Subtypes of depression in a nationally representative sample
Natacha Carragher⁎, Gary Adamson, Brendan Bunting, Siobhan McCann
Psychology Research Institute, University of Ulster, N. Ireland, United Kingdom
Received 28 January 2008; received in revised form 18 May 2008; accepted 18 May 2008
Available online 21 July 2008
Background: Continued research efforts aim to elucidate the heterogeneity in depression. The identification of meaningful and
valid subtypes has implications for research and clinical practice. Based on patterns of depressive symptomatology, this study
identified a typology of depressive syndromes using data from a large, nationally representative survey.
Methods: Analyses were based on a subsample of 12,180 respondents from the 2001–2002 Wave of the National Epidemiologic
Survey on Alcohol and Related Conditions (NESARC). Latent class analysis was applied to the DSM-IV ‘A’ criteria for major
depression to identify homogenous subtypes or classes of depressive syndromes. Associations between the emergent latent classes
and demographic and clinical characteristics were assessed.
Results: Three clinically relevant subtypes were identified, in addition to a class who reported few depressive symptoms: severely
depressed (40.9%), psychosomatic (30.6%), cognitive–emotional (10.2%) and non-depressed (18.3%). The odds of experiencing
negative life events, psychiatric disorders, and having a family background of major depression were significantly higher for the
severely depressed, psychosomatic and cognitive–emotional classes, compared to the non-depressed class. Several unique
differences between the latent classes also emerged.
Limitations: Methodological shortcomings included: reliance on lay interviewer-administered structured interviews to determine
diagnoses; basing sample selection on the endorsement of screener items; and, using measures of ‘any anxiety disorder’, ‘any mood
disorder’, and ‘any personality disorder’ to determine psychiatric disorder prevalence rates.
Conclusions: Significant heterogeneity in depressive symptomatology exists in this U.S. sample. Profiling symptom patterns is
potentially useful as a first step in developing tailored intervention and treatment programmes.
© 2008 Elsevier B.V. All rights reserved.
Keywords: Latent class analysis; Depression; NESARC; Heterogeneity
1. Subtypes of depression in a nationally
Depression is generally considered to be an etiologi-
cally and clinically heterogeneous condition. It is asso-
ciated with a wide range of risk factors and affected
onset, and response to treatment (Kendler et al., 1999).
Recognition of such heterogeneity in depression has long
motivated research interest in the identification of mean-
ingful and valid subtypes. Subtyping holds promise for
guiding research on etiology and better informing clinical
management by improving diagnostic practices and
prevention approaches, and refining differential treatment
Journal of Affective Disorders 113 (2009) 88–99
⁎Corresponding author. School of Psychology, University of Ulster
at Magee, Northland Road, Derry BT48 7JL, Northern Ireland, United
Kingdom. Tel.: +44 28 71375367.
E-mail address: Carragher-N@ulster.ac.uk (N. Carragher).
0165-0327/$ - see front matter © 2008 Elsevier B.V. All rights reserved.
typologies is particularly appealing in the context of de-
burden of the condition (Murray and Lopez, 1997).
Heterogeneity is said to be present when a population
can be separated into distinct subpopulations or clusters.
More specifically, heterogeneity is observed when it is
possible to identify the subpopulations based on an
heterogeneity, the variables giving rise to heterogeneity
are unknown and subpopulation membership must be
are termed latent classes since subpopulation member-
ship is unobserved (Lubke and Muthén, 2005).
Research efforts to elucidate the heterogeneity in
depression have utilised a wide variety of latent variable
techniques including factor analysis (Aggen et al., 2005;
Muthén, 1989; Simon and von Korff, 2006); discriminant
function analysis (Sen, 1987); cluster analysis (Andreasen
et al.,1980;Blashfield and Morey, 1979;Cox et al., 2001;
Scotte et al., 1997); grade of membership analysis (Blazer
et al., 1988, 1989; Davidson et al., 1988); and, more
recently, latent class analysis (Chen et al., 2000; Crum
et al., 2005; Eaton et al., 1989; Kendler et al., 1996;
psychiatric interviews) and the resultant subtypes have
demographic variables, familial liability to psychiatric
illness, and clinical features, including treatment response
and relapse rate. On a final methodological note, these
and clinical samples derived from psychiatric hospitals,
outpatient clinics, and primary care practices. Basing a
typology on data from treated populations may, however,
introduce selection bias since individuals in clinical set-
the population and are likely to display greater symptom
to the wider population, thereby limiting their utility in
terms of public health initiatives (Kendler et al., 1996;
Moss et al., 2007). Based on this rationale, this paper is
based on data from the general population.
A number of variables have been identified as im-
portant in capturing the heterogeneity in depression. One
of the most replicated findings in psychiatric epidemiol-
of major depression among women compared to men
(e.g., Weissman et al., 1996). Similarly, investigations
consistently demonstrate that psychiatric disorders such
as major depression are “over-represented in the lower
social strata” (Miech et al., 1999: 1096). Several lines of
research garner support for a significant association
between major depression and the experience of stressful
background of major depression (Weissman et al., 2005).
minority groups have a lower prevalence of psychiatric
disorders, despite the higher levels of social adversity
encountered (Breslau et al., 2006). Research also
suggests that having a partner acts as a protective buffer
against depression (Dehle et al., 2001). In contrast, the
literature on geographic associations with major depres-
sion is inconclusive. Some mental health surveys lend
whereas other researchers cite lower depression rates in
urban areas (Probst et al., 2006), and others fail to find
significant differences (Kovess-Masfety et al., 2005).
Inconsistent findings are also evident in relation to age
(see Jorm, 2000).
Comorbidity represents another potential source of
heterogeneity and has received considerable attention in
the literature in recent years. Research has documented a
high rate of comorbidity between major depression and
anxiety disorders (Wittchen et al., 2000); substance use
disorders, involving alcohol and illicit drugs (Davis
et al., 2005; Weissman et al., 1996); and personality
disorders (Corruble et al., 1996; Farabaugh et al., 2004).
Comorbidity between major depression and nicotine
dependence is similarly well established (Fergusson
et al., 2003). Epidemiological studies, as used in this
paper, are particularly suited to the investigation of
comorbidity since they are not confounded by treatment
seeking status (Barry et al., 2008).
