Psychiatric and Familial Predictors of Transition Times Between Smoking Stages

Abstract

The modifying effects of psychiatric and familial risk factors on age at smoking initiation, rate of progression from first cigarette to regular smoking, and transition time from regular smoking to nicotine dependence (ND) were examined in 1,269 offspring of male twins from the Vietnam Era Twin Registry. Mean age of the sample was 20.1 years. Cox proportional hazard regression analyses adjusting for paternal alcohol dependence and ND status and maternal ND were conducted. Both early age at first cigarette and rapid transition from initiation to regular smoking were associated with externalizing disorders, alcohol consumption, and cannabis use. Rapid escalation from regular smoking to ND was also predicted by externalizing disorders, but in contrast to earlier transitions, revealed a strong association with internalizing disorders and no significant relationship with use of other substances. Findings characterize a rarely examined aspect of the course of ND development and highlight critical distinctions in risk profiles across stages of tobacco involvement.

1. Introduction

The pathway to nicotine dependence (ND) involves a progression through stages of increasingly frequent smoking behaviors and results in an addiction associated with a myriad of negative health outcomes (Fagerstrom, 2002; Bartecchi et al., 1994). Experimental cigarette use is developmentally normative among adolescents: over 70% of individuals under the age of 18 have tried at least one cigarette in their lifetimes (CDC, 1998; Nelson and Wittchen, 1998). However, only a subset of experimenters go on to become daily smokers (one-third of those who begin smoking; van den Bree et al., 2004) or to develop ND (approximately half of all daily smokers; Breslau et al., 2001), suggesting that distinct mechanisms may underlie different stages of smoking behaviors. Much of the smoking literature has therefore examined risk factors in relation to the likelihood of reaching particular stages of tobacco involvement (e.g., initiation, regular smoking), but few studies have characterized risk associated with the rate of transition between smoking milestones (e.g., transition time from first cigarette to regular smoking). Given the association of early smoking initiation with increased probability of developing smoking-related problems (Grant, 1998; Hu et al., 2006) and the public health implications of identifying individuals at risk for rapid escalation to greater tobacco involvement, investigation of risk factors that modify the rate of progression between smoking milestones is clearly warranted.

1.1 Psychiatric and familial risk for smoking outcomes

Psychiatric disorders are more prevalent among adult smokers than non-smokers (Black et al., 1999; Farrell et al., 2001; John et al., 2004). The odds of meeting criteria for a psychiatric illness among daily smokers were estimated to be 2.7 times that of non-smokers in one population-based study (Lasser et al., 2000). The link between ND and psychiatric disorders has also been well established: nicotine dependent status is associated with elevated rates of psychiatric illness (Breslau et al., 1994; Grant et al., 2004) and a higher number of comorbid disorders (John et al., 2004). Other substance use disorders (SUDs) are the most consistently identified and robust psychiatric correlates of smoking behaviors (Grant et al., 2004; Upadhyaya et al., 2002), occurring at rates approximately two to three times higher among smokers than non-smokers (Black et al., 1999; Lasser et al., 2000), although mood and anxiety disorders have frequently been associated with smoking behaviors as well (Breslau et al., 1991; Fergusson et al., 2003; Patton et al., 1996).

Investigations of adolescent and young adult cigarette smoking have produced similar findings regarding co-occurring psychopathology: psychiatric disorders, especially SUDs, are more commonly found in smokers (Grucza & Bierut, 2006; Nelson & Wittchen, 1998; Patton et al., 2006; Upadhyaya et al., 2002). Familial and psychiatric risk factors for smoking outcomes have also been examined more broadly in younger samples, revealing risk associated with initiation of alcohol and cannabis use (Dierker et al., 2004; van den Bree et al., 2004) and problem behaviors that originate in childhood (i.e., disruptive disorders, school problems; Dierker et al., 2001; Rohde et al., 2004). Some of these studies have included parenting practices as well (Hill et al., 2005; Lewinsohn et al., 2000), but parental smoking behaviors are the most widely studied. Parental smoking history has been associated with increased likelihood of offspring experimental use, regular smoking, and lifetime ND (Bricker et al., 2007; Flay et al., 1998; Hu et al., 2006). Boomsma and colleagues’ work (1994), part of a growing literature on the heritability of smoking behaviors, suggests that the close tie between parental and offspring smoking is attributable to genetic transmission of risk (Boomsma et al., 1994).

