We sought to examine whether DSM-IV and FTND criteria of nicotine dependence could be used to identify subgroups of individuals with distinct patterns of endorsement of these criteria and consequent variations in risk for nicotine dependence and other comorbid psychopathology.
LCA results can shed light on two underlying patterns of data. Either underlying all nicotine dependence criteria is a common unifying dimension (e.g., smoking quantity) or sets of criteria are etiologically distinct and oblique to each other resulting in subtypes of individuals. In the first instance, where endorsement of a set of criteria largely indexes a severity continuum, plotting the endorsement probabilities (as done in ) would result in a series of lines that are largely parallel to each other but have varying positions (i.e., index severity) along the y-axis. On the other hand, if distinct subtypes are identified, these lines would be expected to intersect, such that one class of individuals would be characterized by high endorsement probability on a series of criteria, while another class would show low endorsement probabilities for those criteria with a corresponding increase in endorsement probability for another set of criteria. In our sample of regular smokers, four groups of individuals were identified—those with low endorsement of all criteria, those with high endorsement of all criteria, those with intermediate endorsement of FTND and DSM-IV, with particularly high levels of tolerance, and a class of individuals with intermediate endorsement of FTND criteria but only modest endorsement of DSM-IV–based dependence criteria. Thus, while our findings primarily reflect a severity continuum (high endorsement vs. low endorsement of all criteria—in , the lines denoting the HDHF, MDMF, and LDLF classes are nearly parallel), there is possible and suggestive support for a subgroup (LSMF class) of smokers. This latter grouping is demonstrated by the intersection of endorsement probabilities for the DSM-IV tolerance criterion across the MDMF and LSMF classes (with a corresponding reversal of endorsement probabilities for some of the FTND criteria across these classes (e.g., higher endorsement of hate to give up first cigarette in the LSMF class).
The MDMF class may be viewed as an intermediate risk class, and while the LSMF class appears to index similar vulnerability to the MDMF class, it is distinguished by a contrasting low endorsement of tolerance and a corresponding high endorsement of smoking less than 20 CPD. Furthermore, examination of the psychiatric covariates that aggregate with the classes highlights two key observations. First, compared with those in the LDLF class, those in the LSMF class are more likely to be male, initiate smoking prior to age 13, report greater DSM-IV and FTND symptoms, report craving, and meet criteria for a lifetime history of DSM-IV conduct disorder and ADHD problems. Thus, while these individuals are lighter smokers, they demonstrate increased vulnerability to some psychopathology. Second, despite this increased risk, relative to those in the HDHF and MDMF class, the individuals in the LSMF class report the same levels of, if not significantly less, psychopathology. For instance, while membership in LSMF class is associated with a 2.2-fold increased likelihood of conduct disorder, membership in the MDMF class is associated with a statistically comparable 1.8-fold increased risk. Furthermore, rates of neither AD nor major depressive disorder appear to be elevated in the LSMF class, while those in the MDMF class are 1.1–1.6 times more likely to meet criteria for a lifetime history of these disorders.
In addition to the marked difference in endorsement of tolerance, the apparent distinction between the MDMF and LSMF class is largely attributable to levels of smoking and recency of becoming a regular smoker. Fifty-seven percent of those in the LSMF class reported smoking 11–19 CPD (with the remainder smoking less than 11 CPD), while an overwhelming 62% of those in the MDMF (also 79% of those in the HDHF class) report smoking more than 19 CPD. Additionally, individuals in the LSMF group were more recent regular smokers, perhaps indicating that they were in the early stages of their smoking trajectories. This variation in smoking may also have contributed to the extreme discordance (0% vs. 100%) in endorsement of DSM-IV tolerance, which incorporates elements of CPD.