In current nomenclature, unipolar and bipolar
disorders are conceptualised as being qualitatively
different and categorised as “separate branches on the
mood disorder diagnostic tree” (Cuellar et al., 2005:
309). However, in recent years several converging lines
of research have garnered support for a continuum be-
tween bipolar disorders and major depressive disorder,
in line with Kraepelin's classification of mood disorders
(for an overview see Benazzi, 2005). Longitudinal
analysis, for instance, suggests that patients initially
hospitalised for unipolar depression have a significant
risk of developing bipolar symptoms (e.g., Goldberg
et al., 2001). At the unipolar end of this spectrum,
epidemiological surveys and clinical samples provide
evidence for a progression over time from dysthymia to
major depression, with the two disorders often occurring
concurrently — termed ‘double depression’ (Alonso
et al., 2004; Bell et al., 2004).
This paper employed latent class analysis to identify
latent classes or subtypes of depressive syndromes. The
89 N. Carragher et al. / Journal of Affective Disorders 113 (2009) 88–99
emergent typology was subsequently validated by
reference to covariates cited in the extant literature,
reviewed above, as playing an important role in the
patterning of depression. This flexible modelling
approach is a popular tool in the subtyping literature
used to capture unobserved population heterogeneity,
and is particularly useful since it can simultaneously
accommodate mixed levels of measurement as well as
descriptive and predictive models.
Muthén and Asparouhov (2006) advocate the use of
large samples in investigations of population hetero-
geneity to ensure that sufficient numbers of respondents
endorse the respective criteria. Large nationally repre-
sentative surveys also have the added advantage of
yielding stable parameter estimates. Accordingly, this
study utilised data from the National Epidemiologic
Survey on Alcohol and Related Conditions (NESARC).
Given that the NESARC is the largest prevalence study
of psychiatric disorders conducted to date, the findings
hold promise of providing a clearer picture of depres-
sion. Furthermore, this paper extended earlier efforts to
subtype depression by basing analyses on current DSM-
of the NESARC, conducted by the National Institute on
Alcohol Abuse and Alcoholism (NIAAA; Grant et al.,
2003a). The NESARC is a nationally representative sur-
vey, which targeted the civilian, non-institutionalised
population living in the United States, including the
District of Columbia, Alaska and Hawaii. Face-to-face
personal interviews were conducted with 43,093 respon-
dents, aged 18 years and older. The overall response rate
was 81%. Various subpopulations, including Blacks,
adjusted to reflect over-sampling and household- and
person-level non-response. The weighted data were then
the 2000 Census. For further details regarding the
NESARC methodology see Grant et al. (2005).
2.2. Assessment of depressive symptoms
Depressive symptoms were assessed using the
Alcohol Use Disorder and Associated Disabilities Inter-
view Schedule-DSM-IV (AUDADIS-IV: Grant et al.,
2001), a fully structured diagnostic interview for use by
non-clinician interviewers. Test–retest reliability of the
AUDADIS-IV measures of major depression has been
shown to be good (kappa=0.64–0.67; Canino et al.,
1999; Chatterji et al., 1997; Grant et al., 2003b) and a
clinical reappraisal study reported good correspondence
between AUDADIS-IV diagnoses and psychiatrist
evaluations (kappa=0.64–0.68; Canino et al., 1999).
The AUDADIS-IVincluded twenty-one dichotomous
symptom item questions that separately operationalised
the nine DSM-IV ‘A’ criteria for major depression
(American Psychiatric Association, 2000). Similar to
other diagnosticinterviews (e.g.,Composite International
Diagnostic Interview: World Health Organization, 1990),
to those respondents who endorsed the lifetime occur-
rence of a 2-week period of depressed mood or loss of
interest in activities.
Basing the sample on the endorsement of screener
items precluded the use of these non-independent
questions as indicators of depression in the analyses
(cf. Slade and Andrews, 2005). As screeners, the items
offer little in the way of variability of responses and
therefore their exclusion would be unlikely to signifi-
cantly impact on the findings. In consequence, the ana-
lyses focused on seven, rather than nine, aggregated
diagnostic criteria: (i) appetite/weight change; (ii) sleep
feelings of worthlessness/excessive guilt; (vi) impaired
concentration/indecision; (vii) death/suicidal ideations. A
each criterion; where a criterion had multiple symptom
criterion was considered to be present.
We used the aggregated diagnostic criteria (or
item parcels in other words) because they offer a
number of advantages over over-disaggregated, indivi-
dual symptom items. As Little et al. (2002) point out,
parcels provide more reliable latent variables by reduc-
ing random error and requiring fewer estimated para-
meters. In turn, this increases the reliability of the
model's structural coefficients and provides a more
parsimonious model. This analytic strategy has been
used in previous typology studies of depression (e.g.,
Chen et al., 2000).
A diagnosis of major depressive disorder (MDD)
was not required for inclusion in the analyses (cf.
Kendler et al., 1996; Sullivan et al., 1998) in light of
research documenting the clinical and public health
significance of subthreshold depressive symptomatol-
ogy (e.g., Goldney et al., 2004). Finally, analyses were
restricted to those respondents who had complete
information on all items relating to depressive sympto-
90 N. Carragher et al. / Journal of Affective Disorders 113 (2009) 88–99
2.3. Latent class analysis
Latent class analysis (LCA) was used to empirically
identify subtypes of depressive syndromes. Often
described as a categorical variant of factor analysis,
LCA assumes that observed variables are indicators of
an unobserved, latent variable and attempts to explain
this relationship in terms of a small number of
subgroups or classes. Succinct introductions to LCA
are available elsewhere (e.g., Hagenaars and McCutch-
As Nylund (2007) points out, there is currently no
consensus in the literature regarding a single statistical
index that identifies the most appropriate number of
classes in a given population. Thus, a series of models
with a successive number of classes were specified
and selection of the optimal model that combines
goodness of fit and parsimony was based on conceptual
considerations and various statistical fit indices. Sta-
tistical indices reported here include: the Akaike Infor-
mation Criterion (AIC; Akaike, 1974); Bayesian
Information criterion (BIC; Schwartz, 1978); sample-
size adjusted BIC (SSABIC; Sclove, 1987); Lo–
Mendell–Rubin likelihood ratio test (LMR-LRT; Lo
et al., 2001); and an entropy measure (Ramaswamy
et al., 1993).