Heritability estimates for smoking outcomes span a wide range, from approximately 30% to 80% (Kendler et al., 1999; Maes et al., 2004; True et al., 1997), depending on the sample and phenotype assessed. Much of the genetically-informative research in this area has examined multiple smoking-related constructs in conjunction to determine common and specific liability across phenotypes (Heath et al., 1999; Koopmans et al., 1999; Madden et al., 1999). Cross-phenotype investigations using stages of involvement in tobacco use have provided critical information regarding continuity of genetic contributions over developmental course. For example, Maes et al. (2004) reported high heritability for initiation of tobacco use, regular smoking, and nicotine dependence (75%, 80%, and 62%, respectively) and substantial genetic correlations between phenotypes, but found that over one-third of genetic liability to ND was attributable to factors specific to ND (Maes et al., 2004). Investigations of common heritable influences on initiation and persistence of smoking (Madden et al., 1999; True et al., 1997) and on initiation and ND (Kendler et al., 1999) have reached a similar conclusion: genetic risk for later stages of smoking behaviors is associated with but not fully explained by heritable contributions to smoking initiation. This line of research has also provided evidence for the substantial role of shared environmental influences on the initiation of cigarette use (accounting for approximately 30% to 55% of the total variance) and the virtual absence of such influences in later stage smoking behaviors (Heath et al., 1999; Madden et al., 1999; Stallings et al., 1999; True et al., 1997). In sum, findings indicate that the relative contributions of environmental versus genetic factors and the nature of heritable influences on smoking behaviors change across stages of tobacco involvement.

1.2 Transitions in the pathway to nicotine dependence

Consistent with genetically-informative research, investigations of psychosocial and psychiatric correlates of smoking milestones have revealed both cross-stage and stage-specific risk factors for progressing to more severe smoking behaviors. Alcohol and cannabis use, especially problem use, have been linked to increased risk for all smoking stage transitions: initiation of smoking (Dierker et al., 2001; Kendler et al., 1999; Rohde et al., 2003), progression from experimental use to regular smoking (Dierker et al., 2004; Rohde et al., 2004), and transition from regular smoking to ND (Breslau et al., 2004; Dierker et al., 2001). Delinquency and externalizing disorders have been tied to progression in smoking behaviors across stages as well (Clark & Cornelius, 2004; Dierker et al., 2004; van den Bree, 2004), whereas evidence that risk conferred by mood and anxiety disorders is global rather than stage-specific is inconsistent (Dierker et al., 2004; Kendler et al., 1999). In contrast, findings regarding the role of familial smoking behaviors (with some exceptions, e.g., Otten et al., 2007) suggest that their impact increases in later smoking stage transitions (Bricker et al., 2007; Dierker et al., 2001; Flay et al., 1998). Taken together, these studies provide a foundation for exploring continuity versus change in psychiatric and familial risk factors for transitions over the course of ND development. The next step is to assess the predictive power of these risk factors for determining the rate of progression to ND across stage transitions.

The current investigation used a genetically-informative design to determine the modifying effects of psychiatric and familial risk factors on transition times at three critical junctures in the pathway to ND. Predictors of the rates of transition from non-use to first cigarette (i.e., age at first cigarette), first cigarette to regular smoking, and regular smoking to ND onset were assessed in a survival analysis format with an offspring-of-twins sample.

2. Method

2.1 Participants

The sample was comprised of 1,269 offspring of twins from the Vietnam Era Twin Registry (VETR), a national registry of male like-sex twin pairs in which both served in the military during the Vietnam Era (1965 - 1975). (Information regarding construction of the VETR and method of determining zygosity has been reported in prior publications [Eisen et al., 1987; Eisen et al., 1989]). Data were drawn from a study of alcohol use disorders (AUDs) and related psychopathology that involved twin pairs, their offspring, and biological mothers of offspring. Selection of twin fathers was based on reports of offspring birth dates collected via questionnaire in 1987 and a structured diagnostic interview conducted in 1992 to obtain fathers’ histories of DSM-IIIR drug dependence and other psychiatric diagnoses, including ND and AD (Tsuang et al., 1996). Male twin pairs were eligible for study participation if one or both members of the pair endorsed full DSM-III-R criteria (lifetime) for AD. A random sample of twin pairs in which neither co-twin was alcohol dependent was also included in this high-risk cohort to serve as a control group. At least one member of each twin pair was required to have one or more offspring born between 1974 and 1988 to be eligible for participation. Response rate for the 1,464 eligible twin fathers was 83% (n = 1,213); 11% declined participation and 6% were deceased, disabled, or could not be located.

Twins were contacted first to obtain permission to contact mothers and offspring. The mothers were then contacted for participation and for permission to contact their children. Participating mothers were associated with 1,588 target offspring. Permission was granted by mothers to contact 1,487 offspring (93.6%), 135 (9%) of whom were unavailable, unlocatable or incapacitated. Of the remaining 1,352 offspring, 1270 (93.9%) were interviewed. Paternal alcohol dependence status was not associated with offspring response rate (Scherrer et al., 2004). One participant was excluded from the present study due to missing age.

