Latent Class Analysis Shows Strong Heritability of the
Child Behavior Checklist–Juvenile Bipolar Phenotype
Robert R. Althoff, David C. Rettew, Stephen V. Faraone, Dorret I. Boomsma, and James J. Hudziak
Background: The Child Behavior Checklist (CBCL) has been used to provide a quantitative description of childhood bipolar disorder
(BPAD). Many have reported that children in the clinical range on the Attention Problems (AP), Aggressive Behavior (AGG), and
Anxious-Depressed (A/D) syndromes simultaneously are more likely to meet the criteria for childhood BPAD. The purpose of this study
was to determine if Latent Class Analysis (LCA) could identify heritable phenotypes representing the CBCL-Juvenile Bipolar (CBCL-JBD)
profile and whether this phenotype demonstrates increased frequency of suicidal endorsement.
Methods: The CBCL data were received by survey of mothers of twins in two large twin samples, the Netherlands Twin Registry. The
setting for the study was the general community twin sample. Participants included 6246 10-year-old Dutch twins from the
Netherlands Twin Registry. The main outcome measure consisted of the LCA on the items comprising the AP, AGG, and A/D subscales
and means from the suicidal items #18 and #91 within classes.
Results: A 7 class model fit best for girls and an 8 class fit best for boys. The most common class for boys or girls was one with no
symptoms. The CBCL-JBD phenotype was the least common—about 4%–5% of the boys and girls. This class was the only one that had
significant elevations on the suicidal items of the CBCL. Gender differences were present across latent classes with girls showing no
aggression without the CBCL-JBD phenotype and rarely showing attention problems in isolation. Evidence of high heritability of these
latent classes was found with odds ratios.
Conclusions: In a general population sample, LCA identifies a CBCL-JBD phenotype latent class that is associated with high rates of
suicidality, is highly heritable, and speaks to the comorbidity between attention problems, aggressive behavior, and anxious/depression
Key Words: ADHD, bipolar disorder, Latent Class Analysis
and debate over the past decade (Leibenluft et al 2003). The
general description of this group of children includes prominent
ADHD symptoms coupled with aggression, out of control behav-
ior, and affective instability. Beginning with Biederman et al
(1995), many groups have described a profile on the Child
Behavior Checklist (CBCL) (Achenbach 1991) that occurs in
children with JBD that is discrete from children with ADHD alone
(Biederman et al 1995; Carlson and Kelly 1998; Dienes et al 2002;
Geller et al 1998; Hazell et al 1999; Wals et al 2001). The CBCL
profile includes elevation on the Attention Problems (AP), Ag-
gressive Behavior (AGG), and Anxious/Depressed (A/D) syn-
dromes. In contrast, ADHD children without bipolar disorder
show elevations on the AP syndrome alone. The CBCL-JBD
phenotype has shown consistent associations with the diagnosis
of bipolar disorder across samples, across countries, and across
methodologies. Mick et al’s 2003 meta-analysis of the CBCL
studies found considerable agreement between research sites
indicating that bipolar children are characterized by problems
he phenotype of children with attention-deficit/hyperac-
tivity disorder (ADHD) comorbid with juvenile bipolar
disorder (JBD) has been a source of considerable study
with aggression, mixed mania with depression, and ADHD
symptomatology (Mick et al 2003).
Building on these findings, we have demonstrated that this
phenotype is highly heritable (Hudziak et al 2005). With CBCL
data for 5418, 3562, and 1971 Dutch mono- and dizygotic twin
pairs at ages 7, 10, and 12 years, we investigated the prevalence
of and the genetic and environmental contributions to the
CBCL-JBD phenotype and compared these results with those for
CBCL-Attention Problems (CBCL-AP). With a cutpoint of T-scores
on AP, AGG, and A/D all ? 70, we found that the CBCL-JBD
phenotype occurs in approximately 1% of children at all ages
sampled. Among the children who met criteria for the CBCL-AP
phenotype, 13%–20% also met criteria for the CBCL-JBD pheno-
type. With structural equation models, the variance in the
CBCL-JBD phenotype was explained by a model that includes
additive genetic, and shared and unique environmental factors: a
profile different from the model for CBCL-AP, which showed
dominant genetic, additive genetic, and unique environmental
factors. These findings suggest that the CBCL-JBD phenotype is
different genetically from the CBCL-AP phenotype and that
further refinement of this phenotype might improve gene-finding
explorations. Using a cutpoint approach, however, has the
disadvantage of not including children who might have a score
that is subthreshold on one of these subscales but generally have
an item response profile that is very close to the CBCL-JBD
phenotype. Here, we investigate a strategy for phenotypic refine-
ment with Latent Class Analysis (LCA).
