Download Continuous, categorical and mixture models of DSM

RESEARCH REPORT
doi:10.1111/j.1360-0443.2010.02951.x
Continuous, categorical and mixture models of
DSM-IV alcohol and cannabis use disorders in
the Australian community
add_2951
1246..1253
Andrew J. Baillie1 & Maree Teesson2
Psychology Department, Macquarie University, Sydney, NSW, Australia1 and National Drug and Alcohol Research Centre, University of New South Wales,
Sydney, NSW, Australia2
ABSTRACT
Aims To apply item response mixture modelling (IRMM) to investigate the viability of the dimensional and categorical
approaches to conceptualizing alcohol and cannabis use disorders. Design A cross-sectional survey assessing substance use and DSM-IV substance use disorders. Setting and participants A household survey of a nationally representative sample of 10 641 Australia adults (aged 18 years or older). Measurements Trained survey interviewers
administered a structured interview based on the Composite International Diagnostic Interview (CIDI). Findings Of
the 10 641 Australian adults interviewed, 7746 had drunk alcohol in the past 12 months and 722 had used cannabis.
There was no improvement in fit for categorical latent class nor mixture models combining continuous and categorical
parameters compared to continuous factor analysis models. The results indicated that both alcohol and cannabis
problems can be considered as dimensional, with those with the disorder arrayed along a dimension of severity.
Conclusions A single factor accounts for more variance in the DSM-IV alcohol and cannabis use criteria than latent
class or mixture models, so the disorders can be explained most effectively by a dimensional score.
Keywords
Alcohol, cannabis, diagnosis, DSM-IV.
Correspondence to: Andrew Baillie, Macquarie University, Sydney, NSW 2109, Australia. E-mail: [email protected]
Submitted 2 June 2009; initial review completed 13 July 2009; final version accepted 6 January 2010
INTRODUCTION
Alcohol is the most widely used licit drug and cannabis is
the most widely used illicit drug in developed countries
[1,2], and each has a significant impact on health in the
general population [3–5]. Both alcohol and cannabis
users in the general population frequently endorse criteria for the most common drug phenotypes, abuse and
dependence [3–10]. Risks associated with alcohol use
have long been known and the risks of cannabis use are
being recognized increasingly. The most common description of the phenotype of these consequences is DSM-IV
alcohol and cannabis use disorders, yet it is becoming
increasing clear that the validity and utility of the abuse
and dependence syndrome (and its implementation in
widely used diagnostic schemes) requires attention [11].
The dependence syndrome was first formulated by
Edwards & Gross [12] and Edwards et al. [13], as a
number of psychological and physiological factors asso-
ciated with diminished control over alcohol use. In a
later publication Edwards [14] referred to the ‘bi-axial
concept’, where dependence as described above constitutes one axis of the syndrome and alcohol-related problems forms the other. While the bi-axial concept has
clinical appeal, evidence from large population and clinical samples has called into question this distinction and
the use of categorical descriptions generally providing
evidence for the conceptualization of problems along a
dimension or continuum of severity. In the current paper
we apply item response mixture modelling (IRMM) to
investigate the viability of the dimensional and categorical approaches to conceptualizing alcohol and cannabis
problems.
The aims of this paper are to compare continuous,
categorical and mixture models for the structure of the
DSM-IV [15] criteria in a large population survey using
item response mixture modelling. This is conducted on
the most commonly used licit and illicit drugs, alcohol
© 2010 The Authors. Journal compilation © 2010 Society for the Study of Addiction
Addiction, 105, 1246–1253
Mixture models
and cannabis. Item response mixture modelling is a new
hybrid latent variable model which combines the features
of categorical and dimensional analyses used in the conventional techniques of latent class and factor analysis.
It has been argued that hybrid models are more suitable
for understanding substance use dependence and fit the
data better than conventional models of factor analysis
and latent class analysis (LCA) [16]. Muthén et al.
[16–18] have demonstrated its superiority in nicotine
dependence and alcohol use disorders, but it is untested
as yet in the most commonly used illicit drugs.
