Defining and assessing tobacco dependence Running head: DEFINING AND ASSESSING TOBACCO DEPENDENCE Assessing tobacco dependence: A guide to measure evaluation and selection Megan E. Piper, Danielle E. McCarthy, Timothy B. Baker Megan E. Piper, M.A., Center for Tobacco Research and Intervention, University of Wisconsin Medical School and Department of Psychology, University of Wisconsin-Madison, Madison, WI Danielle E. McCarthy, M.S., Center for Tobacco Research and Intervention, University of Wisconsin Medical School and Department of Psychology, University of Wisconsin-Madison, Madison, WI Timothy B. Baker, Ph.D., Center for Tobacco Research and Intervention, University of Wisconsin Medical School and Department of Psychology, University of Wisconsin-Madison, Madison, WI Corresponding Author: 1
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Defining and assessing tobacco dependence
Running head: DEFINING AND ASSESSING TOBACCO DEPENDENCE
Assessing tobacco dependence:
A guide to measure evaluation and selection
Megan E. Piper, Danielle E. McCarthy, Timothy B. Baker
Megan E. Piper, M.A., Center for Tobacco Research and Intervention, University of Wisconsin Medical School and Department of Psychology, University of Wisconsin-Madison, Madison, WI
Danielle E. McCarthy, M.S., Center for Tobacco Research and Intervention, University of Wisconsin Medical School and Department of Psychology, University of Wisconsin-Madison, Madison, WI
Timothy B. Baker, Ph.D., Center for Tobacco Research and Intervention, University of Wisconsin Medical School and Department of Psychology, University of Wisconsin-Madison, Madison, WI
Corresponding Author:
Megan E. Piper, Center for Tobacco Research and Intervention, 1930 Monroe St., Suite 200, Madison, WI, 53711. Telephone: (608) 265-5472. Fax: (608) 265-3102. Email: [email protected]
This research was conducted at the University of Wisconsin, Madison.
1
Defining and assessing tobacco dependence
Abstract
Tobacco dependence has become a key construct in the field of tobacco research.
This paper describes a construct validation approach to dependence assessment and
describes key conceptual and psychometric criteria on which to evaluate putative
measures of dependence. The authors then examine the standing of five current
dependence scales: the Diagnostic and Statistical Manual, 4th Edition and the Fagerström
Test of Nicotine Dependence, as well as new measures, the Cigarette Dependence Scale,
the Nicotine Dependence Syndrome Scale and the Wisconsin Inventory of Smoking
Dependence Motives, on these critical dimensions. Recommendations for use of each
measure for varying purposes are made.
2
Defining and assessing tobacco dependence
Defining and assessing tobacco dependence: A critique and primer
Introduction
Many tobacco researchers and clinicians are interested in the concept of tobacco
dependence – a hypothetical construct invoked to explain smoking relapse (e.g., USHHS,
1988), heavy drug use and severe withdrawal symptoms. Tobacco dependence1 is an
individual difference veariable that may account for heterogeneous patterns of tobacco
use (e.g., chippers vs. heavy daily smokers). Dependence may also account for
differences in other key targets that are important to both tobacco researchers as well as
clinicians, such as withdrawal and relapse. To date, only modest inroads have been made
in the effort to explain important tobacco-related behaviors and outcomes using
individual difference variables such as dependence.
Progress in this area may depend upon improved measurement. At present,
measures of dependence are only modestly related to important dependence criteria such
Baker, 1994) and extant measures of dependence do not consistently agree with one
another (e.g., Moolchan, et al., 2002). Improved measures of dependence might not only
redress such problems, but might also illuminate dependence mechanisms (i.e., elucidate
underlying processes that produce dependence). In keeping with this perspective, the
goals of the present paper are to review the principles and procedures relevant to the
assessment of a complex construct such as dependence and to evaluate how extant
1 Throughout this paper, we will refer to tobacco dependence rather than nicotine or cigarette dependence. While nicotine is the addictive substance in tobacco, and cigarettes are the most common delivery device, we believe it is unduly restrictive to refer to either the single constituent, nicotine, or the delivery device per se. Marijuana dependence is not referred to as THC dependence or marijuana cigarette dependence. In addition, it should be noted that the chemical alterations in tobacco cigarettes as well as other forms of tobacco (e.g., snus, chew) play a significant role in the delivery and impact of nicotine. Therefore, to maintain a broad perspective, consistent with the PHS Guideline (Fiore, et al., 2000), we have chosen to use the term tobacco dependence.