Based on father AD status and the AD status of his co-twin, offspring were classified into four AD risk status groups. AD risk group 1 consisted of offspring of alcohol dependent fathers (high genetic (G) and high environmental (E) risk; n=587). AD risk group 2 consisted of offspring of unaffected monozygotic (MZ) twins whose co-twins were positive for AD (high G, low E risk; n=162) and AD risk group 3 consisted of unaffected dizygotic (DZ) twins whose co-twins were positive for AD (moderate G, low E risk; n=157). AD risk group 4 was comprised of offspring of unaffected fathers with unaffected co-twins (low G and low E risk; n=363). In addition to AD risk groups, offspring were categorized into one of four ND risk groups derived from paternal reports of lifetime ND in the 1992 twin assessment. Similar to the AD risk groups, ND risk group 1 (n = 611) was comprised of offspring whose fathers met ND diagnostic criteria (high G, high E); ND risk group 2 (n = 102) consisted of offspring of unaffected MZ twins whose co-twins were positive for ND (high G, low E); ND risk group 3 (n = 139) consisted of offspring of unaffected DZ twins whose co-twins were positive for ND (moderate G, low E); and ND risk group 4 (n = 417) was comprised of offspring whose fathers and fathers’ co-twins were unaffected (low G, low E). The rationale for these risk groups has been previously reported (Jacob et al., 2001).

Participants ranged in age from 12 to 28 years, with a mean age of 20.10 years (S.D.=3.98). Female offspring comprised 51.5% of the sample. Approximately 95% of participants self-identified as non-Hispanic Caucasian and 61% of the offspring had fathers who completed 12 or more years of education.

2.2 Procedure

Data were collected from January to November 2001 by the Institute for Survey Research (ISR) at Temple University using Computer Assisted Telephone Interviews (CATI). Interviewers underwent extensive training that involved both ISR staff and project investigators. The Institutional Review Boards at participating institutions approved study protocols prior to the start of data collection. All participants consented to the study prior to the start of interviews.

2.3 Assessment battery

A modified version of the Semi-Structured Assessment for the Genetics of Alcoholism (SSAGA-II) (Bucholz et al., 1994; Hesselbrock et al., 1999) was used to collect psychiatric and psychosocial histories from offspring in telephone interviews. Detailed accounts of symptoms and onset of episodes were obtained in addition to lifetime DSM-IV diagnoses for substance use, mood, anxiety, and externalizing disorders. Offspring also reported on childhood physical and sexual abuse events and on quality of relationships with parents and parental discipline during early childhood (i.e., ages 6-13 years).

Paternal psychiatric diagnoses and demographic information were obtained from histories provided in the 1992 study (Tsuang et al., 1996) in combination with brief telephone interviews conducted at the time of data collection for the current study. Biological mothers were also assessed using an adaptation of the SSAGA-II. Maternal interviews covered mother’s history of AUDs and major depression, and included screens for drug use, ND, mania, and antisocial personality disorder. Mother’s response to the question “How soon after waking do you smoke your first cigarette?” was used as an index of maternal ND. It is the first question in the Fagerstrom Test for Nicotine Dependence and believed to be the best single indicator of ND (Fagerstrom, 2003). Responses were used as an ordinal variable, with a lower number of minutes to first cigarette indicating increased likelihood of ND (5 or fewer: n = 130; 6-30: n = 139; 31-60: n = 112; 60 or more: n = 148; non-smoker: n = 666). Mothers also provided diagnostic information to assess offspring oppositional defiant disorder (ODD) and attention deficit hyperactivity disorder (ADHD). Approximately 82% of the participating offspring’s mothers were interviewed and provided diagnostic data.

2.3.1 Smoking outcomes

Age at first cigarette, age at onset of regular smoking, and age at onset of nicotine dependence were obtained in the history of smoking behaviors provided by offspring in the SSAGA-II. Regular smoking was defined as smoking daily or nearly daily for a period of three weeks or longer. Nicotine dependence referred to full DSM-IV criteria, i.e., 3 or more of the 7 possible ND symptoms reported as occurring in the same 12 month period.

2.3.2 Predictors

Psychiatric and familial correlates of smoking outcomes derived from the literature were assessed as possible predictors of transitions between smoking milestones. Risk factors included parental psychopathology and problem substance use (Brook et al., 2006), history of childhood physical and sexual abuse (Nelson et al., 2006; Nichols & Harlow, 2004), separation from father prior to age 18 (Otten et al., 2007), inconsistent parental discipline and lack of closeness to parents (O’Byrne et al., 2002), female gender (Ridenour et al., 2006), offspring alcohol and cannabis use and use disorders (Dierker et al., 2004; van den Bree et al., 2004), and offspring psychiatric disorders (Nelson & Wittchen, 1998; Upadhyaya et al., 2002). Rates of risk factors in the sample are displayed by domain in Table 1.