If superior, these hybrid models may be useful in
improving our conceptualization of alcohol and cannabis
problems. The focus of this paper is therefore to explore
whether categorical or latent class models, dimensional
or factor analysis/item response theory models or the
new hybrid models are superior in describing problems
associated with alcohol and cannabis use. Both the relationship between the abuse and dependence criteria
and the appropriateness of the criteria to alcohol and
cannabis use disorders are examined with each measurement model. The DSM-IV specifies 11 criteria for substance use disorders, equally applicable to all classes
of psychoactive substances including alcohol, cocaine,
opiates, cannabis, sedatives, stimulants and hallucinogens. Dependence is measured by seven criteria, at least
three of which must be endorsed for a diagnosis to be
reached. Abuse is measured by four additional criteria,
and a diagnosis is made if at least one criterion is
endorsed (and a diagnosis of dependence is absent).
Criteria for each diagnostic outcome are assumed to
have equal weighting.
The application of statistical modelling to substance
use criteria is increasing. Most investigations of this type
have focused upon clinical populations (with associated
range restriction) and employed a range of factor analytical techniques, yielding inconsistent results. In the
alcohol literature some studies have found evidence
for two separate, although related factors [19,20], while
others have identified single dimensions [21]. The
two studies to assess the current DSM-IV criteria in
population-based samples supported a one-factor solution [22,23]. The position of the items along this latent
dimension differed considerably, with abuse indicating
severity in the Australian analysis [22] and abuse items
being spread more widely across the continuum in the US
analysis [23]
In terms of cannabis, some studies have found
evidence for the coherence of cannabis dependence
[10,24–27], while others have identified multiple dimensions [28] For example, Kosten et al. [28] suggested a
three-factor model for disorders of cannabis use, based
on profiles of endorsement of DSM-III-R criteria for
substance use disorders in a clinical population.
1247
The majority of these studies have restricted their
examination to the structure of the cannabis dependence criteria only. Teesson et al. [3], Lynskey et al. [4]
and Compton et al. [5] examined the structure of both
DSM-IV cannabis abuse and dependence in population
samples and found that the one-factor solution was
the most parsimonious. Langenbucher et al. [29] found
similar results in a mixed clinical and general population sample.
The WHO cross-national study by Nelson et al. [27]
tested the seven DSM-IV dependence and four abuse criteria for alcohol and cannabis using confirmatory factor
analysis (CFA) using community and treatment centre
samples. With this sample they found a two-factor solution no better than the one-factor solution; but when they
‘trimmed’ the data of extreme respondents (those who
responded ‘no’ to all criteria or ‘yes’ to 10 or 11 criteria)
they found the two-factor solution to be superior. This
result implies that symptoms behave differently at different levels of severity and supports the exploration of
hybrid models with both dimensional and categorical
conceptualizations.
The dimensionality of the substance use criteria
is not entirely clear. Our work to date has supported
unidimensionality; however, it is recognized that categories are useful in the clinical response to disorders [19].
In 2006, Muthén [17] raised the possibility that both
dimensional and continuous models may fit diagnostic
criteria. He proposed the application of new hybrid
latent variable models for phenotypical analyses. He
tested these models on tobacco dependence criteria
and alcohol use disorder criteria in males and found
that the hybrid latent models fitted the data more
effectively than conventional models of factor analysis
(IRT) and LCA. The current paper extends our existing
work with Australian general population data [3,22]
to examine these hybrid models. Based on previous
work [16–18] we plan to compare the fit of one- and
two-factor models which assess substance use disorders
as a single dimension or as separate dependence and
abuse dimensions, as two, three, four, five or six latent
classes in which individuals are conceptualized as
forming separate classes or categories, and hybrid
models with both categorical and dimensional entities.