The AIC, BIC and SSABIC are goodness of fit
measures commonly used for comparison across
competing models: the lowest value on each criterion
indicates the best best-fitting model. The BIC has been
recently found to be one of the most reliable indicators
in determining the number of latent classes (Nylund,
2007). Another useful tool for class enumeration is the
LMR-LRT, which assesses the improvement in fit
between competing models: a non-significant value
(pN0.05) suggests that the model with one fewer class
provides a more parsimonious fit to the data. Based on
the posterior class membership probabilities, entropy
evaluates how well each of the classes is separated and
represented by the data; values range from 0 to 1, with
high values preferred. The average conditional prob-
abilities for class assignment similarly assess classifica-
tion quality and accuracy; values approaching or
exceeding 0.80 are preferred. Finally, selection of the
best-fitting model was based on whether the model
reflected coherent, conceptually meaningful subgroups
and adequately accounts for the heterogeneity in the
All analyses were implemented using Mplus version
5.1, a statistical modelling programme which can
accommodate complex design methodology (Muthén
and Muthén, 1998–2008). Specifically, the NESARC
complex survey features of stratification, clustering and
sampling weights were taken into account in the
parameter estimations and the standard error calcula-
tions. Methods appropriate for subpopulation analyses
were employed and models were specified using
maximum-likelihood estimation with robust standard
errors (MLR). The MLR estimator is well suited for
analysing data based on complex survey designs.
Finally, in mixture analysis, models are vulnerable to
converging on local rather than global solutions. In
order to avoid the issue of local maxima and to ensure
that all values converged on identical solutions, 500
random sets of starting values were used initially and 10
final stage optimisations.
2.4. Inclusion of covariates in the measurement model
Following identification of the best-fitting latent class
help describe the heterogeneity in depression and to sub-
stantiate the validity of the emergent classes or subtypes.
Odds ratios (ORs) and accompanying confidence inter-
vals (CIs) were calculated to evaluate these associations.
2.4.1. Demographic variables
Demographic variables of interest included: age;
educational attainment; total personal income in the last
12 months; race/ethnicity; current marital status;
urbanicity; and, gender.
2.4.2. Family history of major depression
whether any of their first- and second-degree relatives had
ever experienced major depression. First-degree relatives
comprised biological parents or children, and full siblings.
Second-degree relatives related to full siblings of both
biological parents, and biological paternal and maternal
grandparents. For the purposes of the present study, family
background of major depression was treated as a con-
degree relatives only’, ‘first degree relatives only’, and
Institute on Alcohol Abuse and Alcoholism, 2006).
2.4.3. Negative life events
Respondents were asked whether they had experi-
enced 12 different types of negative life events in the
1The analysis was also performed treating family background of
major depression as an ordinal categorical variable. Apart from being
computationally demanding, there was little, if any, difference in the
91N. Carragher et al. / Journal of Affective Disorders 113 (2009) 88–99
12 months prior to interview. A continuous measure of
stress was created based on the number of negative life
events reported (cf. Dawson et al., 2007).
2.4.4. Psychiatric disorders
Several disorders were diagnosed in the NESARC,
the reliability and validity of which have been docu-
mented in several studies (e.g., Grant et al., 2003b). For
the purposes of this paper, three mood disorder
diagnoses (i.e., dysthymia, mania, hypomania) were
aggregated to create a single variable indicating the
presence of ‘any other mood disorder’. Similarly, a
dichotomous measure of ‘any anxiety disorder’ was
coded as positive for individuals who met criteria for
panic disorder with/without agoraphobia, social phobia,
specific phobia, or generalised anxiety disorder. A
dichotomous measure of ‘any personality disorder’ was
created to reflect a diagnosis of antisocial, avoidant,
dependent, obsessive–compulsive, paranoid, schizoid,
or histrionic personality disorder (cf. Dawson et al.,
2007). These measures all reflect lifetime estimates.
Finally, respondents were asked a series of questions
relating to alcohol use and, on this basis, past-year
diagnoses of alcohol abuse and/or dependence were
2.4.5. Illicit drug use and nicotine
The NESARC enquired into the use of ten different
types of illicit drugs in the past year: amphetamines,
opioids, sedatives, tranquillisers, cocaine, inhalants/
solvents, hallucinogens, cannabis, heroin, and other
drugs (e.g., antidepressants). Illicit drug use was
considered to be positive if the respondent reported
using any of these drugs (cf. Dawson et al., 2007).
Finally, respondents answered an extensive list of
symptom questions assessing nicotine dependence and
a lifetime diagnosis was made accordingly.
2.5. Further validation of the measurement model
In a recent paper based on the NESARC, Hasin et al.
(2005) reported a MDD lifetime prevalence of 13.23%.
In an effort to further substantiate the validity of the
measurement model, we examined the lifetime MDD
prevalence rates of the emergent latent classes.
3.1. Demographic characteristics
The demographic distribution of the subsample of
respondents who screened positive for depressive symp-
tomatology and the remaining NESARC sample are
displayed in Table 1.
3.2. Estimation of the number of latent classes
Five latent class models were fitted to the data,
beginning with the most parsimonious one-class model
Demographic distribution of the subsample of respondents who
screened positive for depressive symptomatology and the remaining
Unweighted Weighted Unweighted Weighted
Total personal income in the last 12 months, US$
Current marital status
Never married 2910
92N. Carragher et al. / Journal of Affective Disorders 113 (2009) 88–99
through to a five-class model. The goodness of fit
indices, provided in Table 2, suggested that the best-
fitting model was a four-class solution. The AIC, BIC,
and SSABIC were markedly lower for the four-class
model compared to the earlier models. There were only
small decreases in these indices thereafter, providing
weak support for a five-class solution. The LMR-LRT
further confirmed that the five-class model was not a
significant improvement over the four-class model
(pN0.05). Classification quality, based upon the average
conditional probabilities for most likely class member-
ship, indicated that the four classes were reasonably well
defined: 0.85 for latent class one, 0.77 for latent class
two, 0.73 for latent class three, and 0.87 for latent class
four. The entropy measure (0.67) similarly suggested
that the data was adequately defined by a four-class
3.3. Symptom profiles
On a more substantive note, the four-class solution
appeared to be conceptually meaningful. Individuals
were assigned to the latent classes on the basis of their
response profile and the estimated probabilities of
endorsing the seven DSM-IV criteria, for the subsample
and four latent classes, are presented in Table 3. To aid
interpretation, the estimated probabilities are plotted in
Latent class one comprised the majority of respon-
dents (40.9%) who highly endorsed each of the depres-
sive criteria, with probabilities ranging from 0.69 for
death/suicidal ideations to 0.98 for sleep disturbances.