Table 1. Prevalence of Risk Factors.

Familial Psychiatric History, Family Environment, and Gender		Offspring Substance Use and Psychiatric Disorders
Parental Psychiatric History (Lifetime)		Substance Use and Abuse/Dependence
Paternal antisocial personality disorder	2.8%	Alcohol use	79.4%
Paternal major depressive disorder	8.2%	Alcohol dependence	10.0%
Paternal drug dependence	11.9%	Cannabis use	46.2%
Maternal alcohol dependence	9.5%	Cannabis abuse	8.7%
Maternal major depressive disorder	17.7%	Externalizing Disorders
Parenting (Offspring Report For Ages 6-13)		Attention deficit hyperactivity disorder	6.2%
Not close to father	10.7%	Conduct disorder	5.8%
Not close to mother	4.8%	Oppositional defiant disorder	6.2%
Father inconsistent with discipline	11.8%	Internalizing Disorders
Mother inconsistent with discipline	12.0%	Major depressive disorder	14.5%
Offspring Abuse History		Generalized anxiety disorder	2.5%
Physical abuse (ages 6-13)	4.1%	Panic disorder	2.1%
Sexual abuse	6.7%	Social phobia	8.9%
Gender: female	51.5%

2.4 Data Analysis

Three models were constructed to determine the role of psychiatric and familial risk factors in predicting 1) time to first cigarette, 2) time from first cigarette to onset of regular smoking, and 3) time from regular smoking to onset of ND. Given that not all of the offspring had lived through the period of risk for the development of ND, Cox proportional hazards models were chosen to conduct multivariate regression with time-to-event data. Thus, all data was utilized up until the time of censorship (i.e., time of interview) when calculating hazard ratios. Analyses were conducted with STATA, version 8.2 (Statcorp, 2005). Standard deviations of means and confidence intervals for Cox regression models were adjusted for family clustering using Huber-White robust standard errors. The Efron approximation (Hosmer & Lemeshow, 1999) was used to handle ties. With the exception of mother-reported data (i.e., offspring ODD and ADHD diagnoses and maternal psychopathology), complete data were available for nearly all participants (rates of missing data were below 1.5% for variables based on father and offspring reports). Missing data were deleted listwise in Cox regression analyses.

Three dummy variables representing the four AD risk groups (with AD risk group 4 as the common reference group) were included in the models to account for sampling design. Three dummy variables representing the four ND risk groups (with ND risk group 4 as the common reference group) were also included in the models with the maternal ND variable to adjust for genetic and environmental contributions to smoking outcomes associated with paternal and maternal ND status. Of note, potential covariates included risk factors that varied with regard to age of onset such that onset preceded the age at which a given smoking milestone was reached in some cases but not others (e.g., AD and ND onset). In order to accurately model the contribution of these factors to risk for smoking outcomes, “person year data” was created using SAS, version 9.1 (SAS Institute Inc, 2002) to represent time-varying covariates. Data were constructed with one line of data representing a single year of life for each case. For cases that were positive for a given risk factor, corresponding variables were coded as absent in each year up to the age of onset of the risk factor and present from the year of onset onwards. Time-varying covariates were used for offspring sexual abuse history, substance use disorders, and psychiatric disorders. The assessment of parenting behaviors and physical abuse specifically referenced the period from age 6 to 13 years and parental psychiatric and SUDs were lifetime diagnoses. As a result, parental psychopathology, physical abuse history, and parenting behaviors were time-invariant.

Each of the covariates listed as predictors was tested with maternal ND and the six dummy variables representing paternal AD and ND risk group in two stages. First, covariates were tested by domain (see Table 1). For example, conduct disorder (CD) was tested in combination with the other two offspring externalizing disorders. If the hazard ratios were statistically significant (i.e., confidence intervals did not include 1.00), covariates were retained and then combined with significant covariates from the other domains assessed. The reduced model was derived through manual backward deletion. Covariates that were no longer significant in the context of other predictors were removed one at a time to produce the final model.