Following on from Nelson et al. [27], who removed from
analysis those individuals who endorsed no criteria or
all criteria from some analyses, we examine one- and
two-factor models in conjuction with a zero class
accounting for those who endorsed no criteria. By comparing the extent to which these models explain the
observed data on the same fit indices, this paper provides
an empirical test of the whether the DSM-IV criteria
behave as one or more dimensions, categories or a
mixture of the two.
© 2010 The Authors. Journal compilation © 2010 Society for the Study of Addiction
Addiction, 105, 1246–1253
1248
Andrew J. Baillie & Maree Teesson
We use data from the Australian National Survey of
Mental Health and Wellbeing (NSMHWB) [30], a study
of a large and representative sample of the general population. A detailed description of this study is provided
elsewhere [30].
METHODS
The NSMHWB was a multi-stage sample of private dwellings across Australia. Dwellings were selected using
random stratified multi-stage area sampling, so that each
person had a known chance of participation. One person
aged at least 18 years was selected randomly (next
birthday) from each dwelling and asked to participate. Approximately 13 600 private dwellings were
approached, with a final sample size of 10 641 people
giving a response rate of 78%.
Mental disorders were assessed by a modified version
of the CIDI [30], which yielded diagnoses of both ICD-10
and DSM-IV disorders. The CIDI has been used in a range
of epidemiological studies, and has been shown to be a
reliable and valid survey instrument; details of this have
been provided elsewhere [30] based upon studies using
an earlier version of the CIDI. The substance use disorders section has changed very little between this
version and that used in the NSMHWB. Interviews were
conducted face to face with trained interviewers and
questioning was restricted to symptoms in the last 12
months.
Alcohol abuse and dependence were assessed in all
people who had consumed at least 12 alcoholic drinks
in the past 12 months (n = 7746). Cannabis abuse and
dependence criteria were assessed in all people who had
used cannabis more than five times in the past 12 months
(n = 722).
Statistical methods
Although the NSMHWB data have been obtained by
complex sampling procedures, the factor analysis
methods used in this paper assumed simple random sampling. Muthén et al. [19] argue that this simplified
method is acceptable because multivariate methods
are generally less sensitive than univariate methods to
complex sampling features.
All analyses in this paper were carried out using
maximum-likelihood estimation in the Mplus program
[31,32]. Three models are tested on the data; one and two
CFA, two to six LCA and the new hybrid item response
or factor mixture model. Muthén & Asparouhov [16]
provide a description of the three models. Maximum likelihood estimation was used to allow comparison of the fit
of different models by comparison of the –2log-likelihood
estimates.
RESULTS
Alcohol use disorders
Indices of the fit of the different models proposed
to underlie DSM-IV alcohol use disorders are shown
in Table 1. CFA/item response theory models using
maximum likelihood estimation are reported first. Two
separate abuse and dependence factors gave a significantly better fit (deviance c2 = 5.35, df = 1, P < 0.05)
than a single substance use factor. The correlation
between these two latent factors was very high (0.954).
Latent class models were examined next. As can be
seen in Table 1, the best-fitting LCA employed three latent
classes [log-likelihood = 11 834.084, parameters = 35,
Bayesian information criteria (BIC) = 23 981.611] but
higher log-likelihood and BIC indicate a worse fit while
using a greater number of parameters than the CFA
models. Figure 1 shows the profile of criteria for each of
the three latent classes in the best-fitting LCA model. The
estimated proportion falling into each of these three
classes were class 1, 10.64% (n = 824); class 2, 2.07%
(n = 160); and class 3, 87.29% (n = 6762). In comparison, the existing DSM-IV algorithm would assign a
diagnosis of alcohol dependence to 5.6% of drinkers and
alcohol abuse to 2.6% of drinkers [19]. Class 1 can be
taken to represent dependent drinkers with a lower
threshold for diagnosis than expressed in DSM-IV and
class 2 reflects a small proportion of drinkers who are
likely to experience criteria LARGER and CUTDOWN and
TOLERANCE to a lesser extent, but not other criteria.