Latent Class Analysis allows the investigator to test empirically
for the existence of discrete groups who endorse similar patterns
of symptoms (Hudziak et al 1998). Using this strategy, distinct
classes of responding with regard to ADHD symptoms (Hudziak
et al 1998; Neuman et al 1999, 2001; Rasmussen et al 2002a,
2002b, 2004; Rohde et al 2001; Todd et al 2001), the CBCL
subscales of anxious/depression (Wadsworth et al 2001), AP
(Hudziak et al 1999), Aggression (van Lier et al 2003), as well as
ADHD with the Conners’ Parent and Teacher forms (Althoff et al,
submitted) have been identified. Furthermore, our group and
From the Department of Psychiatry (RRA), Massachusetts General Hospital
and Harvard Medical School, Boston, Massachusetts; Department of
ical Genetics Research Program and Department of Psychiatry and Be-
havioral Sciences (SVF), State University of New York Upstate Medical
University, Syracuse, New York; and The Free University (DIB), Amster-
dam, The Netherlands.
Address reprint requests to James J. Hudziak, M.D., University of Vermont,
Division of Behavioral Genetics, Department of Psychiatry, Given Building,
B229, 1 South Prospect Street, Burlington, VT 05405; E-mail: james.
Received May 16, 2005; revised February 8; accepted February 10, 2006.
BIOL PSYCHIATRY 2006;60:903–911
© 2006 Society of Biological Psychiatry
others have shown that heritability within latent classes are
higher than across latent classes, suggesting their utility for
phenotypic refinement in ADHD (Neuman et al 1999; Rasmussen
et al 2004). Todd et al (2003) extended this work, showing that
they could use LCA to uncover an association between a single
nucleotide polymorphism in the nicotinic acetylcholine recep-
tor ? subunit gene and the inattentive latent class of ADHD.
We hope to begin a similar enterprise by first identifying a
CBCL-JBD phenotype with LCA and show that heritability
within a latent class is stronger than across latent classes.
One argument against the use of proxies such as the CBCL-
JBD phenotype has been the lack of impairment information that
is necessary for a diagnosis in the current DSM-oriented nosology
(American Psychiatric Association 1994). Given that JBD has
been shown to be a risk factor for suicidality (Brent et al 1988)
and that suicidality is one of the most extreme examples of
impairment in childhood psychopathology, we also investigated
the prevalence of suicidal ideation as indicated on the CBCL.
Methods and Materials
Participants and Procedure
The data of the present study are derived from a large
ongoing longitudinal study that examines the genetic and envi-
ronmental influences on the development of problem behavior
in families with 3–12-year-old twins. The families are volunteer
members of the Netherlands Twin Register, kept by the Depart-
ment of Biological Psychology at the Free University in Amster-
dam (Boomsma 1998). Starting in 1987, families with twins were
recruited a few months after birth. Currently, 40%–50% of all
multiple births are registered by the Netherlands Twin Registry.
For the present study, we included data of mother report for
10-year-old twin pairs. Mothers of twins were asked to fill out
questionnaires about problem behavior for the eldest and young-
est twin at age 10 years. After 2 months a reminder was sent to
the non-responders, and after 4 months those who still did not
respond were telephoned. The continued participation rate for
the Netherlands Twin Registry is 80%. This study was approved
by the institutional review boards of both the Free University,
Amsterdam, and the University of Vermont.