This class is best characterised as a ‘severely depressed’
class. Members of the second group, which comprised
30.6% of the sample, had high probabilities of expe-
riencing appetite (0.61) and sleep disturbances (0.86),
psychomotor complaints (0.70), and impaired concen-
tration/indecision (0.70). This class was labelled as a
‘psychosomatic’ class. In contrast, the third class
(10.2%) consisted of individuals with relatively high
probabilities of experiencing feelings of worthlessness/
excessive guilt (0.81), impaired concentration/indeci-
sion (0.67), and death/suicidal ideations (0.50). This
class was termed as a ‘cognitive–emotional’ class.
Finally, latent class four (18.3%) was characterised by
respondents who screened positive for depressive symp-
tomatology but displayed relatively low endorsement
rates on the DSM-IV criteria, as the conditional pro-
babilities are all below 0.26. As such, this class was
considered as a ‘non-depressed’ class.
Fit indices for a one-class model through to a five-class model
Model Fit statistics
Note. AIC, Akaike Information Criteria; BIC, Bayesian Information Criteria; SSABIC, Sample-Size Adjusted Bayesian Information Criteria;
LRT, p-value for the Lo–Mendell–Rubin Likelihood Ratio Test.
Best-fitting model in bold type.
Estimated probabilities of endorsing each of the DSM-IV criteria for the four latent classes
DSM-IV criterion Observed proportion
endorsing each criterion
Probability of endorsing each criterion
Severely depressed Psychosomatic Cognitive–emotional Non-depressed
Feelings of worthlessness/excessive guilt 0.557
93 N. Carragher et al. / Journal of Affective Disorders 113 (2009) 88–99
3.4. Validity of the latent classes
To validate the measurement model and to help
included in the model (see Table 4 for a list of the
covariates). This resulted in an improvement in fit over
the original unrestricted model: AIC=90877.905, BIC=
91685.328, SSABIC=91338.938. The entropy result
was similar (0.68) and the LMR-LRT remained sig-
nificant (p=0.007). Additionally, the average condi-
tional probabilities and prevalence rates were similar:
e.g., class 1=37.0%, class 2=30.3%, class 3=13.1%,
and class 4=19.6%. It would seem prudent to compare
the four- and five-class solutions with the inclusion of
covariates in order to ensure stability of model fit. The
LMR-LRT indicated that the five-class solution was not
a significant improvement over the four-class solution
are presented in Table 4, with the ‘non-depressed’ class
treated as the reference class. The OR describes the
proportionate change in the odds associated with a one-
unit change in the independent variable. As Table 4
indicates, the odds of experiencing negative life events
and having a family background of major depression
were significantly elevated for members of the ‘severely
depressed’, ‘psychosomatic’ and ‘cognitive–emotional’
classes, compared to the ‘non-depressed’ class. The
‘severely depressed’, ‘psychosomatic’ and ‘cognitive–
emotional’ classes were all significantly more likely to
experience psychiatric disorders, though the magnitude
of risk varied accordingly. For example, the odds ratio
associated with a personality disorder diagnosis was
highest for the ‘cognitive–emotional’ class (OR=4.06,
CI=2.88–5.71), whereas the odds ratio associated
with an anxiety disorder diagnosis was highest for the
‘severely depressed’ class (OR=4.54, CI=3.57–5.78).
was highest for the ‘severely depressed’ class
(OR=16.54, CI=10.52–26.01); in fact, this was the
largest odds ratio observed in the study. Those
individuals with a high school education and American
Indians/Alaska Natives were significantly less likely to
be present in the ‘severely depressed’, ‘psychosomatic’
Having a lifetime diagnosis of nicotine dependence
(OR=1.39, CI=1.12–1.73 and OR=2.47, CI=1.98–
3.09) or being female (OR=0.41, CI=0.32–0.53 and
OR=0.49, CI=0.38–0.63) were associated with
increased risks of being in the ‘psychosomatic’ and
‘severely depressed’ classes. Belonging to the highest-
earning income bracket signalled a decreased risk of
being in the ‘severely depressed’ (OR=0.75, CI=0.57–
0.97) or ‘cognitive–emotional’ class (OR=0.66, CI=
0.45–0.98). Interestingly, there were no significant dif-
ferences among the classes in respect of a drug use
Fig. 1. Latent class profile plot of the seven DSM-IVSSS major depression criteria.
94N. Carragher et al. / Journal of Affective Disorders 113 (2009) 88–99
Among individual factors, being male (OR=1.94,
CI=1.36–2.78) or having a past-year alcohol depen-
dence diagnosis (OR=3.40, CI=1.20–9.67) or a past-
year alcohol abuse and dependence diagnosis (OR=
2.30, CI=1.01–5.27) were consistent with increased
risks of being in the ‘cognitive–emotional’ class. The
likelihood of being widowed, separated, or divorced
(OR=1.58, CI=1.07–2.35) was significantly higher for
the ‘psychosomatic’ class, compared to the ‘non-
depressed’ class. Individuals who were aged 30–
44 years (OR=0.70, CI=0.54–0.91) or Hispanic
to be present in the ‘psychosomatic’ class. Finally,
individuals who resided in a rural area (OR=0.73,
CI=0.56–0.94) or who were never being married
(OR=1.47, CI=1.21–1.78) had significantly higher
to the ‘non-depressed’ class.
3.5. Major depressive disorder (MDD) diagnoses
Using the posterior probabilities for most likely class
membership, the lifetime prevalence estimates of MDD
were examined for each latent class to further sub-
stantiate the emergent typology. As one might expect, a
considerable proportion (92.1%) of the severely
depressed class had a lifetime diagnosis of MDD, whilst
the prevalence rate was zero in the non-depressed class.