3
Defining and assessing tobacco dependence
dependence measures “stack-up” against such psychometric criteria. This paper arises, in
part, from a belief that a greater familiarity with a construct validation approach to
assessment would allow tobacco scientists and clinicians to critically examine diverse
measures of dependence and to choose measures that have maximal validity for a target
problem/question. In addition, this paper provides guidelines for tobacco scientists who
wish to develop new dependence measures that have greater fidelity to their theoretical
views of dependence, have better psychometric properties, and enhance cumulative
progress in this area.
A Construct Validation Approach to Tobacco Dependence Measurement
A construct is “some postulated attribute of people, assumed to be reflected in test
performance” (Cronbach & Meehl, 1955, p. 283). A construct cannot be measured
directly and no single variable can provide a complete index of it. Instead, a construct is
defined by the interlocking system of laws that relate it to other constructs and to
observable properties or manipulations of the environment (Wiggins, 1973). Cronbach &
Meehl (1955) dubbed this interlocking system of laws the nomological network, which
comprises constructs as well as observable variables that are interconnected through a set
of theory-guided lawful relations. To date, measures of tobacco dependence have not
been clearly embedded in a theoretically-grounded nomological network that defines
dependence in terms of its relations to other constructs and observables. We will attempt
to explicate the nomological networks defining dependence as it is assessed by extant
measures.
In a construct validation approach, one’s theory of dependence will suggest a
nomological network that comprises both hypothesized unobservables (e.g., postulated
4
Defining and assessing tobacco dependence
mechanisms of dependence) as well as observables (e.g., responses on test items and
criterion measures). Thus, a theory of dependence should suggest: (a) the test items that
measure dependence, (b) how these test items are related to one another, and (c) how the
test items are related to theoretically, societally, and clinically meaningful outcomes
(criteria) that are attributed to dependence (see Figures 1 and 2; Meehl, 1995). Once a
theory of dependence has been articulated, and dependence items have been developed
and dependence criteria identified, the measure must be evaluated in terms of its
psychometric characteristics and utility. Below, we discuss these steps that are critical to
the development and evaluation of an assessment of a complex construct.
<Insert Figure 1 about here>
Measurement Development
A theory of dependence should suggest or specify features or mechanisms of
dependence. The mechanisms or features should suggest test “items” that reflect the
presence or amount of the dependence features/mechanism. Each of these elements of
dependence can be assessed at a variety of levels, ranging from neuropharmacological
and cognitive processes to expectancies and behaviors. For practical reasons, however,
most dependence measures will use self-report strategies and therefore, the measures will
target constructs and events available to awareness. Figure 1 depicts an example of a
multifaceted dependence construct in which dependence is defined on the basis of five
processes or features and each of these is measured with multiple items. For instance, the
dependence process by which conditional stimuli acquire motivational significance might
be assessed with items that ask about the extent to which cigarette cues attract attention
5
Defining and assessing tobacco dependence
or trigger craving. Clark and Watson (1995) provide an excellent review of the steps
necessary for theory-based assessment development.
Another critical issue to consider when evaluating any measure is its structure and
underlying assumptions or predictions of the construct implied by this structure. For
example, it is important to know whether a measure captures a unidimensional construct
or assesses multiple facets or components of an overarching construct. In assessing
tobacco dependence it is important to understand whether the measure is attempting to
assess dependence as a unidimensional construct or to assess various facets of
dependence using a multidimensional approach.
Selecting Dependence Criteria
A criterion is an outcome that a measure is intended to predict, and it should be
defined or chosen based on three considerations. First, there must be a theoretical link
between the criterion and the target construct (i.e., dependence). Second, a criterion is of
clinical, societal or research import. In other words, in addition to being theoretically
linked to the construct of interest, criteria are important constructs or observables that we
wish to assess or wish to predict for specific clinical, social or research reasons. Finally,
a criterion should be chosen based on the measurement goal; i.e., the goal driving the
measure development effort should guide criteria selection.
We claim that heavy use, withdrawal, and relapse are the core criteria of tobacco
dependence (e.g., USDHHS, 1988). In advancing these criteria we would have to argue
that these criteria have social, research or clinical import, and they can be clearly linked
to hypothesized dependence mechanisms or features. Because the goal of this paper is
not to explicate a particular dependence model we will not discuss at length mechanisms
6
Defining and assessing tobacco dependence
that might link the dependence construct to the criteria. However, anyone developing or
validating a dependence measure should be able to do so based upon the guiding theory.