However, this raises the question of whether nicotine dependence and its correlated impediment of successful smoking cessation is even a concern in lighter smokers. Our results show that despite lower CPDs, nearly half the individuals in the LSMF class reported withdrawal or withdrawal relief, which is a key predictor of failed cessation (Madden et al., 1997
; Rubinstein, Benowitz, Auerback, & Moscicki, 2008
; Xian et al., 2005
). Several studies have demonstrated that light smoking, even less than 10 CPD, can be associated with diminished autonomy over smoking, persistent smoking, and nicotine dependence (Coggins, Murrelle, Carchman, & Heidbreder, 2009
). Furthermore, studies have found light smoking (defined variously; Husten, 2009
) to be somewhat unstable with the social context of smoking (e.g., peers), as well as cooccurring alcohol use and psychopathology, contributing to escalation of smoking quantity and frequency in lighter smokers (Hukkinen, Kaprio, Broms, Koskenvuo, & Korhonen, 2009
; Levy, Biener, & Rigotti, 2009
; White, Bray, Fleming, & Catalano, 2009
). Furthermore, all groups reported persistent desire and repeated unsuccessful attempts to cut back or quite smoking—in fact, this criterion failed to provide any discriminative utility across classes. Therefore, measurement of nicotine dependence in light smokers is necessary for research and practice.
A related and growing concern is that rates of daily light smoking (as well as light and intermittent smoking/LITS) have increased, particularly, in the United States. Some argue that the surge in lighter smoking patterns is attributable to greater social sanctions and prohibitions imposed on smoking (Shiffman, 2009
). In our data as well, the FTND criteria of smoking where prohibited along with giving up important activities were considerably elevated in the LSMF class, even though these individuals smoked 19 or fewer CPD. With increasing denormalization of smoking, lighter smokers may continue to grow in numbers, making their characterization a priority.
The LSMF group requires further study. While they appear unique to this sample, none of the covariates tested were successful in distinguishing these individuals from other groups. It is possible that these individuals will eventually transition to the MDMF or HDHF class, and longitudinal data will be extremely informative in confirming whether this is the case. In contrast, as the individuals in this group smoke less than those in the MDMF group, longitudinal data can also be used to determine whether this group will experience successful smoking cessation, making this group a prime target for treatment and interventions. This will be examined in subsequent studies. However, the current data failed to successfully distinguish this group from the others based on post-hoc covariate analyses. Given the methodological caveats discussed below, we cannot be certain that this group is of substantive relevance until further characterized.
There are some key methodological caveats that should also be considered when viewing these findings. In some instances, such as when there is only partial conditional independence (i.e., sets of observed variables are correlated over and above their relationship via the latent classes), overextraction of classes can occur. In such cases, factor mixture models (where factors are nested within classes and where endorsement probabilities reflect changes in factor loadings and thresholds) have been shown to facilitate superior interpretation of data. Because of this, and also due to the strong observed support for the latent classes reflecting a severity continuum, we conducted an exploratory factor analysis of these data. The three factor solution fit best with the first factor loading well on all the DSM-IV criteria as well as the FTND criteria of difficulty smoking where prohibited and smoking when ill; the second factor loading primarily on the FTND criteria, particularly time to first cigarette, hate to give up first cigarette of the day, and smoking within the first hour of waking; and the third factor showing very high loadings on DSM-IV tolerance (0.997) and the FTND criterion of CPD (0.996), which are key to the LSMF class. Thus, while a factor mixture model may provide an alternate interpretation of these data, a model that accounts for the complexity of both factors and classes is somewhat intractable for this situation (particularly with this sample size, as several complex models failed due to empty cells in the joint distribution). We performed modified LCA using latent class factor analysis (where a single factor is allowed to have varying means across classes) and also partially relaxed the assumption of conditional independence (e.g., a factor with loadings on tolerance and CPD) but neither of these produced any significant improvement in model fit. Additionally, we performed these LCA by dichotomizing the two ordinal FTND measures of time to first cigarette (dichotomized as either <6 min or 6 min and longer) and CPD (dichotomized as 26+ CPD or less)—this eliminated the LSMF class resulting in a severity continuum of LDLF, MDMF, and HDHF that indicates that the identification of the LSMF class is reliant on jointly modeling tolerance with an indicator of light smoking (e.g., CPD ≥ 11 or CPD = 11–19).