Odds ratios and 95% confidence intervals for the latent class model with covariates
Covariate Latent classes
Severely depressed PsychosomaticCognitive–emotional
OR (95% CI) OR (95% CI)OR (95% CI)
Less than high school (referent)
Some college or high
Income in the last 12 months, US$
American Indian/Alaska Native
Asian/Native Hawaiian/Pacific Islander
Current martial status
Married or cohabiting (referent)
Widowed, separated or divorced
Negative life events (last 12 months)
Family history of major depression
Lifetime nicotine dependence diagnosis
Any lifetime mood disorder
Any other lifetime personality disorder
Any lifetime anxiety disorder
Alcohol abuse diagnosis (last 12 months)
Alcohol dependence diagnosis (last 12 months)
Alcohol abuse and dependence diagnosis (last 12 months)
Any drug use disorder diagnosis (last 12 months)
Bold type indicates a significant odds ratio.
95 N. Carragher et al. / Journal of Affective Disorders 113 (2009) 88–99
a MDD lifetime prevalence rate of 48.4% and 35.5%,
In this paper, we applied LCA to a nationally re-
presentative sample and identified three clinically
relevant homogeneous subtypes of depressive syn-
dromes. A fourth subgrouping of respondents reported
few, if any, depressive symptoms. The ‘severely
depressed’ subtype comprised the majority of respon-
dents who exhibited high endorsement rates across the
DSM-IV criteria. The profiles of the ‘psychosomatic’
and ‘cognitive–emotional’ subtypes were comparatively
less uniform. Specifically, the ‘psychosomatic’ subtype
was characterised by appetite and sleep disturbances,
psychomotor complaints, and impaired concentration/
indecision. Finally, respondents belonging to the
‘cognitive–emotional subtype’, the smallest subgroup,
reported feelings of worthlessness/excessive guilt,
impaired concentration/indecision, and death/suicidal
The subtypes share a number of characteristics in
common with other typologies, particularly that by Chen
et al. (2000) who identified ‘psychomotor’, ‘severely
depressed’ and ‘non-depressed’ latent classes in their
analysis of data from the Epidemiologic Catchment Area
Study. Similar to the present study, Eaton et al. (1989)
the present paper failed to find evidence in support of a
gradient of severity, as reported by Garrett and Zeger
(2000)inthe their LCAusing Markovchain Monte Carlo
techniques. Having said that, detailed comparisons with
such as differences in psychological measurements.
into the subtypes. For instance, a significant association
was observed between the ‘severely depressed’ subtype
residents are more likely than their urban counterparts to
experience circumstances, conditions, and behaviours
poorer physical health and typically have less access to
primary health care, specialists, health-related technolo-
gies, and other health and social services. Other factors,
lifemay contributetothe significant associationobserved
in this study. In light of research documenting a relation-
ship between social isolation and suicide (e.g., Singh and
Siahpush, 2002), the present finding has important impli-
cations in terms of allocation of resources towards the
alleviation of the public health burden of depression in
On a related note, the ‘cognitive–emotional’ subtype
was characterised by suicidal ideation symptomatology
and was significantly associated with males with a
lifetime diagnosis of alcohol dependence or abuse and
dependence. This gender and alcohol use disorder find-
ing is consistent with previous research (Sher, 2005).
Indeed, given that suicide was the eighth leading cause
of death for males in the U.S. in 2004 (Centers for
Disease Control and Prevention, CDC, 2005), this cur-
rent findings further underscores the need for an in-
creased understanding of the associations between
alcohol use and suicidality and the targeting of resources
to address these issues.
A number of similarities were apparent across the
‘severely depressed’, ‘psychosomatic’ and ‘cognitive–
class, all three subtypes were significantly more likely to
have a family background of depression, to experience
negative life events, and to be vulnerable to developing
public health and economic implications as comorbidity
is significantly associated with functional status and
quality of life (Gijsen et al., 2001); increases in service
utilisation and health care costs (Druss and Rosenheck,
1999); and negatively influences treatment outcome,
prognosis, and course as individuals typically display a
Brady, 2003; Swendsen and Merikangas, 2000). Rela-
tive to respondents with the lowest formal educational
attainment, high school graduates, on the whole, were
less likely to report depressive symptomatology. Further
consistent with the literature linking socially disadvan-
taged groups to mental illness (Miech et al., 1999),
individuals earning ≥$70,000 had a decreased risk of
reporting severely depressed or cognitive–emotional
symptoms.AsShavers (2007) suggests,individualswith
a higher level of education are likely to possess superior
information processing and problem solving skills and
able to interact effectively with healthcare providers.
promoting behaviour and lifestyles, and have better
occupational and economic conditions, as well as psy-
chological resources. Due to the disabilities and eco-
nomic cost incurred by poor mental health, intervention
strategies targeting improvements in education appear
Significant health disparities also emerged, with
Hispanics significantly less likely to report psychoso-
matic symptomsrelative totheirWhite counterparts, and
American Indians/Alaska Natives had a decreased risk,
96N. Carragher et al. / Journal of Affective Disorders 113 (2009) 88–99
in general, of reporting depressive symptomatology.
Congruent with recent findings from Breslau et al.
(2006) and Beals et al. (2005), this finding underscores
the importance of exploring factors which offer protec-
tion to these ethnic minority populations.
The ‘severely depressed’ and ‘psychosomatic’ sub-
This gender disparity in depression inwellestablished in
the literature, possibly reflecting biological, environ-
mental, or psychological differences between genders
(Kuehner, 2003). Similarly, research consistently indi-
cates that having a partner has a protective effect for
mental and physical health (Dehle et al., 2001),
providing a valuable source of companionship, as well
as emotional and financial support (Waite and Lehrer,
2003). However, a surprising feature of the present
between the subtypes and a drug use disorder. This is
inconsistent with community and clinical studies which
frequently cite a significant association between depres-
sion and drug use (Davis et al., 2005; Weissman et al.,
Some notes of caution, however, should temper the
above findings. Firstly, due to the structure of the
AUDADIS-IV, this study was curtailed to individuals
who reported a 2-week periodof depressed mood/loss of
interest in activities. This may limit the generalisability
of the findings to individuals in the general population
who do not endorse these questions. Secondly, the
NESARC utilised lay interviewer-administered struc-
tured interviews to determine mental health diagnoses.
Having said that, the interviews were conducted by
professional interviewers from the U.S. Bureau of the
Census and adhered to strict quality control procedures.