To model rate of transition from first cigarette to onset of regular smoking, two dummy variables were created to represent age at first cigarette in order to adjust for variability in proximity of age at first cigarette to the risk period for development of regular smoking. (Chi-square analyses conducted with three levels of age at first cigarette and three levels of transition time to regular smoking were highly significant; χ² = 77.75; p<.001). The two dummy variables reflected the lowest and highest thirds of the distribution for age at first cigarette, with median age (14-15 years) as the common reference group. “Early age at first cigarette” was defined as initiating smoking at age 13 or younger and “late age at first cigarette” as initiating smoking at age 16 or older. Rate of transition from regular smoking to onset of ND did not differ significantly by age at first cigarette (based on Chi-square analyses conducted with three levels of transition time and three levels of age at first cigarette; χ² = 8.26; p>.05), so adjustments for age at smoking initiation were not made to the third model.

The proportional hazards assumption that risk remains constant over time was assessed using the Grambsch and Therneau test of the Schoenfeld residuals (Grambsch & Therneau, 1994). The proportional hazard assumption was met for the third model (time from regular smoking to onset of ND) but was violated in the first and second models (time to first cigarette and time from first cigarette to onset of regular smoking). In the first model, maternal ND violated the proportional hazards assumption. To adjust the model, the distribution for years at risk was split into three approximately equal divisions based on frequency: 0-13, 14-15, and 16 or more years, and interaction terms were created between each of the three divisions with maternal ND status. Interaction terms were entered into the model one at a time to test for changes in the proportional hazard estimations. The addition of the interaction term representing maternal ND by the risk period from 14-15 years of age resolved the violation of the proportional hazard assumption. The same procedure was followed to adjust proportional hazard violations for alcohol use and early age at first cigarette in the second model. The distribution of years at risk from first cigarette to onset of regular smoking was split by frequency into three divisions: 0-1, 2-3, and 4 or more years and interaction terms created with the two covariates. Violations were corrected by including interaction terms representing early age at first cigarette with the 0-1 year risk period and early age at first cigarette with the 2-3 year risk period in the model.

3. Results

3.1 Smoking milestones and transition times

The majority of participants had smoked one or more cigarettes during their lifetimes (64.5%; n = 817), 26.2% (n = 332) of the total sample progressed to regular smoking, and 9.6% (n = 122) met lifetime criteria for ND. First cigarette was smoked on average at 14.17 (S.D. = 2.96) years of age, regular smoking began at a mean age of 16.62 (S.D. = 2.31) years, and mean onset of ND occurred at 18.40 (S.D. = 2.69) years of age. Rate of transition from first cigarette to regular smoking ranged from 0 to 11 years, with approximately half of regular smokers reporting 2 or fewer years from first cigarette to the onset of regular smoking. Transition times from regular smoking to ND onset covered a comparable range (0 to 10 years), but greater than two thirds reported 2 or fewer years from the start of regular smoking to the development of ND.

3.2 Predictors of age at first cigarette

Results from the Cox proportional hazards regression model predicting age at first cigarette are shown in Table 2. After adjusting for risk conferred by paternal ND and AD status and maternal ND, offspring use of other substances predicted age at initiation of smoking. Cannabis and alcohol use were associated with nearly two and a half fold increases in the rate of onset for cigarette smoking. ODD (HR: 1.69) and maternal AD (HR: 1.37) also predicted early age at first cigarette, as did both maternal ND (HR: 1.08) and paternal ND (ND risk group 1) (HR: 1.22). The interaction term representing maternal ND by the risk period 14-15 years of age, included in the model to adjust for proportional hazards violations, was associated with a slower rate of onset for first cigarette (HR: 0.88), indicating that the influence of maternal ND on smoking initiation was diminished during that portion of the risk period.

Table 2. Cox proportional hazards model for time to first cigarette (n = 1,269).

	Hazard Ratio (95% CI)
Cannabis use*	2.50 (1.99 - 3.15)
Alcohol use*	2.42 (1.99 - 2.94)
Oppositional defiant disorder*	1.69 (1.34 - 2.13)
Maternal alcohol dependence*	1.37 (1.10 - 1.70)
Nicotine dependence risk group (based on paternal status)a
Group 1: high genetic (G), high environmental (E)*	1.22 (1.01 - 1.49)
Group 2: high G, low E	1.14 (0.80 - 1.62)
Group 3: moderate G, low E	1.05 (0.79 - 1.40)
Alcohol dependence risk group (based on paternal status)a
Group 1: high G, high E	1.07 (0.87 - 1.33)
Group 2: high G, low E	1.12 (0.86 - 1.47)
Group 3: moderate G, low E	1.03 (0.79 - 1.35)
Maternal nicotine dependence (Fagerstrom proxy)	1.08 (1.02 - 1.16)
Maternal ND X risk period from 14-15 years of age b*	0.88 (0.79 - 0.98)

^aHazard ratios relative to Group 4 (control)

^bInteraction term included in the model to adjust for violation of proportional hazard assumption for maternal nicotine dependence.