Class 3 represents a zero class. The lines representing the
profiles for classes 1 and 2 might be interpreted as parallel
or indicating that these two profiles may differ on severity
alone. Notably, absent is a latent class with high probability of endorsement for the abuse criteria. Consistent with
our earlier modelling using item response theory [21],
criteria GIVE-UP and TIME SPENT discriminated between
classes 1 and 2, as did WITHDRAWAL, CONTINUE and,
to a lesser extent, the abuse criteria.
The first item response or factor mixture model (model
8 in Table 1) examines a single factor with a zero class.
This model proposes that (i) there are people who drink
alcohol but report none of the 11 criteria (the zero class)
and (ii) that the patterns of criteria reported by the
remainder of drinkers are accounted for by a single
dimension.
Model 9 (see Table 1) estimates a zero class and two
related but separable dependence and abuse factors. This
model follows from Nelson et al. [27], who concluded that
there were separate abuse and dependence dimensions
when they removed those who endorsed all or none of the
11 criteria. Higher –2log-likelihood and BIC indicate a
worse fit than the two-factor CFA/IRT model and worse
than the three-class LCA.
© 2010 The Authors. Journal compilation © 2010 Society for the Study of Addiction
Addiction, 105, 1246–1253
4855.682
4875.443
23
26
The procedure for fitting models to the 11 DSM-IV cannabis use disorder criteria was the same as above. As can
be seen in Table 1, a single dimension underlying all 11
DSM-IV cannabis use disorder criteria was not a significantly worse fit to the data based on comparison of
–2log-likelihoods (deviance c2 = 0.010, d f = 1, P = ns)
and showed lower BIC than separate abuse and dependence factors. Of the latent class models fitted, model 3
with two latent classes had the lowest BIC. However, this
model showed a poorer fit to the data than the best CFA
model a single factor.
Figure 2 shows the probability of endorsing each of
the criteria for the two latent classes in the best-fitting
latent class model. The second class, which accounted for
547 (75.76%) of the 722 cannabis users, approximates a
zero class with an approximately 20% chance of endorsing the CUTDOWN criteria. In contrast, the first latent
class containing 175 (24.24%) of the 722 cannabis users
shows much greater endorsement of the TOLERANCE,
WITHDRAWAL, LARGER and CUTDOWN criteria.
Application of the DSM-IV algorithm gives a prevalence
of cannabis dependence as 20.9% [standard error
(SE) = 2.2], so this class can be thought of as reflecting a
dependence-like syndrome with a lower threshold.
Adding a zero class to a single dimension underlying
the 11 criteria (model 8) showed no significant improvement in –2log-likelihood [deviance c2 = 1.07, df = 1,
P = not significant (NS)] and an increase in the BIC, indicating a worsening of fit to the data compared to the
single-factor CFA model. As with drinkers above, there
was no improvement in fit for a two-factor model with a
zero class.
-2352.148
-2352.155
DISCUSSION
BIC: Bayesian information criteria.
23
26
-11 814.226
-12 088.402
1249
Cannabis use disorders
23 834.416
24 409.631
4910.052
4923.976
4978.043
5018.716
5086.560
23
35
47
59
71
-2379.333
-2346.802
-2334.334
-2315.188
-2309.618
23
35
47
59
71
-12 200.049
-11 834.094
-11 817.335
-11 799.682
-11 760.249
24 606.062
23 981.611
24 055.551
24 127.705
24 156.297
4851.244
4857.807
22
23
-2353.220
-2353.210
23 835.681
23 833.936
22
23
-11 819.336
-11 813.986
Continuous confirmatory factor analysis/item response theory models
1. One substance use disorder factor
2. Two separate abuse and dependence factors
Categorical latent class models
3. Two latent classes
4. Three latent classes
5. Four latent classes
6. Five latent classes
7. Six latent classes
Item response or factor mixture models
8. One factor with a zero class
9. Two factors with a zero class
Log-likelihood
Log-likelihood
Model
Table 1 Model fit results.