For instance, a negative reinforcement model of dependence might hold that strength of
withdrawal symptoms and reinforcement from withdrawal-relief would constitute
important mechanisms of dependence (contrast with the different view of dependence
mechanisms depicted in Figure 1). In theory, such mechanisms would explain the
tobacco user’s high rates of tobacco use and a tendency to relapse (e.g., see Baker, et al.,
2004). Such a theory of dependence would suggest items that elicit information about
withdrawal severity or smoking rewards after abstinence (e.g., morning smoking after
overnight abstinence). The validity of these items would be appraised by correlating
them with the theoretically identified criteria (e.g., actual withdrawal symptoms upon
drug abstinence, relapse latency/likelihood).
Other potential core criteria might be suggested by other conceptualizations of
dependence. Examples of such criteria include continued drug use despite awareness of
harmful consequences or opposing motivation (i.e., a “harmful consequences” criterion;
Edwards, 1986), tolerance (Siegel, 1983), or loss of control (DiFranza, et al., 2002).
Once again, it is important that the underlying theory guide the selection of criteria.
If one is interested in the criteria, why not assess them directly? Why take the
circuitous route of assessing the target construct and then using the presence or
magnitude of the target construct to predict criteria? In essence, this is the DSM
approach in which the criteria become the items. For the purposes of classification it is,
no doubt, often desirable to measure dependence criteria directly (i.e., directly measure
the endpoints of addiction). But, this approach has limitations. First, endpoints may be
7
Defining and assessing tobacco dependence
very crude, insensitive indicators of dependence. For instance, a tendency to relapse back
to smoking may be influenced by numerous variables, and therefore may be a less
sensitive index than a “purer” measure of smoking motivation or need. Second, it is
often desirable to be able to predict dependence development, or to detect the initial
manifestations of dependence. Major criteria of dependence may not appear until late in
the course of the disorder, so that even if such criteria are measured dimensionally and
continuously, they will be insensitive to subtle manifestations. Third, direct assessment
of criteria often provides little added knowledge about the nature of dependence: i.e.,
little added scientific insight or explanation. For instance, directly measuring relapse
latency does not provide insight into the mechanisms that lead to relapse. Finally, in
many cases it is neither practical, economically feasible, nor possible to measure criteria
directly and accurately. For instance, while structured interview techniques may yield
reliable answers to questions, it remains unclear whether smokers and other addicted
individuals can validly report on the severity of their withdrawal (in general, over time)
or on the extent to which they spend time procuring/using drug (tobacco). Thus, while
any approach to assessing dependence may pose measurement problems, the assessment
of criteria is often so difficult so as to inspire the term “the criterion problem” (Wiggins,
1973).
As Figures 1 and 2 illustrate, similar constructs may occur both as components or
facets of dependence, as well as criteria. For instance, one may assume that heavy
tobacco use is a necessary causal influence on dependence, and yet, it also seems vital
that a dependence measure predicts level of tobacco use. This overlap between
dependence mechanisms or components and criteria raises the prospect of criterion
8
Defining and assessing tobacco dependence
contamination. Criterion contamination occurs when a researcher claims that the relation
between different measures provides insight into the nature of a construct but inspection
reveals significant content overlap between the two measures. The problems inherent in
such overlap can be reduced in two ways. First, different types of measures can be used
to measure the dependence features (e.g., using self-report items) and criteria (e.g.,
biochemical indices) and thereby reduce the extent to which method variance contributes
to associations. Second, and perhaps more important, researchers can ensure that items
predict more than just the most similar criterion (i.e., that “heavy use” items predict
criteria other than heavy use, such as relapse vulnerability).
Figure 2 also depicts secondary criteria in addition to core criteria. Secondary
criteria are theoretically linked to the construct and can provide information about the
construct, but do not have the importance of the core criteria. For instance, they may be
less tightly related to the guiding theory than are the core criteria, they may be integral to
dependence but may not be of social, clinical, or research import, or they may not be
relevant to the measurement goal. For example, an investigator might view demand
elasticity as a secondary criterion (e.g., Bickel & Marsch, 2001; Vuchinich & Tucker,
1988) because, while it is theoretically relevant, it has little intrinsic clinical import and
its estimation may not be the goal of the assessment.