For 822 same gender twin pairs, zygosity was based on
blood group polymorphisms (n ? 424) or DNA (n ? 398). For
the remaining twins, zygosity was determined by question-
naire items completed by the mother about physical similarity
and frequency of confusion of the twins by family and
strangers (Goldsmith 1991). The classification of zygosity was
based on a discriminant analysis, relating the questionnaire
items to zygosity on the basis of blood/DNA typing in a group
of same-gender twin pairs. The zygosity was correctly classi-
fied by questionnaire in nearly 95% of the cases (Rietveld et al
A family was excluded when one of the twin pair had a
disease or handicap that interfered severely with normal daily
functioning (about 2%). Table 1 gives an overview of the
number of participants, broken down by zygosity. An earlier
comparison of the parental socioeconomic status distribution
with those obtained for the general Dutch population showed
a slightly higher frequency of the middle and higher socio-
economic status groups (for details see Rietveld et al 2003).
Attrition rates as well as a detailed discussion on the repre-
sentativeness of the sample at each age are discussed in detail
elsewhere (van Beijsterveldt et al 2003).
Problem behavior was measured with the CBCL/4–18
(Achenbach 1991), a questionnaire of 118 items developed to
measure problem behavior in 4–18-year-old children. Parents
were asked to rate the behavior of the child for the preceding 6
months on a three-point scale.
For the CBCL/4–18, eight syndrome scales were composed
according to the 1991 profile (Achenbach 1991) that has been
normed specifically for the Dutch (Achenbach et al 1987; Ver-
hulst et al 1988). We specifically used the 44 items from the AP,
AGG, and A/D subscales. Because items on the CBCL are listed
on a Likert scale from 0 to 2, items from the AP, AGG, and A/D
subscales were first truncated to create dichotomous variables
with either 1 (“somewhat true”) or 2 (“often true”) considered as
positive responses and 0 (“not true”) considered as a negative
Twin pairs where one twin had missing items on one or more
of these subscales were not included in the analysis. The
numbers of excluded participants by zygosity is listed in Table 1.
Latent Class Analysis is a form of categorical data analysis that
hypothesizes that it is possible to account for the observed
symptom (or item) endorsement profiles of respondents in terms
of some small number of mutually exclusive respondent classes
(M), with each class having its own set of symptom endorsement
probabilities. Latent Class Analysis presupposes the existence of
discrete latent categories or classes, distinguishing it from factor
analysis, which assumes continuous latent variables are present.
Local independence is assumed (i.e., under an M-class solution,
the conditional probabilities of endorsing a set of items are
statistically independent for a given class) (Goodman 1974). If
the underlying latent variable is continuous rather than categor-
ical, then the LCA-derived classes will reflect differences in
severity, whereas discrete classes of responding will emerge
from the analysis if the underlying latent structure is categorical.
The parameter estimates that result from LCA are: 1) probabilities
of class membership assignment for individuals, and (2) symp-
tom endorsement probabilities for each class.
Latent class models were fitted by means of an Expectation
Maximization (EM) algorithm (Dempster et al 1977), with the
program Latent Gold (Vermunt and Magidson 2000). Models
estimating 1-class through 10-class solutions were compared. To
calculate the best fitting model, we first ensured goodness of fit
with the bootstrapping algorithm built into Latent Gold—a step
that is essential when dealing with sparse data matrices such as
these—and then compared the change in the Bayesian Informa-
tion Criterion (BIC) when moving from an M to an M ? 1 class
solution. The BIC is a goodness-of-fit index that considers the
Table 1. Sample Composition and Breakdown by Zygosity
Number of Participants
MZ, monozygotic; DZ, dizygotic; DOS M_F, dizygotic opposite gender
male eldest; DOS F_M, dizygotic opposite gender female eldest.
904 BIOL PSYCHIATRY 2006;60:903–911
R.R. Althoff et al
Faraone SV, Althoff RR, Hudziak JJ, Monuteaux M, Biederman J (2005): The
CBCL predicts DSM bipolar disorder in children: A receiver operating
characteristic curve analysis. Bipolar Disord 7:518–524.
Geller B, Warner K, Williams M, Zimerman B (1998): Prepubertal and young
adolescent bipolarity versus ADHD: Assessment and validity using the
WASH-U-KSADS, CBCL and TRF. J Affect Disord 51:93–100.
Goldsmith HH (1991): A zygosity questionnaire for young twins: A research
note. Behav Genet 21:257–269.
able and unidentifiable models. Biometrika 61:215–231.
Hazell PL, Lewin TJ, Carr VJ (1999): Confirmation that Child Behavior Check-
list clinical scales discriminate juvenile mania from attention deficit hy-
Hudziak JJ, Althoff RR, Derks EM, Faraone SV, Boomsma DI (2005): Preva-
lar disorder. Biol Psychiatry 58:562–568.