Alcohol (n = 7746)
No. of free
parameters
BIC
Cannabis (n = 722)
No. of free
parameters
BIC
Mixture models
The purpose of these analyses was to compare explicitly
categorical, continuous and new mixture models of
DSM-IV alcohol and cannabis criteria in a large representative epidemiological sample. Single- or two-factor
(highly correlated) models were found to be the bestfitting models for both alcohol and cannabis use disorders. While two-factor models showed a better fit, the
correlation between the two factors was so high that a
single-factor model is the most parsimonious solution.
Hybrid item response mixture models incorporating both
dimensional and categorical conceptions of substance
use disorders do not provide substantial gains in fit over
straight categorical or straight dimensional conceptions
of alcohol and cannabis problems.
The current results are somewhat inconsistent with
previous analyses undertaken on tobacco and alcohol use
disorders [16–18], which report that these item response
© 2010 The Authors. Journal compilation © 2010 Society for the Study of Addiction
Addiction, 105, 1246–1253
1250
Andrew J. Baillie & Maree Teesson
Criterion Profiles from Latent Class Models of Drinkers
1.2
Criterion Probability
1
0.8
Latent Class 1
Latent Class 2
Latent Class 3
0.6
0.4
0.2
Social
5
Legal
TImeSp
4
Hazard
Cutdown
3
Major Role
Larger
2
Coninue
Withdrawal
1
Give up
Tolerance
0
6
7
8
9
10
11
DSM-IV Criteria
Figure 1 Criterion profiles for a three latent class model of drinkers
Criterion Profiles from Latent Class Models of Cannabis Users
0.9
0.8
Criterion Probability
0.7
0.6
0.5
LC(1)
LC(2)
0.4
0.3
0.2
0.1
L
SO
C
IA
_
AL
G
AZ
H
LE
AR
D
LE
M
AJ
R
U
TN
N
C
IV
G
E
EU
P
P
ES
TI
M
N
W
TD
C
U
G
ER
R
LA
D
TH
W
TL
R
R
W
N
C
L
E
0
DSM-IV Criteria
Figure 2 Criterion profile for a two latent class model of cannabis users
mixture models give a richer account of obtained data
for tobacco dependence and alcohol use disorders. Their
results suggest that these new methods, which allow
joint testing of categorical and continuous models of the
underlying disorder, are a closer match to the theoretical
descriptions of substance dependence. They argue that
the hybrid models may provide a richer description of the
disorders. The current study did not find any substantial
gains in fit over straight dimensional conceptualizations.
The analyses reported here are consistent with previous analyses we have reported on this data set [22]. In
this analysis of the 11 DSM-IV alcohol use disorder criteria we found that a two-factor model gave the best fit to
the data. The differences between the one- and two-factor
© 2010 The Authors. Journal compilation © 2010 Society for the Study of Addiction
Addiction, 105, 1246–1253
Mixture models
models were small in both analyses and the correlation
between the two factors in a two-factor model were very
high (0.95). Thus, the one-factor model represents the
most parsimonious solution.
A model containing two highly correlated factors
might seem to support the notion that substance abuse
represents the social and role impairment that develops
before symptoms of neuroadaptation and dependence.
However, in previous analyses we have shown that the
abuse criteria generally are markers of a more severe
state of substance use disorder. We would argue that the
finding that abuse criteria indicate a more severe state,
and that there is only a relatively small number of people
who meet criteria for abuse and not dependence, means
that the abuse diagnosis is a residual [33]. These results
are contrary to findings reported by Langenbucher &
Chung [34], who argue that the abuse criteria reflect an
early stage in the development of substance use disorders
in a clinical sample; however, it may be that there is something about the abuse criteria that brings people into
contact with treatment services, so that clinical samples
give different results. Thus it is critical to examine the
substance use disorder construct in population samples
to reduce such potentially confounding effects.
