
Reich W, Herjan ic B, W elner Z, Gandhy PR (1982), D evelopmenr of a Arch Gen Psychiatry 44:847-85 3
structu red psychiarric int erview for children . Part 2. Agreemenr on Weissman MM, My ers JK (1978), AfFective disorder s in a US urb an
d iagnosis comparing child and parenr inrerviews. ] Abnorm Child community. Arch Gen Psychiatry 35:13 04-13 1I
PsychoII 0:325-3 36 Weissman MM, Prusoff BA, Gammo n G D, M erikangas KR, Leckman JF ,
Roberts RE (1991) , Mental health su rveys. In: Methods and Applications Kidd KK (198 4), Psychopathology in the children (ages 6-18) of
in Mental Health Surveys, Suzuki S, Roberts R, eds. Tokyo: U niversity depressed and nor mal paren ts. ] Am Acad Child Psychiatry 23:78-84
of Tokyo Press, pp 3-19 We issman MM, Wic krama rarne P, Adams PB et al. (1993), The relation ship
Rob ins LN , Regier DA (1991), Psychiatric Disorders in America: The between pan ic d isorder and major depr ession : a new family study. Arch
Epidemiologic Catchment Area Study. New York : The Free Press Gen Psychiatry 50:767- 780

288 J. AM. ACAD. CHILD ADOLESC. PSYCH IAT RY, 35:3, MARCH 1996