Hudziak JJ, Heath AC, Madden PF, Reich W, Bucholz KK, Slutske W, et al
(1998): Latent class and factor analysis of DSM-IV ADHD: A twin study of
Hudziak JJ, Wadsworth ME, Heath AC, Achenbach TM (1999): Latent class
Leibenluft E, Charney DS, Towbin KE, Bhangoo RK, Pine DS (2003): Defining
clinical phenotypes of juvenile mania. Am J Psychiatry 160:430–437.
Lewinsohn PM, Klein DN, Seeley JR (2000): Bipolar disorder during adoles-
cence and young adulthood in a community sample. Bipolar Disord
Mick E, Biederman J, Pandina G, Faraone SV (2003): A preliminary meta-
analysis of the child behavior checklist in pediatric bipolar disorder. Biol
Neuman RJ, Heath A, Reich W, Bucholz KK, Madden PAF, Sun L, et al (2001):
Latent class analysis of ADHD and comorbid symptoms in a population
sample of adolescent female twins. J Child Psychol Psychiatry 42:933–
Rasmussen ER, Neuman RJ, Heath AC, Levy F, Hay DA, Todd RD (2002a):
Replication of the latent class structure of Attention-Deficit/Hyperactiv-
ity Disorder (ADHD) subtypes in a sample of Australian twins. J Child
Rasmussen ER, Neuman RJ, Heath AC, Levy F, Hay DA, Todd RD (2004):
Familial clustering of latent class and DSM-IV defined attention-deficit/
hyperactivity disorder (ADHD) subtypes. J Child Psychol Psychiatry 45:
Comparison of male adolescent-report of attention-deficit/hyperactiv-
Heritability of attention problems in children: I. Cross-sectional results
from a study of twins, age 3–12 years. Am J Med Genet 117B:102–113.
Heritability of attention problems in children: Longitudinal results from
a study of twins, age 3 to 12. J Child Psychol Psychiatry 45:577–588.
(2001): Factor and latent class analysis of DSM-IVADHD symptoms in a
Seiffge-Krenke I, Kollmar F (1998): Discrepancies between mothers’ and
fathers’ perceptions of sons’ and daughters’ problem behaviour: A lon-
gitudinal analysis of parent-adolescent agreement on internalising and
Todd RD, Lobos EA, Sun LW, Neuman RJ (2003): Mutational analysis of the
nicotinic acetylcholine receptor alpha 4 subunit gene in attention defi-
cit/hyperactivity disorder: Evidence for association of an intronic poly-
morphism with attention problems. Mol Psychiatry 8:103–108.
Todd RD, Rasmussen ER, Neuman RJ, Reich W, Hudziak JJ, Bucholz KK, et al
activity disorder in a population sample of adolescent female twins. Am
J Psychiatry 158:1891–1898.
van Beijsterveldt CE, Bartels M, Hudziak JJ, Boomsma DI (2003): Causes of
nal genetic analysis in Dutch twins. Behav Genet 33:591–605.
van Lier PA, Verhulst FC, van der Ende J, Crijnen AA (2003): Classes of
disruptive behaviour in a sample of young elementary school children.
of syndromes derived from the child behavior checklist for American
Vermunt JK, Magidson J (2000): Latent GOLD User’s Guide. Belmont, Massa-
chusetts: Statistical Innovations.
Wadsworth ME, Hudziak JJ, Heath AC, Achenbach TM (2001): Latent class
analysis of child behavior checklist anxiety/depression in children and
Wals M, Hillegers MH, Reichart CG, Ormel J, Nolen WA, Verhulst FC (2001):
Whittington CJ, Kendall T, Fonagy P, Cottrell D, Cotgrove A, Boddington E
Systematic review of published versus unpublished data. Lancet 363:
Wozniak J, Spencer T, Biederman J, Kwon A, Monuteaux M, Rettew J, et al
(2004): The clinical characteristics of unipolar vs. bipolar major depres-
sion in ADHD youth. J Affect Disord 82(suppl 1):S59–S69.
R.R. Althoff et al
BIOL PSYCHIATRY 2006;60:903–911 911