^*p <.05

3.3 Predictors of transition time from first cigarette to regular smoking

In the second regression model, which estimated rate of progression from first cigarette to regular smoking, other substance use and externalizing problems were also associated with an elevated rate of transition, even after adjusting for age at first cigarette, paternal ND and AD status, and maternal ND (see Table 3). Cannabis use (HR: 2.38) was the most potent predictor of a rapid transition from first cigarette to regular smoking, followed by CD (HR: 1.64) and alcohol use (HR: 1.56). Paternal ND was associated with an increased rate of transition as well (HR: 1.66). The two interaction terms added to the model to adjust for violations of the proportional hazard assumption, which represented early age at first cigarette by the risk periods up to 1 year and 2-3 years after first cigarette, predicted increased rate of progression to regular smoking (HR: 2.68 and 6.51, respectively). That is, the effects of early first cigarette on rapid transition to regular smoking were most pronounced in the first 3 years after smoking initiation.

Table 3. Cox proportional hazards model for transition time from first cigarette to onset of regular smoking (n = 817).

	Hazard Ratio (95% CI)
Cannabis use*	2.38 (1.83 - 3.11)
Conduct disorder*	1.64 (1.15 - 2.34)
Alcohol use*	1.56 (1.12 - 2.18)
Age at first cigarettea
Early (<= 13 years)*	0.43 (0.25 - 0.73)
Late (>16 years)	0.76 (0.54 - 1.05)
Nicotine dependence risk group (based on paternal status)b
Group 1: high G, high E*	1.66 (1.20 - 2.29)
Group 2: high G, low E	1.62 (1.00 - 2.61)
Group 3: moderate G, low E	1.36 (0.89 - 2.10)
Maternal nicotine dependence (Fagerstrom proxy)	1.05 (0.97 - 1.13)
Alcohol dependence risk group (based on paternal status)b
Group 1: high G, high E	0.81 (0.61 - 1.07)
Group 2: high G, low E	0.96 (0.66 - 1.41)
Group 3: moderate G, low E	0.91 (0.62 - 1.35)
Early first cigarette X risk period up to 1 year after first cigarettec*	2.68 (1.40 - 5.13)
Early first cigarette X risk period 2-3 years after first cigarettec*	6.51 (3.17 - 13.32)

^aHazard ratios relative to median age at first cigarette (14-15 years)

^bHazard ratios relative to Group 4 (control)

^cInteraction terms included in the model to adjust for violation of proportional hazard assumption for alcohol use and for early age at first cigarette.

^*p <.05

3.4 Predictors of transition time from regular smoking to onset of ND

Results of the regression model predicting rate of progression from regular smoking to onset of ND are summarized in Table 4. Consistent with the models depicting initiation of cigarette smoking and transition to regular smoking, externalizing disorders were associated with elevated risk for a rapid transition from regular smoking to onset of ND. Hazard ratios were 3.62 and 2.32 for CD and ODD, respectively, but it was the internalizing disorders that demonstrated the largest effects. The most robust predictor of rapid escalation to ND was panic disorder, with a hazard ratio of 13.28. Greater than four-fold increases in rate of progression from regular smoking to ND were also found for both major depressive disorder and social phobia.

Table 4. Cox proportional hazards model for transition time from regular smoking to onset of nicotine dependence(n = 332).

	Hazard Ratio (95% CI)
Panic disorder*	13.28 (4.12 - 42.75)
Social phobia*	4.78 (2.44 - 9.37)
Major depressive disorder*	4.31 (1.80 - 10.31)
Conduct disorder*	3.62 (2.11 - 6.23)
Oppositional defiant disorder*	2.32 (1.32 - 4.06)
Nicotine dependence risk group (based on paternal status)a
Group 1: high G, high E	0.88 (0.47 - 1.64)
Group 2: high G, low E	0.36 (0.13 - 1.00)
Group 3: moderate G, low E	0.70 (0.23 - 2.16)
Maternal nicotine dependence (Fagerstrom proxy)	0.92 (0.77 - 1.10)
Alcohol dependence risk group (based on paternal status)a
Group 1: high G, high E	0.81 (0.40 - 1.63)
Group 2: high G, low E	1.04 (0.42 - 2.57)
Group 3: moderate G, low E	1.42 (0.60 - 3.38)

^aHazard ratios relative to Group 4 (control)

^*p <.05

4. Discussion

In the current study, psychiatric and familial risk factors for escalation in smoking behaviors were examined across three stage transitions using an offspring-of-twins design, thus extending the current literature in two ways. First, the adjustments for maternal ND and for paternal ND and AD risk conferred by genetic and environmental exposures provided a robust test for the association of psychiatric disorders and the family environment with transitioning to greater tobacco involvement. Second, by characterizing risk for rapid movement between stages the current investigation captured an aspect of developmental course that is not addressed when simply predicting whether a stage of ND development is ever reached.