In addition, the reliability and validity of the diagnoses
have been established (Grant et al., 2003b, 2005). A
final limitation concerns our reliance on measures of
‘any anxiety disorder’, ‘any mood disorder’, and ‘any
personality disorder’ to determine psychiatric disorder
prevalence rates. Albeit a somewhat crude measure for
analyses involving common disorders, this metric
facilitated examination of less prevalent disorders and
interpretation of class structures (cf. Whitesell et al.,
Notwithstanding these limitations, this paper high-
lights the utility of latent class analysis in elucidating the
heterogeneity in depression in the general population.
Rather than a generic one size fits all approach, profiling
patterns of depressive symptomatology is a potentially
useful first step in informing tailored intervention and
treatment strategies (cf. Moss et al., 2007). The second
wave of the NESARC holds promise for examining
class transitions over time and contributing to the pre-
diction of outcome.
Role of funding source
Conflict of interest
All authors declare that they have no conflict of interests.
Aggen, S.H., Neale, M.C., Kendler, K.S., 2005. DSM criteria for
major depression: evaluating symptom patterns using latent-trait
item response models. Psychol. Med. 35, 475–487.
Akaike, H., 1974. A new look at the statistical model identification.
IEEE Trans. Automat. Contr. 19, 716–723.
Alonso, J., Angermeyer, M.C., Bernert, S., Bruffaerts, R., Brugha, T.S.,
Bryson, H., de Girolamo, G., Graaf, R., Demyttenaere, K., Gasquet,
I., Haro, J.M., Katz, S.J., Kessler, R.C., Kovess, V., Lépine, J.P.,
Ormel, J., Polidori, G., Russo, L.J., Vilagut, G., Almansa, J.,
Arbabzadeh-Bouchez, S., Autonell, J., Bernal, M., Buist-Bouwman,
M.A., Codony, M., Domingo-Salvany, Ferrer, M., Joo, S.S.,
Martínez-Alonso, M., Matschinger, H., Mazzi, F., Morgan, Z.,
Morosini, P., Palacín, C., Romera, B., Taub, N., Vollebergh, W.A.,
2004. 12-Month comorbidity patterns and associated factors in
Europe: results from the European Study of the Epidemiology of
Mental Disorders (ESEMeD) project. Acta Psychiatr. Scand., Suppl.
American Psychiatric Association, 2000. Diagnostic and Statistical
Manual of Mental Disorders, Text Revision, 4th Ed. American
Psychiatric Association, Washington.
Andreasen, N.C., Grove, W.M., Maurer, R., 1980. Cluster analysis and
the classification of depression. Br. J. Psychiatry 137, 256–265.
Barry, D., Pietrzak, R.H., Petry, N.M., 2008. Gender differences in
associations between body mass index and DSM-IV mood and
anxiety disorders: results from the National Epidemiologic Survey
Beals, J., Novins, D.K., Whitesell, N.R., Spicer, P., Mitchell, C.M.,
Manson, S.M., 2005. Prevalence of mental disorders and
utilization of mental health services in two American Indian
reservation populations: mental health disparities in a national
context. Am. J. Psychiatry 162, 1723–1732.
Bell, B., Chalklin, L., Mills, M., Browne, G., Steiner, M., Roberts, J.,
Gafni, A., Byrne, C., Wallik, D., Kraemer, J., Webb, M., Jamieson,
illness in adults in a Canadian primary care setting: high rates of
psychiatric illness in the offspring. J. Affect. Disord. 78, 73–80.
Benazzi, F., 2005. The relationship of major depressive disorder to
bipolar disorder: continuous or discontinuous? Curr. Psychiatry
Rep. 7, 462–470.
Bijl, R.V., Ravelli, A., van Zessen, G., 1998. Prevalence of psychiatric
disorder in the general population: results of the Netherlands
Mental Health Survey and Incidence Study (NEMESIS). Soc.
Psychiatry Psychiatr. Epidemiol. 33, 587–595.
Blashfield, R.K., Morey, L.C., 1979. The classification of depression
through cluster analysis. Compr. Psychiatry 20, 516–527.
Blazer, D., Woodbury, M., Hughes, D.C.,George, L.K., Manton, K.G.,
Bachar, J.R., Fowler, N., 1989. A statistical analysis of the
classification of depression in a mixed community and clinical
sample. J. Affect. Disord. 16, 11–20.
97N. Carragher et al. / Journal of Affective Disorders 113 (2009) 88–99
Blazer, D., Swartz, M., Woodbury, M., Manton, K.G., Hughes, D.,
George, L.K., 1988. Depressive symptoms and depressive
diagnoses in a community population: use of a new procedure
for analysis of psychiatric classification. Arch. Gen. Psychiatry 45,
Breslau, J., Aguilar-Gaxiola, S., Kendler, K.S., Su, M., Williams, D.,
Kessler, R.C., 2006. Specifying race-ethnic differences in risk of
psychiatric disorder in a USA national sample. Psychol. Med. 36,
Canino, G., Bravo, M., Ramirez, R., Febo, V.E., Rubio-Stipec, M.,
Fernandez, R.L., Hasin, D., 1999. The Spanish Alcohol Use
Disorder and Associated Disabilities Interview Schedule (AUDA-
DIS): reliability and concordance with clinical diagnoses in a
Hispanic population. J. Stud. Alcohol 60, 790–799.
Centers for Disease Control and Prevention, 2005. Web-based
Injury Statistics Query and Reporting System (WISQARS),
National Center for Injury Prevention and Control, CDC.
Retrieved 14 September 2007, from http://www.cdc.gov/ncipc/
Chatterji, S., Saunders, J.B., Vrasti, R., Grant, B.F., Hasin, D.S.,
Mager, D., 1997. Reliability of the Alcohol Use Disorder and
Associated Disabilities Interview Schedule-Alcohol/Drug Revised
(AUDADIS-ADR): an international comparison. Drug Alcohol
Depend. 47, 171–185.
Chen, L., Eaton, W.W., Gallo, J.J., Nestadt, G., 2000. Understanding
the heterogeneity of depression through the triad of symptoms,
course and risk factors: a longitudinal population-based study.
J. Affect. Disord. 59, 1–11.
Corruble, E., Ginestet, D., Guelfi, J.D., 1996. Comorbidity of
personality disorders and unipolar major depression: a review.
