Psychosomatic Medicine
HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrowRequest Permissions
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Rodriguez, D.
Right arrow Articles by Audrain-McGovern, J.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Rodriguez, D.
Right arrow Articles by Audrain-McGovern, J.
Related Collections
Right arrow Depression
Psychosomatic Medicine 67:200-210 (2005)
© 2005 American Psychosomatic Society

ORIGINAL ARTICLES

Developmental Heterogeneity in Adolescent Depressive Symptoms: Associations With Smoking Behavior

Daniel Rodriguez, PhD, Howard B. Moss, MD and Janet Audrain-McGovern, PhD

From the Department of Psychiatry, University of Pennsylvania, Philadelphia, PA.

Address correspondence and reprint requests to Daniel Rodriguez, Department of Psychiatry, University of Pennsylvania, 3535 Market Street, Suite 4100, Philadelphia, PA 19104. E-mail: drodrig2{at}mail.med.upenn.edu


   ABSTRACT
TOP
 ABSTRACT
INTRODUCTION
 METHODS
 RESULTS
 DISCUSSION
 CONCLUSION
 NOTES
 REFERENCES
 
Objective: Previous research has indicated an association between smoking and depression in adolescents, although the nature of the relationship is controversial. We sought to understand this relationship better in a prospective study by investigating whether there are subpopulations of adolescents with different relationships between smoking and depressive symptoms.

Methods: Our sample was 925 adolescents attending one of five Northern Virginia high schools, grades 9 to 12. We used General Growth Mixture Modeling as our method because it allowed identification and characterization of depressive symptoms trajectories and assessment of the effects of trajectory on 12th grade smoking. We defined the binary variable 12th grade current smoking as smoking on 1 or more of the past 30 days and more than 100 cigarettes smoked in a lifetime, versus not having smoked in the past 30 days.

Results: We identified three trajectories: high, medium, and low depressive symptoms. For adolescents with high symptoms, 9th grade (baseline) smoking was associated with an overall deceleration of depressive symptoms, whereas for adolescents with moderate symptoms, baseline smoking was associated with an overall acceleration in depressive symptoms. Baseline smoking was not associated with rate of change in depressive symptoms for adolescents with low symptoms, nor was it associated with baseline depressive symptoms in any trajectory.

Conclusion: These findings demonstrate that there is a relationship between smoking and depressive symptoms in adolescents, and that the relationship can vary by developmental trajectory, suggesting etiological heterogeneity.

Key Words: depression • smoking • trajectories • General Growth Mixture Modeling

Abbreviations: GGMM = General Growth Mixture Modeling; CES-D = Center for Epidemiological Studies Depression; GMM = Growth Mixture Modeling; BIC = Bayesian Information Criterion; LMR LRT = Lo-Mendell-Rubin Likelihood Ratio; OR = odds ratio; CI = confidence interval.


   INTRODUCTION
TOP
 ABSTRACT
 INTRODUCTION
METHODS
 RESULTS
 DISCUSSION
 CONCLUSION
 NOTES
 REFERENCES
 
Tobaccosmoking is the leading preventable cause of death and disease in the United States, resulting in more than 400,000 premature deaths per year (1). Eighty percent of adult smokers started smoking before age 18 years, and nearly 5000 adolescents a day try their first cigarette (1). In order to inform prevention and intervention efforts, it is essential to identify factors that place adolescents at greater risk of initiating smoking and that facilitate progression to regular smoking.

One factor that influences smoking in adolescents is depression. Cross-sectional research has found significant comorbidity between depression and cigarette smoking in adolescents (2–8). For instance, the odds of smoking and being nicotine-dependent were shown to be 2 to 4 times greater for depressed adolescents than for nondepressed adolescents (5,6).

Prospective studies support the cross-sectional findings, although the direction of the relationship is unclear. Some studies have found that depression predicts smoking (9,10), whereas others have found that smoking precedes depression in adolescents (11–14). The findings of other studies support a bidirectional relationship between smoking and depression in adolescents (15–19). For example, one study found serious and persistent depression at baseline predicted cigarette smoking 2 years later. In addition, heavy and persistent smoking at baseline predicted depressive symptoms at 2-year follow-up (19).