The examination of three key transition points in the development of ND provided the opportunity to address continuity versus specificity of psychiatric and familial risk across stage transitions in the progression to greater tobacco involvement. Similar to investigations examining predictors of reaching a given smoking milestone, findings from the current study indicated that certain risk factors are associated with speed of transitions across multiple stages of development of the disorder, whereas others impact the rate of escalation only during specific stages.

4.1 Cross-stage predictors of transitions

Results produced evidence of continuity in risk conferred by externalizing disorders for rapid transitions across the three smoking stages: ODD predicted early initiation, CD was associated with escalation to regular smoking, and both had significant effects on the transition to ND. Childhood antisocial behavior has been linked to early experimentation with cigarettes in numerous previous investigations (Clark & Cornelius, 2004; Dierker et al., 2001; Rohde et al., 2004; van den Bree et al., 2004) and elevated rates of regular smoking and ND among individuals with externalizing psychopathology are well documented (Disney et al., 1999; Galera et al., 2004; Kollins et al., 2005; Tercyak et al., 2002). The current findings suggest that externalizing disorders impact yet another dimension of ND development: the rate of progression between smoking stages. Furthermore, as the study design included control for genetic factors associated with ND and AD, our results indicate that this association is not fully accounted for by genetic risk shared with these disorders.

Consistent with previous studies demonstrating an association between alcohol consumption and early age at first cigarette (Grant, 1998; Rohde et al., 2003), earlier onset of smoking initiation was predicted by alcohol use, as was rapid transition to regular smoking. Although rarely investigated in relation to the rate of progression across stages of tobacco involvement, there is a large literature tying alcohol use to smoking (Grucza & Bierut, 2006; Dierker et al., 2004; Koopmans et al., 1997; Weitzman & Chen, 2005) that includes evidence of comorbid alcohol use and smoking trajectories spanning young adulthood (Jackson et al., 2005). Shared heritability (adjusted for in the current investigation) is among the best documented risk factors (Bierut et al., 2004; Connor et al., 2007; True et al., 1999; Young et al., 2006). Thus, though the link between alcohol use and smoking is predictable, the robustness of these findings is noteworthy. Associations were found in analyses that accounted for maternal ND in addition to paternal genetic and environmental contributions to AD and ND and used time-varying covariates to model order of onset for smoking and alcohol use behaviors.

Cannabis use produced a similar pattern of results. It was associated with earlier age at first cigarette and rapid transition from first cigarette to regular smoking. Findings were consistent with the increased probability of experimentation with cannabis found in smokers (Ellickson et al., 2004a; Faeh et al., 2006) and the high rates of smoking among cannabis users (Degenhardt et al., 2001; Swift et al., 2001; Reid et al., 2000), but added to this literature by portraying the role of cannabis use in escalating tobacco involvement. Although suggestive of an atypical sequence of substance involvement, with tobacco use generally believed to precede initiation of cannabis use (Kandel, 1975; Kandel et al., 1992), findings are compatible with a growing literature demonstrating that the sequence of substance use is not universal (Amos et al., 2004; Guerra et al., 2000; Wetzels et al., 2003). Our results parallel recent evidence that cannabis use predicts both smoking initiation and progression to ND (Patton et al., 2005; Timberlake et al., 2006).

The absence of significant findings for alcohol and cannabis use (or alcohol/cannabis use disorders) on the rate of transition from regular smoking to ND merits comment as well. Rather than indicating that they are unrelated to the rate of escalation to ND, the lack of effects may reflect the absence of unique variance accounted for by alcohol and cannabis use - independent of the variance accounted for by genetic risk for ND and AD. An increase from early to later stage transitions in the overlap of variance in genetic risk for ND and AD with variance in alcohol and cannabis use would also explain this finding. Evidence for changes in the source of genetic contributions to smoking outcomes across stages (i.e., heritability of smoking initiation is not perfectly correlated with heritability for progression to heavier use [Fowler et al., 2006] or ND [Kendler et al., 1999; Maes et al., 2004]) lends support to this interpretation of the results.

4.2 Stage-specific predictors of transitions

Maternal AD predicted early smoking initiation, but was not associated with the speed of later stage transitions, paralleling findings from an earlier investigation of alcohol stage transitions conducted by our group (Sartor et al., 2007). The emergence in the final stage transition of internalizing disorders as accelerators of the shift from regular smoking to ND, by contrast, has rarely been reported and is best interpreted in the context of the larger literature on mood and anxiety disorders as they relate to smoking behaviors.