J. Affect. Disord. 37, 157–170.
Cox, B.J., Enns, M.W., Larsen, D.K., 2001. The continuity of
depression symptoms: use of cluster analysis for profile identifica-
tion in patient and student samples. J. Affect. Disord. 65, 67–73.
Crum, R.M.,Storr, C.L.,Chan, Y.F., 2005.Depressionsyndromeswith
risk of alcohol dependence in adulthood: a latent class analysis.
Drug Alcohol Depend. 79, 71–81.
Cuellar, A.K., Johnson, S.L., Winters, R., 2005. Distinctions between
bipolar and unipolar depression. Clin. Psychol. Rev. 25, 307–339.
Davidson, J., Woodbury, M.A., Pelton, S., Krishnan, R., 1988. A study
of depressive typologies using grade of membership analysis.
Psychol. Med. 18, 179–189.
Davis, L.L., Rush, J.A., Wisniewski, S.R., Rice, K., Cassano, P.,
Husain, M.M., Quitkin, F.M., McGrath, P.J., 2005. Substance use
disorder comorbidity in major depressive disorder: an exploratory
analysis of the Sequenced Treatment Alternatives to Relieve
Depression cohort. Compr. Psychiatry 46, 81–89.
Dawson, D.A., Grant, B.F., Li, T.K., 2007. Impact of age at first drink
on stress-reactive drinking. Alcohol. Clin. Exp. Res. 31, 69–77.
Dehle, C., Larsen, D., Landers, J.E., 2001. Social support in marriage.
Am. J. Fam. Ther. 29, 307–324.
Druss, B.G., Rosenheck, R.A., 1999. Patterns of health care costs
associated with depression and substance abuse in a national
sample. Psychiatr. Serv. 1999, 214–218.
Eaton, W.W., Dryman, A., Sorenson, A., McCutcheon, A., 1989.
DSM-IIImajordepressive disorder in the community:a latentclass
analysis of data from the NIMH Epidemiologic Catchment Area
Programme. Br. J. Psychiatry 155, 48–54.
Farabaugh, A., Mischoulon, D., Fava, M., Guyker, W., Alpert, J.,
2004. The overlap between personality disorders and major
depressive disorder (MDD). Ann. Clin. Psychiatry 16, 217–224.
Fergusson, D.M., Goodwin, R.D., Horwood, L.J., 2003. Major
depression and cigarette smoking: results of a 21-year longitudinal
study. Psychol. Med. 33, 1357–1367.
Gijsen, R., Hoeymans, N., Schellevis, F.G., Ruwaard, D., Satariano,
W.A., van den Bos, G.A., 2001. Causes and consequences of
comorbidity: a review. J. Clin. Epidemiol. 54, 661–674.
Goldberg, J.F., Harrow, M., Whiteside, J.E., 2001. Risk for bipolar
illness in patients initially hospitalized for unipolar depression.
Am. J. Psychiatry 158, 1265–1270.
Goldney, R.D., Fisher, L.J., Dal Grande, E., Taylor, A.W., 2004.
Subsyndromal depression: prevalence, use of health services and
Epidemiol. 39, 293–298.
Grant, B.F., Hasin, D.S., Stinson, F.S., Dawson, D.A., Chou, P.S.,
Ruan, W.J., Huang, B., 2005. Co-occurrence of 12-month mood
and anxiety disorders and personality disorders in the US: results
from the National Epidemiologic Survey on Alcohol and Related
Conditions. J. Psychiatr. Res. 39, 1–9.
Grant, B.F., Kaplan, K., Shepard, J., Moore, T., 2003a. Source and
Accuracy Statement for Wave 1 of the 2001–2003 National
Epidemiologic Survey on Alcohol and Related Conditions.
National Institute on Alcohol Abuse and Alcoholism, Bethesda.
Grant, B.F., Dawson, D.A., Stinson, F.S., Chou, P.S., Kay, W.,
Pickering, R., 2003b. The Alcohol Use Disorder and Associated
Disabilities Interview Schedule-IV (AUDADIS-IV): reliability of
alcohol consumption, tobacco use, family history of depression
and psychiatric diagnostic modules in a general population sample.
Drug Alcohol Depend. 71, 7–16.
Grant, B.F., Dawson, D.A., Hasin, D.S., 2001. The Alcohol Use
Disorder and Associated Disabilities Interview Schedule-DSM-IV
Version (AUDADIS-IV). National Institute on Alcohol Abuse and
Hagenaars, J., McCutcheon, A., 2002. Applied Latent Class Analysis
Models. Cambridge University Press, Cambridge.
Hasin, D.S., Goodwin, R.D., Stinson, F.S., Grant, B.F., 2005.
Epidemiology of major depressive disorder: results from the
National Epidemiologic Survey on Alcoholism and Related
Conditions. Arch. Gen. Psychiatry 62, 1097–1106.
Haslam, N., Beck, A.T., 1994. Subtyping major depression: a
taxometric analysis. J. Abnorm. Psychol. 103, 686–692.
Jorm, A.F., 2000. Does old age reduce the risk of anxiety and
depression? A review of epidemiological studies across the adult
life span. Psychol. Med. 30, 11–22.
Kendler, K.S., Gardener, C.O., Prescott, C.A., 1999. Clinical
characteristics of major depression that predict risk of depression
in relatives. Arch. Gen. Psychiatry 56, 322–327.
Kendler, K.S., Eaves, L.J., Walters, E.E., Neale, M.C., Heath, A.C.,
Kessler, R.C., 1996. The identification and validation of distinct
depressive syndromes in a population-based sample of female
twins. Arch. Gen. Psychiatry 53, 391–399.
Kuehner, C., 2003. Gender differences in unipolar depression: an
updateof epidemiological findingsandpossibleexplanations. Acta
Psychiatr. Scand. 108, 163–174.
Kovess-Masfety, V., Lecoutour, X., Delavelle, S., 2005. Mood
disorders and urban/rural settings: comparisons between two
French regions. Soc. Psychiatry Psychiatr. Epidemiol. 40,
Leskela, U., Rytsala, H., Komulainen, K., Melartin, T., Sokero, P.,
Lestela-Mielonen, P., Isometsa, E., 2006. The influence of ad-
versity and perceived social support on the outcome of major
98N. Carragher et al. / Journal of Affective Disorders 113 (2009) 88–99
depressive disorder in subjects with different levels of depressive
symptoms. Psychol. Med. 36, 779–788.