Finally, Figure 2 depicts related constructs such as impulsivity, neuroticism, and
biological vulnerabilities (e.g., genotype). A related construct is not intrinsic to the target
construct, as specified by the theory of dependence. For instance, it does not reflect a
mechanism of dependence. Rather, it reflects a protective factor or a potential
vulnerability or a diathesis to developing the target construct (e.g., dependence).
9
Defining and assessing tobacco dependence
Impulsivity, therefore, might be considered such a construct (e.g., Bickel, Odum &
Madden, 1999; Mitchell, 1999). Thus, related constructs are conceptually distinct from
dependence, but may provide information on who is more likely to be more dependent
(e.g., individuals high in neuroticism may be more likely to be more dependent; Breslau,
information on factors that may modulate the magnitude of dependence. Therefore, a
sensitive appraisal of a dependence measure may require that such factors be included in
an evaluation study.
Assessing Relations Within the Nomological Network and Evaluation Context
Elaboration of the nomological network and the full set of variables relevant to
the evaluation (i.e., the evaluation context) provides a theoretical framework within
which to design studies and evaluate empirical results. That is, the nomological net
comprises not only the hypothetical features or mechanisms of dependence, and the items
used to tap them, but also all of the other variables and constructs theoretically linked to
dependence (e.g., withdrawal and the measures used to assess it). Therefore, the
nomological net not only defines the construct, but also provides a “context” for the
evaluation of the dependence measures (e.g., showing exactly which variables should be
related to dependence).
Measurement development is a process of “bootstrapping” – an iterative process
in which theory guides the refinement of a measure even after its publication, and
information gained in the process of measure validation is used to refine or modify the
theory, which then may have further impact on the measure (Cronbach & Meehl, 1955;
10
Defining and assessing tobacco dependence
Lakatos, 1970). This process of “bootstrapping” hinges on the ability to interpret the
empirical evidence.
The evaluation context depicted in Figure 2 explicitly acknowledges that heavy
use, withdrawal and relapse are multiply determined and that dependence is related to
these core criteria in a linear, probabilistic manner, such that greater dependence would
predict greater use, more severe withdrawal and greater likelihood of relapse. However,
these criteria are also related to other control variables, independent of dependence.2
Theory provides a priori information regarding which variables need to be controlled
within the experimental design or analyses in order to conduct a rigorous test of how well
the measure assesses dependence. When the role of control variables and related
constructs is explicit in the theoretical model, researchers can be proactive and create a
research design that will permit analysis of the separate roles of dependence, related
constructs and control variables. In fact, almost any study can collect data that can
provide important information on the validity of a dependence measure.
There are at least four different relations to examine when analyzing empirical
data in the context of construct validation.
1. Examine how items are related to one another, and how they are related to critical
constructs, including traditional tests used to establish the psychometric properties
2 It is important to note that while the target construct is theoretically linked to criteria, the target construct is not equivalent to a person’s status on the criteria. The core criteria are multiply determined and therefore their relation to the construct of dependence, as defined by the theory, is merely probabilistic and a good measure of the dependence construct merely allows one to predict the probability or magnitude of criteria. In the language of factor analysis, the target construct may be the general factor influencing the criteria, but researchers need to take into account additional specific influences. One vital product that research on the definition and measurement of dependence might produce is a clearer understanding of how important a person-factor such as dependence is in predicting relapse, versus environmental or contextual factors such as presence of stressors. Conversely, high values on particular criteria may not necessarily indicate dependence (e.g., medical patients may use drug (e.g., opiates) heavily, but not to meet other criteria such as relapse back to drug use). In other words, none of the three core criteria is necessary or sufficient to indicate dependence.
11
Defining and assessing tobacco dependence
of an assessment instrument. For instance, analyses should address the internal
consistency of an instrument (its reliability) and its structure (do the items all
correlate highly with one another or do correlational/covariance patterns appear to
reflect subgroupings of items suggesting that a construct has distinct “facets”).
There should be significant relations among the components, but if specific facets
or subfactors of dependence are also found (e.g., severity of withdrawal vs. high
rates of use), this suggests the need to determine the reliability of each facet. In
addition, analyses should address whether an item shows its strongest relations
with other items that are intended to assess the same target construct, or construct
facet. Thus, factor analyses should show that items have their highest loadings on
appropriate factors.