Elevated rates of regular smoking and ND have been found for individuals with affective and anxiety disorders in a number of studies (Breslau et al., 1991; Fergusson et al., 2003; John et al, 2004; Patton et al., 1996), but some reports have indicated that the association disappears when examined in the context of additional psychiatric and SUDs (Black et al., 1999; Grant et al., 2004; Rohde et al., 2004). Investigations involving multiple stages of tobacco involvement suggest that inconsistencies across studies may be attributable in part to differences in the operationalization of the smoking phenotypes examined. In a longitudinal study spanning adolescence to young adulthood for example, social phobia was associated with increased likelihood of transitioning from regular smoking to ND but not of progressing to regular smoking (Sonntag et al., 2000). Persistent depression and anxiety similarly predicted progression to ND but not onset of daily smoking in an Australian sample followed for 10 years (Patton et al., 2006). A large, population-based study in Germany also revealed clear distinctions in the role of internalizing disorders for initiation of smoking versus the transition to dependent smoking status. Across mood and anxiety disorders, effects were non-significant for the earlier stage transition and significant for the transition from regular smoking to ND (Schumann et al., 2004). The convergence of findings is striking and suggestive of a later stage specificity in the impact of internalizing disorders on progressive tobacco involvement, but given the limited literature in this area, conclusions are best viewed as preliminary. Moreover, the approach used in the current study is distinct from that used in prior literature in its characterization of the speed of progression through stages, making direct comparisons to previous findings difficult. A complete account of mechanisms underlying the stage transitions described in our results is beyond the scope of this study, but the integration of these findings into an etiological model of the development of the disorder and an exploration of their clinical implications are clearly warranted.

4.3 Clinical implications

The findings from the current study portray a three-stage pathway from smoking initiation to the onset of ND. The beginning of the pathway, in which externalizing behaviors and other substance use are tied to early smoking initiation and a rapid movement to regular smoking, is frequently described in the literature as a problem behavior syndrome (Donovan & Jessor, 1985; Jessor & Jessor, 1977) that is perpetuated by environmental exposures and responses to antisocial behaviors, e.g., association with deviant peers (Ary et al., 1999; Dishion et al., 2004; Dodge & Pettit, 2003). By contrast, the prominent role of internalizing disorders in combination with the continued influence of externalizing disorders on escalation from regular smoking to ND suggests that the mechanisms underlying this later stage transition differ from (or are more diverse than) those that contribute to the early stages of tobacco involvement. The clinical implications of these findings are two-fold. First, the pathway of progressive tobacco involvement is determined by modifiable risk factors in addition to genetics. The risk factors described in our results were identified after stringently controlling for paternal and maternal genetic contributions to ND (as well as paternally transmitted genetic risk for AD). Furthermore, the distinctions in risk profiles across stage transitions underscore the importance of ongoing assessment and treatment of comorbid disorders that shape the developmental course of tobacco involvement. Those individuals at risk for progression early on in the course of ND development (e.g., behaviorally undercontrolled adolescents who drink and use cannabis) may not be the same as those who move rapidly from regular smoking to ND (e.g., anxiety-prone or depressed youth).

4.4 Limitations and future directions

The current study has some limitations, which suggest possible directions for future research in this area. First, as the sample is relatively well-educated and largely Caucasian, replication with more diverse samples would be needed to determine whether variability by socioeconomic status (Jerfferis et al., 2003; Tyas et al., 1998) and ethnicity (Blitstein et al., 2003; Ellickson et al., 2004b) in smoking outcomes extend to distinctions in psychiatric and familial risk factors for rate of stage transitions. Second, although bias was minimized through the use of survival analytic strategies, the sample included adolescents not yet through the age of risk for ND, so findings may more accurately characterize smoking transitions among early onset cases of ND. Third, smoking histories were reported retrospectively, and though covering a relative short period of time, prospective assessments would likely produce more reliable accounts. Fourth, a broad range of psychiatric risk factors encompassing the most common forms of psychopathology were examined, but not every disorder associated with smoking behaviors was assessed (e.g., psychotic disorders, Hughes et al., 1986; McCreadie et al., 2002). Fifth, findings suggest that treating co-morbid disorders may have an impact the course of ND development, but direct tests of this hypothesis have yet to be conducted. Finally, the present investigation was focused on familial and psychiatric correlates of smoking stage transitions and did not address friends’ or siblings’ influences on smoking. The integration in future studies of stage-specific peer smoking behaviors would create the opportunity to explore the interaction of peer and environmental selection with genetic risk for ND in the progression of smoking behaviors.