Latent mixture models, such as LCA, do not directly address the relative utility and relevance of individual DSM
and FTND criteria in the diagnosis of nicotine dependence. From a statistical perspective, approaches such as factor analysis (or item response modeling) are best suited to such interpretations. Several such factor analyses have been conducted. For instance, Saha et al. (2010)
found that all DSM-IV
nicotine dependence criteria loaded well on an underlying unidimensional construct. In contrast, B. O. Muthen and Asparouhov (2006)
have argued that DSM-IV
nicotine dependence is best conceptualized as a factor mixture model with three classes (including a zero class) and a single factor nested across the two nonzero classes. For FTND, both one- and two-factor (smoking pattern and morning smoking) solutions have been suggested (Haddock, Lando, Klesges, Talcott, & Renaud, 1999
). However, across these studies, tolerance and CPD have been observed to have robustly high factor loadings, suggesting that they are central to the diagnosis of nicotine dependence.
We did not find evidence for increased genetic vulnerability or risk attributable to environmental influences of parental smoking to be over-represented in any class. Two possible explanations exist—first, sample size may have limited our statistical power to distinguish across these groups and second, excluding nonregular smokers may have accounted for a majority of heritable influences and the prominent role of rearing environment. This latter theory is somewhat supported by the increased numbers of members across all four classes in the high genetic (and environment) risk categories (). Additionally, multiple twin studies (Heath, Martin, Lynskey, Todorov, & Madden, 2002
; Kendler et al., 1999
; Lessov et al., 2004
; Madden, Pedersen, Kaprio, Koskenvuo, & Martin, 2004
; Pergadia, Heath, Martin, & Madden, 2006
) show that after accounting for the genetic overlap between regular smoking and nicotine dependence (and persistence), most of the variation in liability to dependence/persistence is individual specific (Rose, Broms, Korhonen, Dick, & Kaprio, 2009
). These studies also note that while familial environment plays a role in smoking initiation, after accounting for these early stages, there may not be additional specific shared environmental factors that impact dependence alone (Broms, Silventoinen, Madden, Heath, & Kaprio, 2006
; Madden et al., 2004
; Maes et al., 1999
Finally, it is important to note that time to first cigarette played an important role in class membership—those in the MDMF class were more likely to endorse smoking their first cigarette over an hour after waking (37.6%) compared with only 14.5% of those in the LSMF class. Recent studies (Baker et al., 2007
; Haberstick et al., 2007
; Muscat, Stellman, Caraballo, & Richie, 2009
; Niaura, Shadel, Goldstein, Hutchinson, & Abrams, 2001
) have begun to focus on the salience of time to first cigarette as a marker of phenotypic and genetic vulnerability to problematic smoking, and our analyses underscore the need for studies focused on this aspect of the FTND.
Some limitations of this study are noteworthy. First, this is an OOT study with a unique sampling design—to what extent these classes would generalize to other populations remains to be seen. However, a considerable strength of this sample is the availability of both DSM-IV–
and FTND-based criteria on a large cohort of adolescents and young adults. Second, while the mean age of the sample is 21.4 years, it is likely that some individuals may not be past the period of risk for the emergence of nicotine dependence symptomatology. However, as likelihood of class membership was invariant with age, this sample characteristic did not influence class assignment. Third, despite the considerable wealth of data, some measures (e.g., social context of smoking) were absent. Fourth, some distinctions across classes were attributable to levels of CPD—to what extent individuals can reliably distinguish between smoking 11–19 cigarettes versus smoking 20–25 cigarettes, or more than 25 cigarettes, may have contributed to the results. Fifth, other instruments exist for the assessment of nicotine dependence, such as the Nicotine Dependence Syndrome Scale (Shiffman, Waters, & Hickcox, 2004
), the Hooked on Nicotine Checklist (DiFranza & Wellman, 2005
; Wellman et al., 2005
), and the Wisconsin Inventory of Smoking Dependence Motives (Piper et al., 2004
) as well as others. Psychometric comparisons of these scales with the DSM-IV
and FTND exist (e.g., Wellman et al., 2006
)—however, our interview did not include these assessments.
As we approach the new era of DSM-V, our findings suggest that while measurement and diagnosis of liability to problematic smoking may vary, a combination of assessment instruments, when possible, may afford the most accuracy. In our study, using both DSM-IV and FTND criteria provided a more comprehensive overview of vulnerability to nicotine dependence. While each instrument may capture different aspects of nicotine dependence, our analyses suggest that the DSM-IV and FTND criteria work synergistically and where possible should be used in tandem to provide a finer-grained view of cigarette smoking behavior.