The original formulation of the dependence syndrome
[12,13] was as a number of psychological and physiological factors associated with diminished control over
alcohol use. Later, this was conceptualized further as
‘bi-axial’ [14], where dependence as described above constitutes one axis of the syndrome and alcohol-related
problems formed the other, and this has influenced conceptualizations in DSM-IV. While these concepts have
clinical and intuitive appeal, their examination using
IRMM does not support such categorical conceptualizations in the current paper. The implications of this paper
for DSM-V is that substance use disorders, as experienced
in the general community, are better conceptualized as
unidimensional, with criteria indicating severity. Importantly, our findings are consistent across both alcohol and
cannabis use disorders.
A single factor performs best in accounting for variance in the DSM-IV alcohol and cannabis use criteria.
This may mean that a single dimension is sufficient to
describe the most common presentations with these
disorders in the community. An alternative is that the
DSM-IV criteria do not encompass all the essential
features of substance use disorders. The DSM-IV criteria
might represent a premature narrowing of the construct.
Some work on the latent structure of alcohol problems by
Chick [35] found support for more than one factor in a
set of 21 symptoms based on the Edwards & Gross [12]
dependence syndrome assessed in a clinical sample of
men attending an alcoholism treatment unit. It may be
that more than one factor emerges from clinical samples,
1251
because most criteria are met leading to ceiling effects
and a restriction in the covariation of symptoms. Future
research should examine the possibility that multiple-factor models of substance use disorders might be
accommodated under a single higher-order factor.
Having demonstrated superior fit and parsimony for a
single latent variable to account for DSM-IV alcohol and
cannabis use disorders, further research could examine
the incremental validity [36] of assessments in addition
to a severity score based on the current criteria. That is,
other constructs associated with additional clinical
information over and above those are accounted for by a
severity score based on the DSM-IV criteria. It may also be
instructive to examine diagnostic orphans, those whose
patterns of symptoms are at odds with the prevailing
single-factor model.
This paper relies upon a structured diagnostic interview to derive symptoms and diagnoses. Although this is
common practice in psychiatric epidemiology, it is likely
that such methods do not capture the breadth of each
concept under study. For example, the DSM conveys the
notion that for the diagnosis to be given there must be
significant clinical impairment or distress. This may not
be captured well in structured diagnostic interviews.
However, the current findings are consistent with previous studies examining the performance of DSM criteria
in clinical samples [27,29,37].
The hybrid item-response/factor-mixture models used
in the present study following the recommendations of
Muthén [16,17] allow for the comparison of continuous,
categorical and combined or ‘mixture’ models of substance use disorders and psychopathology more generally
[38]. In this paper these new models do not provide a
better account of the symptoms of substance use disorders in the community than earlier continuous models.
However, it is important to continue to refine further the
diagnostic criteria and models of these criteria to define
phenotypes more tightly for genetic analyses and to
examine treatment responses more closely.
Declarations of interest
None.
Acknowledgement
The NSMHWB was designed, developed and conducted
with funding from the Mental Health Branch of the
Australian Commonwealth Department of Health and
Family Services. Development and testing of the computerized survey instrument was undertaken by Gavin
Andrews, Lorna Peters and other staff at the Clinical
Research Unit for Anxiety Disorders, and the WHO Collaborating Centre in Mental Health at St Vincent’s Hospital, Sydney. The design of the survey was overseen and
© 2010 The Authors. Journal compilation © 2010 Society for the Study of Addiction
Addiction, 105, 1246–1253
1252
Andrew J. Baillie & Maree Teesson
approved by the Technical Advisory Group comprising:
Professors A. Scott Henderson (Chair), Gavin Andrews,
Wayne Hall, Helen Herman, Assen Jablensky and Bob
Kosky. Survey fieldwork and implementation and the
enumeration, compilation and initial analyses of the data
were undertaken by the Australian Bureau of Statistics.
References
1. World Health Organisation (WHO). Global Status Report on
Alcohol 2004, 2nd edn. Geneva, Switzerland: WHO; 2005.
2. Hall W., Johnston L., Donnelly N. Epidemiology of cannabis
use and its consequences. In: Kalant H., Corrigall W. A., Hall
W., Smart R., editors. The Health Effects of Cannabis, Toronto,
ON: Centre for Addiction and Mental Health; 1999, p.