2. If the measure of dependence has different facets or subfactors, determine how
these are related to one another. Are they meaningfully related to one another as
one would expect if they are all components of the same construct (dependence)?
Does a single second-order factor explain their covariation?
3. Determine whether the new measure predicts the criteria it was designed to
predict and whether it does so in the manner predicted by the theory (e.g., Is the
measure related to criteria in a linear fashion or step function? Do the relations
reveal a taxon or a continuous construct?). In addition, if the measurement model
assessed in the previous step revealed different facets, then the relation of each
facet and the criteria should be examined. Are there distinct types of dependence
or is it a coherent construct in which all of its facets are similarly linked with the
criteria?
12
Defining and assessing tobacco dependence
4. Are the criteria meaningfully related to one another? If they are essentially
unrelated to one another, then the attempt to link them to a common influence
(i.e., dependence) will be doomed. These evaluations should be done after control
variables are used to remove variance that is not theoretically linked to
dependence.
5. Determine whether the “related constructs” perform as predicted. For instance,
are those high in trait neuroticism or impulsivity more likely to become dependent
upon exposure to tobacco?
The analyses described above would allow researchers to validate numerous
components of the nomological network and evaluation context, thereby allowing
researchers to focus on areas of the network that may need alteration, modification or
further assessment. For example, if withdrawal were found to be relatively unrelated to
either the target construct of dependence or other criteria, then perhaps withdrawal should
not be considered a core criterion.
Tobacco Dependence Measures: Traditional and New
Having reviewed psychometric principles relevant to measure development and
evaluation, we will now review five extant measures of tobacco dependence3 in light of
such principles. Specifically, we will evaluate dependence instruments with respect to
psychometrics (e.g., distributional properties, internal consistency), and validity:
convergent, discriminative and incremental validity. Convergent validity means that a
measure should correlate well with other valid measures of dependence, while
discriminative validity means that the measure should not correlate highly with
3 The measures reviewed here focus on adult measures of tobacco dependence. There are other measures that assess tobacco dependence in children and adolescents.
13
Defining and assessing tobacco dependence
theoretically unrelated constructs. Incremental validity refers to the extent to which a
measure predicts important criteria such as relapse risk over and above the predictions
yielded by other predictors. Finally, an ideal dependence measure should yield insight.
That is, it should provide an improved understanding of the mechanisms and processes
underlying dependence that can explain how the measure can predict the relevant criteria.
FTQ and FTND
The most commonly used tobacco dependence measures are the Fagerström
Tolerance Questionnaire (FTQ; Fagerström, 1978) and the Fagerström Test of Nicotine
Dependence (FTND; Heatherton, Kozlowski, Frecker, & Fagerström, 1991). The FTQ
was developed to measure the construct of physical dependence based on Schuster &
Johanson’s definition of physical dependence as “a state produced by chronic drug
administration, which is revealed by the occurrence of signs of physiological dysfunction
when the drug is withdrawn; further, this dysfunction can be reversed by the
administration of drug” (as cited in Fagerström, 1978). The FTND (Heatherton, et al.,
1991) was derived from the FTQ and was intended to overcome the psychometric and
validity limitations of the FTQ (e.g., Hughes, Gust & Pechacek, 1987; Lichtenstein &
Mooney, McClernon, & Jensen, 1999; Kenford, et al., 1994; Procyshyn, Tse, Sin, &
Flynn, 2002; Silagy, Mant, Fowler, & Lodge, 1994). In addition, some evidence suggests
that these dependence measures lack incremental validity; i.e., predict better than simpler,
competing measures. In this regard, studies have shown that using the number of
cigarettes smoked per day as a covariate eliminates the FTQ and FTND as significant
predictors of relapse (Dale, et al., 2001; Razavi, et al., 1999). Some evidence suggests
that the FTND is only slightly better than the FTQ at predicting dependence criteria (e.g.,
cotinine levels, withdrawal), if at all (e.g., Payne, et al., 1994; Pomerleau, et al., 1994)
Insight. Fagerström originally designed the FTQ to assess physical dependence,
which was loosely defined as the extent to which withdrawal symptoms were affected by
15
Defining and assessing tobacco dependence
tobacco abstinence and administration. Inspecting the items of the newer, more
psychometrically sound FTND reveals that they do not address withdrawal severity. For
the most part, they assess the individual’s rating of the aversiveness of abstinence and the
tendency or need to smoke in response to abstinence (e.g., in a response to overnight
abstinence). Thus, a content analysis suggests that the FTND equates dependence with
the motivational impact of withdrawal, and therefore it does suggest a relatively specific
motivational basis of dependence. The FTND’s ability to predict relapse (e.g., Alterman,
et al., 1999; Campbell, et al., 1996; Patten, et al., 2001; Westman, et al., 1997), suggests
that relapse is highly related to negative reinforcement motives or to abstinence-induced
changes in drug incentive valued (Baker, Piper, McCarthy, Majeskie & Fiore, 2004;
Robinson & Berridge, 1993).