Little, T.D., Cunningham, W.A., Shahar, G., Widaman, K.F., 2002. To
parcel or not to parcel: exploring the question, weighing the merits.
Struct. Eq. Modeling 9, 151–173.
Lo, Y., Mendell, N., Rubin, D.B., 2001. Testing the number of
components in a normal mixture. Biometrika 88, 767–778.
Lubke, G.H., Muthén, B., 2005. Investigating population hetero-
geneity with factor mixture models. Psychol. Methods 10, 21–39.
Miech, R.A., Caspi, A., Moffitt, T.E., Wright, B.E., Silva, P.A., 1999.
Low socioeconomic status and mental disorders: a longitudinal
study of selection and causation during young adulthood. Am. J.
Sociol. 104, 1096–1131.
Moss, H.B., Chen, C.M., Yi, H.Y., 2007. Subtypes of alcohol
dependence in a nationally representative sample. Drug Alcohol
Depend. 91, 149–158.
Murray, C.J., Lopez, A.D., 1997. Alternative projections of mortality
and disability by cause 1990–2020: Global Burden of Disease
Study. Lancet 349, 1498–1504.
Muthén, B.O., 1989. Dichotomous factor analysis of symptom data.
Sociol. Methods Res. 18, 19–65.
Muthén, B., Asparouhov, T., 2006. Item response mixture modeling:
application to tobacco dependence criteria. Addict. Behav. 31,
Muthén, L.K., Muthén, B.O., 1998–2008. Mplus User's Guide,
(5th Ed). Muthén & Muthén, Los Angeles.
Myrick, H., Brady, K., 2003. Current review of the comorbidity of
affective, anxiety, and substance use disorders. Curr. Opin.
Psychiatry 16, 261–270.
National Institute on Alcohol Abuse and Alcoholism, 2006. Alcohol
Use and Alcohol Use Disorders in the United States: Main
Findings from the 2001–2002 National Epidemiological Survey
on Alcohol and Related Conditions (NESARC), U.S. Alcohol
Epidemiologic Data Reference Manual Volume 8, Number 1.
National Institute on Alcohol Abuse and Alcoholism, Bethesda.
Nylund, K.L., 2007. Latent transition analysis: modeling extensions
and an application to peer victimization. PhD Dissertation,
University of California, Los Angeles.
Probst, J.C., Laditka, S.B., Moore, C.G., Harun, N., Powell, M.P.,
Baxley, E.G., 2006. Rural–urban differences in depression pre-
valence: implications for family medicine. Fam. Med. 38, 653–660.
Ramaswamy, V., DeSarbo, W., Reibstein, D., Robinson, W., 1993. An
empirical pooling approach for estimating marketing mix elasti-
cities with PIMS data. Marketing Sci. 12, 103–124.
Schwartz, G., 1978. Estimating the dimensionof a model. Ann. Stat. 6,
Sclove, S.L., 1987. Application of model-selection criteria to some
problems in multivariate analysis. Psychometrika 52, 333–343.
Scotte, C.K., Maes, M., Cluydts, R., Cosyns, P., 1997. Cluster analytic
integration of quantitative and qualitative distinctions between
unipolar depressive subtypes. Psychiatry Res. 71, 181–195.
Sen, B., 1987. An analysis of the nature of depressive phenomena in
primary health care utilising multivariate statistical techniques.
Acta Psychiatr. Scand. 76, 28–32.
Shavers, V.L., 2007. Measurement of socioeconomic status in health
disparities research. J. Natl. Med. Assoc. 99, 1013–1023.
Sher, L., 2005. Alcohol use and suicide rates. Med. Hypotheses 65,
Simon, G.E., von Korff, M., 2006. Medical co-morbidity and validity
of DSM-IV depression criteria. Psychol. Med. 36, 27–36.
suicide mortality, 1970–1997. Am. J. Public Health 92, 1161–1167.
Slade, T., Andrews, G., 2005. Latent structure of depression in a com-
munity sample: a taxometric analysis. Psychol. Med. 35, 489–497.
Sullivan, P.F., Kessler, R.C., Kendler, K.S., 1998. Latent class analysis
of lifetime depressive symptoms in the National Comorbidity
Survey. Am. J. Psychiatry 155, 1398–1406.
Sullivan, P.F., Prescott, C.A., Kendler, K.S., 2002. The subtypes of
major depression in a twin registry. J. Affect. Disord. 68, 273–284.
Swendsen, J.D., Merikangas, K.R., 2000. The comorbidity of
depression and substance use disorders. Clin. Psychol. Rev. 20,
Waite, L.J., Lehrer, E.L., 2003. The benefits from marriage and
religion in the United States: a comparative analysis. Popul. Dev.
Rev. 29, 255–275.
Weissman, M.M., Wickramaratne, P., Nomura, Y., Warner, V., Verdeli,
H., Pilowsky, D.J., Grillon, C., Bruder, G., 2005. Families at high
and low risk for depression: a 3-generation study. Arch. Gen.
Psychiatry 62, 29–36.
Weissman, M.M., Bland, R.C., Canino, G.J., Faravelli, C., Greenwald,
S., Hwu, H.G., Joyce, P.R., Karam, E.G., Lee, C.K., Lellouch, J.,
Lepine, J.P., Newman, S.C., Rubio-Stipec, M., Wells, J.E.,
Wickramaratne, P.J., Wittchen, H., Yeh, E.K., 1996. Cross-national
epidemiology of major depression and bipolar disorder. JAMA
Whitesell, N.R., Beas, J., Mitchell, C.M., Novins, D.K., Spicer, P.,
Manson, S.M., 2006. Latent class analysis of substance use:
comparison of two American Indian reservation populations and a
national sample. J. Stud. Alcohol 67, 32–43.
Wittchen, H.U., Kessler, R.C., Pfister, H., Lieb, M., 2000. Why do
people with anxiety disorders become depressed? A prospective-
longitudinal community study. Acta Psychiatr. Scand., Suppl. 406,
World Health Organization, 1990. Composite International Diagnostic
Interview. World Health Organization, Geneva.
99 N. Carragher et al. / Journal of Affective Disorders 113 (2009) 88–99