69–125.
3. Teesson M., Lynskey M., Manor B., Baillie A. The structure
of DSM-IV cannabis use disorders in the community. Drug
Alcohol Depend 2002; 68: 255–62.
4. Lynskey M. T., Agrawal A. Psychometric properties of DSM
assessments of illicit drug abuse and dependence: results
from the National Epidemiologic Survey on Alcohol and
Related Conditions (NESARC). Psychol Med 2007; 9: 1345–
55.
5. Compton W. M., Saha T. D., Conway K. P., Grant B. F. The
role of cannabis use within a dimensional approach to cannabis use disorders. Drug Alcohol Depend 2009; 100: 221–7.
6. Anthony J. C., Warner L. A., Kessler R. C. Comparative epidemiology of dependence on tobacco, alcohol, controlled
substances, and inhalants: basic findings from the National
Comorbidity Survey. Exp Clin Psychopharmacol 1994; 2:
244–68.
7. Compton W. M., Grant B. F., Colliver J. D., Glantz M. D.,
Stinson F. S. Prevalence of marijuana use disorders in the
United States: 1991–1992 and 2001–2002. JAMA 2004;
291: 2114–21.
8. Stinson F. S., Grant B. F., Dawson D. A., Ruan W. J., Huang
B., Saha T. Comorbidity between DSM-IV alcohol and specific drug use disorders in the United States: results from the
National Epidemiologic Survey on Alcohol and Related Conditions. Drug Alcohol Depend 2005; 80: 105–16.
9. Warner L. A., Kessler R. C., Hughes M., Anthony J. C.,
Nelson C. B. Prevalence and correlates of drug use and
dependence in the United States. Results from the National
Comorbidity Survey. Arch Gen Psychiatry 1995; 52: 219–
29.
10. Swift W., Hall W., Teesson M. Characteristics of DSM-IV and
ICD-10 cannabis dependence among Australian adults:
results from the National Survey of Mental Health and Wellbeing. Drug Alcohol Depend 2001; 63: 147–53.
11. Hasin D. S., Liu X., Alderson D., Grant B. F. DSM-IV alcohol
dependence: a categorical or dimensional phenotype?
Psychol Med 2006; 36: 1695–705.
12. Edwards G., Gross M. Alcohol dependence: provisional
description of a clinical syndrome. BMJ 1976; 1: 1058–61.
13. Edwards G., Arif A., Hodgson R. Nomenclature and classification of drug and alcohol related problems: a WHO memorandum. Bull World Health Organ 1981; 50: 225–42.
14. Edwards G. The alcohol dependence syndrome: a concept as
stimulus to enquiry. Br J Addict 1986; 81: 171–83.
15. American Psychiatric Association (APA). Diagnostic and
Statistical Manual of Mental Disorders DSM-IV, 4th edn.
Washington, DC: APA; 1994.
16. Muthén B., Asparouhov T. Item response mixture modeling:
application to tobacco dependence criteria. Addict Behav
2006; 31: 1050–66.
17. Muthén B. Should substance use disorders be considered as
categorical or dimensional? Addiction 2006; 101: 6–16.
18. Kuo P. H., Aggen S. H., Prescott C. A., Kendler K. S., Neale M.
C. Using a factor mixture modeling approach in alcohol
dependence in a general population sample. Drug Alcohol
Depend 2008; 98: 105–14.
19. Muthén B. O., Grant B. F., Hasin D. The dimensionality of
alcohol abuse and dependence: factor analysis of DSM-III-R
and proposed DSM-IV criteria in the 1988 National Health
Interview Survey. Addiction 1993; 88: 1079–90.
20. Harford T. C., Muthén B. O. The dimensionality of alcohol
abuse and dependence: a multivariate analysis of DSM-IV
symptom items in the National Longitudinal Survey of
Youth. J Stud Alcohol 2001; 62: 150–7.
21. Hasin D. S., Muthén B., Wisnicki K. S., Grant B. Validity of
the bi-axial dependence concept: a test in the US general
population. Addiction 1994; 89: 573–9.