The insight provided by the FTND is compromised by several considerations.
First, the FTND tends not to predict withdrawal severity well (Fagerström & Schneider,
1090; Hughes & Hatsukami, 1986), casting some doubt on its ability to index sensitively
abstinence effects. Second, to the extent that it is mulitfactorial (Haddock, et al., 1999;
Payne, et al., 1994), this suggests that it is not a “pure” measure of the hypothesized
dependence mechanism. Finally, if it is indeed the case that dependence may be
produced by diverse processes (e.g., Hudman et al., 2004), then the FTND would not
reflect all dependence facets.
Conclusion. The FTQ/FTND appear to predict relapse back to smoking, and they
suggest that relapse is associated with the motivational impact of abstinence. However,
their value is somewhat compromised by marginal reliability, an apparently unstable
factor structure, and an inability to predict well some dependence criteria such as
16
Defining and assessing tobacco dependence
withdrawal severity. However, the scales’ brevity and relations with relapse make it a
valuable clinical tool.
DSM-IV.
The Diagnostic and Statistical Manual, 4th Edition (DSM-IV; APA, 1994)
provides another commonly used measure of tobacco dependence, especially for the
purpose of clinical diagnosis and epidemiological research. The DSM-IV definition of
Substance Dependence is, “…a cluster of cognitive, behavioral, and physiological
symptoms indicating that the individual continues use of the substance despite significant
substance-related problems. There is a pattern of repeated self-administration that
usually results in tolerance, withdrawal, and compulsive drug-taking behavior.” (APA,
1994, p. 176). The DSM-IV is based on a syndromal medical model, rather than a
theoretical model of dependence. The developers of DSM-IV arrived at a consensus of
the principal features of tobacco dependence, and then generated items that would tap
these features. Thus, the development of the DSM was essentially atheoretical – there
was no need to postulate mechanisms or substrata of dependence. The instrument was
designed to tap phenotypic features – some of which would be considered criteria in other
models (e.g., withdrawal, relapse likelihood). Therefore, in the case of DSM-IV, no
construct validation approach was employed and no explicit or coherent theory guides the
ongoing process of measure development and validation. However, the fact that the
DSM-IV directly indexes major features of dependence affords it face validity and
general buy-in from clinicians and researchers.
Psychometrics. Typically, the DSM-IV questions are administered via an
individual interview, as opposed to a paper and pencil questionnaire. The interviews may
17
Defining and assessing tobacco dependence
be unstructured or semi-structured, such as the Diagnostic Interview Schedule (DIS;
Robins, Helzer, Croughan, & Ratcliff, 1981). An individual is given the categorical
diagnosis of substance dependence if s/he exhibits three symptoms within the same 12-
month period. The concordance between lay interviewers and psychiatrists using the DIS
for DSM-III is very good (ĸ=.78; Robins, et al., 1981). However, self-report
questionnaires, such as the Tobacco Dependence Screener (TDS), have recently been
developed to assess DSM-IV symptoms. The TDS is a 10-item self-report measure is
based on the atheoretical ICD-10, DSM-III-R and DSM-IV definitions of dependence
(Kawakami, Takatsuka, Inaba & Shimizu, 1999) and generates a continuous dependence
score. Developed in Japan, mainly with male smokers, it demonstrated good internal
consistency in different samples (α =.74-.81).