22. Proudfoot H., Baillie A. J., Teesson M. The structure of
alcohol dependence in the community. Drug Alcohol Depend
2006; 81: 21–6.
23. Saha T., Chou S. P., Grant B. F. Toward an alcohol use disorder continuum using item response theory: results from the
National Epidemiologic Survey on Alcohol and Related Conditions. Psychol Med 2006; 36: 931–41.
24. Hasin D. S., Grant B. F., Harford T. C., Endicott J. The drug
dependence syndrome and related disabilities. Br J Addict
1998; 83: 45–55.
25. Morgenstern J., Langenbucher J. W., Labouvie E. W.
The generalizability of the dependence syndrome across
substances: an examination of some properties of the
proposed DSM-IV dependence criteria. Addiction 1994; 89:
1105–13.
26. Feingold A., Rounsaville B. J. Construct validity of the
dependence syndrome as measure by DSM-IV for different
psychoactive substances. Addiction 1995; 90: 1661–9.
27. Nelson C. B., Rehm J., Ustin T. B., Grant B., Chatterji S.
Factor structures for DSM-IV substance disorder criteria
endorsed by alcohol, cannabis, cocaine and opiate users:
results from the WHO reliability and validity study. Addiction 1999; 94: 843–55.
28. Kosten T. R., Rounsaville B. J., Babor T. F., Spitzer R. L.,
Williams J. B. Substance-use disorders in DSM-III-R. Evidence for the dependence syndrome across different psychoactive substances. Br J Psychiatry 1987; 151: 834–43.
29. Langenbucher J., Martin C. S., Labouvie E., Sanjuan P. M.,
Bavly L., Pollock N. K. Toward the DSM-V: the withdrawalgate model versus the DSM-IV in the diagnosis of alcohol
abuse and dependence. J Consult Clin Psychol 2000; 68:
799–809.
30. Teesson M., Hall W., Lynskey M., Degenhardt L. Alcoholand drug-use disorders in Australia: implications of the
National Survey of Mental Health and Wellbeing. Aust NZ J
Psychiatry 2000; 3: 206–13.
31. Muthén B. O., Muthén L. K. Mplus: Statistical Analysis with
Latent Variables. User’s Guide (V2.01). Los Angeles, CA:
Muthén & Muthén; 2001.
32. Muthén B., Muthén L. MPlus. Available at http://www.
statmodel.com (accessed 11 March 2010). Archived at
http://www.webcitation.org.5o9LJhpBS (11 March 2010).
33. Hasin D. Classification of alcohol use disorders. Alcohol Res
Health 2003; 27: 5–17.
© 2010 The Authors. Journal compilation © 2010 Society for the Study of Addiction
Addiction, 105, 1246–1253
Mixture models
34. Langenbucher J. W., Chung T. Onset and staging of DSM-IV
alcohol dependence using mean age and survival-hazard
methods. J Abnorm Psychol 1995; 104: 346–54.
35. Chick J. Is there a unidimensional alcohol dependence
syndrome? Br J Addict 1980; 75: 265–80.
36. Garb H. N. Incremental validity and the assessment of psychopathology in adults. Psychol Assess 2003; 15: 508–20.
37. Wu L. T., Pan J. J., Blazer D., Tai B., Stitzer M. L., Brooner
1253
R. K. et al. An item response theory modeling of alcohol and
marijuana dependence: a national drug abuse treatment
clinical trials network study. J Stud Alcohol Drugs 2009; 70:
414–42.
38. Krueger R. F., Markon K. E. Reinterpreting comorbidity:
a model-based approach to understanding and classifying psychopathology. Annu Rev Clin Psychol 2006; 2: 111–
33.
© 2010 The Authors. Journal compilation © 2010 Society for the Study of Addiction
Addiction, 105, 1246–1253
This document is a scanned copy of a printed document. No warranty is given about the accuracy of the copy.
Users should refer to the original published version of the material.