Validity. There is little doubt that the small set of dichotomous DSM items can
distinguish between light vs. heavy smoking. However, there is little evidence that DSM
items, as traditionally administered, provide a sensitive index of dependence as a
dimensional (continuous) construct. Moreover, an epidemiological study found that the
DSM-III-R was a significant, though weak, predictor of cigarette abstinence over one
year, but that the FTND was a better predictor and that number of cigarettes smoked per
day was the best predictor (Breslau & Johnson, 2000).4 Data on the TDS, the continuous
version of DSM items, indicate that it is correlated with smoking heaviness measures
(e.g., number of cigarettes smoked per day, carbon monoxide levels) and years of
smoking (Kawakami, et al., 1999; Piper, et al., 2004). In addition, individuals who quit
4 It is possible that future research will show the FTND or the DSM-IV to be accurate predictors of relapse and withdrawal. It is important to note that some of the negative findings cited above arose from studies that were underpowered and that did not use control variables to reduce error.
18
Defining and assessing tobacco dependence
smoking after a health risk appraisal were found to have lower TDS scores (Kawakami,
et al., 1999). Aside from this, there is no published data on the validity of a continuous
measure of DSM symptoms to predict such dependence criteria as withdrawal severity or
relapse vulnerability.
The DSM-IV and the FTND also evidence a lack of convergent validity
(Moolchan, et al., 2002; see also Kawakami, Takatsuka, Shimizu, & Takai, 1998). In
addition to differential rates of diagnosis, the FTND and DSM demonstrate little
covariance with one another, with the exception of assessing heavy use (Breslau &
Johnson, 2000), and have inconsistent relations with other constructs (e.g., depression;
Breslau & Johnson, 2000; Moolchan, et al., 2002; Tsoh, et al., 2001).
Insight. While the DSM-IV is a helpful diagnostic tool, it provides little insight
into the mechanisms of tobacco dependence. This is because DSM-type measures are
designed to assess reputed dependence criteria directly. Thus, questions assess use of
tobacco despite costs and a tendency to relapse back to drug use. The nature of items,
therefore, does not provide insight into processes or mechanisms that lead to such
outcomes. In addition, since the DSM approach is not based upon a particular model of
dependence, the validity of the measure is irrelevant to the validity of any particular
model or theory of dependence. Moreover, its items were not selected to provide reliable
measures of distinct facets of dependence. Measuring tobacco dependence with the
DSM-IV criteria is similar to measuring end-state organ damage to assess high blood
pressure. In fact, the DSM-IV could be considered a list of criteria, rather useful for
validating other dependence measures.
19
Defining and assessing tobacco dependence
As a rather small set of dichotomous items the DSM-IV can distinguish between
heavy vs. light use of tobacco but appears to be limited in terms of its ability to assess
dependence as a continuum. Therefore, unless the DSM-IV items were somehow made
more continuous it seems that DSM-type measures would be inappropriate for assessing
degree of dependence among regular smokers. Moreover, as a supposed unidimensional
measure such measures would be inappropriate for tapping different facets of
dependence.
Conclusion. The DSM-IV was not designed to provide theoretical insight into
the construct of dependence, nor was it designed to provide a continuous measure of what
may be a multidimensional construct. In fact, relatively little evidence exists that
supports the convergent and incremental validity of the DSM items in terms of their
ability to predict such outcomes as withdrawal severity and relapse. The DSM-IV
dependence criteria have substantial face validity which promotes general acceptance and
supports their use as validation criteria for other dependence measures.
CDSThe Cigarette Dependence Scale (CDS) was developed by surveying smokers via
the mail and Internet to assess what signs smokers felt indicated addiction to cigarettes
(Etter, Le Houezec & Perneger, 2003). The authors then used the strength of smokers’
endorsements, psychometric consideration, and content coverage to select the final sets of
items that would constitute the 5- and 12-item scales. Unlike a construct validation
approach, this empirical approach to measure development was not based on a theory of
dependence. However, the authors explicitly selected items to ensure coverage of DSM-
IV and ICD-10 symptoms. Moreover, while not stated as an explicit goal, the authors
also apparently sought some coverage of the FTND content domain: e.g., two of the
20
Defining and assessing tobacco dependence
CDS-5 and CDS-12 items are essentially the same as two FTND items. Thus, the CDS is
an amalgam of DSM/ICD and FTND items that is intended to generate a continuous
measure of dependence.
Psychometrics. To date, only one study has published data on the CDS versions,
using data collected via the mail or Internet. The two scales both had good internal
consistency (CDS-12 α =.90; CDS-5 α =.84). The CDS-12 and CDS-5 were both slightly
skewed toward higher values. Test-retest correlations were ≥.60 for all items and ≥ .83
for the full scales. Factor analysis suggested a unidimensional structure for the CDS-12.
Validation. The CDS scales were significantly correlated with number of
cigarettes smoked per day, whether a smoker was a daily or occasional smoker, the
strength of urges during the last quit attempt, and cotinine level (Etter, et al., 2003). The
CDS-12 and CDS-5 appeared somewhat better than the FTND at differentiating
occasional from daily smokers and more strongly correlated with urge to smoke during
the participants’ previous quit attempt. However, it is unknown whether these
differences in validity coefficients achieved statistical significance. Curiously, the FTND
and CDS-5 were more strongly correlated with cotinine levels than was the CDS-12.
Some respondents had participated in a cessation trial. None of the three dependence
measures (i.e., the FTND, CDS-5 or CDS-12) was a significant predictor of relapse
likelihood. However, only a third of potential respondents participated in the follow-up
study, which may have produced considerable response bias.
Insight. Similar to the FTND and DSM-IV, there is overlap between criteria and
items. For instance, the CDS elicits information on smoking heaviness, and smoking
heaviness is also used as a criterion in validation research (Etter, et al., 2003). Moreover,
21
Defining and assessing tobacco dependence
like the DSM measures, the CDS scales largely target outcomes or consequences of
dependence (e.g., subjective sense of the difficulty of quitting and sense of being addicted
to cigarettes). In addition, the CDS-12 (and presumably the CDS-5) has a
unidimensional structure. Therefore, these measures are not designed to elucidate
dependence processes or mechanisms, or to assess the multidimensional nature of
dependence.
Conclusion. The CDS does appear to reflect the DSM-IV/ICD-10 as well as
FTND content, and demonstrate sound psychometric properties. Moreover, as opposed
to the former instruments the CDS was designed to yield a continuous measure of
dependence. The CDS is a new instrument and therefore only a modest amount of
validity evidence supports its use. In particular, there is no evidence that it assess relapse
vulnerability. The instrument was designed to index dependence outcomes aned not
dependence mechanisms and it appears to have a unidimensional structure. Therefore, it
cannot index distinct dependence mechanisms. The face validity of its content, its high
reliability, and the availability of brief forms should promote its clinical use once more
validity data are available.
NDSS
The Nicotine Dependence Syndrome Scale, a 19-item self-report measure, was
developed as a multidimensional scale to assess nicotine dependence using Edwards’
theory of the dependence syndrome (Shiffman, Waters & Hickcox, 2004). Edwards
developed a construct of an alcohol dependence syndrome comprising several essential
elements: a narrowing in the repertoire of drinking behavior, salience of drink-seeking
behavior, increased tolerance, withdrawal symptoms, use of alcohol to avoid or relieve
22
Defining and assessing tobacco dependence
withdrawal symptoms, subjective awareness of a compulsion to drink, and a
reinstatement of the syndrome after abstinence (Edwards, 1976). As with any syndrome,
no elements are necessary nor sufficient for the syndrome to be present, but with
increasing severity, the syndrome will become increasingly coherent (Edwards, 1986).
Edwards also stated that while greater dependence would result in increased alcohol
intake and diminished responsiveness to social controls, these were secondary to the
syndrome itself. Applying this syndromal approach to tobacco dependence, Shiffman
and colleagues originally developed a 23-item scale, but psychometric analysis lead to a
revised 30-item scale, which was ultimately pared to a 19-item version. It is important to
note that none of the psychometric, nor validity data presented in the sole publication on
this instrument is based on this specific 19-item set. Much of the validity data (prediction
of relapse and withdrawal) was derived from the initial 23-item instrument. Some
additional convergent validity information was obtained using the subsequent 30-item
instrument (e.g., relations with heaviness of smoking, self-rated addiction). The final 19-
item NDSS is a subset of the earlier 30-item instrument and no validity data have been
published on this particular item set and its associated factor scores.
Scoring of the NDSS is somewhat complex in that it requires that item responses
be multiplied by a set factor loading before summing the items into their scale scores. It
is important to note that the NDSS was intended to complement, not replace, traditional
dependence measures such as the DSM. The authors note that there is little content
overlap between the two types of measures.
Psychometrics. To date, one paper presents data only one study has published
data on the NDSS (but this contains data from three studies). There are no published data
23
Defining and assessing tobacco dependence
on the final 19-item NDSS. Therefore, psychometric data presented here are based on the
revised 30-item scale. The authors developed items to address 5 different dimensions of