Thus, the nature of the relationship between smoking and depression in adolescents remains unclear. The discrepant findings may be caused, in part, by the existence of latent subpopulations of adolescents with different patterns (trajectories) of depression across time. To date, no study has assessed the possibility of multiple latent trajectories of depression and how they relate to smoking, although research findings do suggest different trajectories of depression for adolescent boys and girls, and for adolescents differing on factors like income, attribution style, and stress (20–23). Essentially, previous research has assumed a single population (trajectory) of adolescents when assessing the relationship between smoking and depression. Although more recently, studies have modeled developmental heterogeneity in the assessment of this relationship (19), they nevertheless assume that extraneous variables (eg, previous depression, previous smoking) influence adolescents equally (a single trajectory assumption), despite the presence of developmental heterogeneity (24). Further, those studies suggesting multiple trajectories of depressive symptoms (eg, 22) have modeled trajectories based on known differences in the effects of covariates on adolescent depression (eg, gender), instead of empirically testing the fit of models with different numbers of trajectories to the data.

The present study sought to understand better the relationship between smoking and depression based on the assumption of multiple but latent subpopulations of adolescents with different developmental trajectories of depressive symptoms. Adolescents with different trajectories of depressive symptoms may exhibit different smoking behavior at baseline and follow-up. For instance, 9th grade smoking may be associated with increasing depressive symptoms for some adolescents and decreasing depressive symptoms for others, or it may not be associated with depressive symptoms at all. Further, different depressive symptoms trajectories may be associated with different levels of 12th grade smoking. Thus, the goal is to identify different depressive symptoms trajectories, assess their relationship to 9th and 12th grade smoking, and establish the characteristics, if any, that discriminate trajectory membership. This would have implications for smoking prevention and cessation efforts depending on which characteristics affect an adolescent’s smoking behavior. The analysis of different trajectories of depressive symptoms may also help us understand the positive, negative, and bidirectional relationships between smoking and depression found in previous studies.

In order to address this issue, we used a longitudinal sample of adolescents followed from the 9th to the 12th grade (ages 14–18 years), yielding five waves of data. To assess the possibility of multiple trajectories, we used a factor mixture modeling method, General Growth Mixture Modeling (GGMM), which allowed us to assess intraindividual variability in depressive symptoms, classify adolescents into their most likely trajectory, and characterize adolescents on covariates (25–27). This approach also permitted us to assess the effects of trajectory membership on 12th grade smoking as a binary distal outcome. We hypothesized the existence of multiple trajectories of depressive symptoms, representing discrete subpopulations of adolescents. In addition, we hypothesized that adolescents belonging to different trajectories would be distinguishable on covariates related to smoking and depressive symptoms. The overall aim of this study was to apply a novel analytic method to the question of the relationship between smoking and depressive symptoms in a nonclinical sample of adolescents.


   METHODS
TOP
 ABSTRACT
 INTRODUCTION
 METHODS
RESULTS
 DISCUSSION
 CONCLUSION
 NOTES
 REFERENCES
 
Participants and Procedures
Participants were high school students (52% female and 64% Caucasian) taking part in a longitudinal study of the biobehavioral predictors of adolescent smoking adoption. Participants were enrolled in five public high schools in northern Virginia. This cohort was drawn from the 2393 students identified through class rosters at the beginning of 9th grade. Students were ineligible to participate in this study if they had a special classroom placement (ie, English as a second language and/or severe learning disability). Based on the selection criteria, a total of 2120 (89%) students were eligible to participate. Of the 2120 eligible students, 1533 (72%) parents provided a response. Of these 1533, a total of 1151 (75%) parents consented to their teen’s participation in the study, yielding an overall consent rate of 54%. An analysis of differences between students whose parents did and did not consent to participation in the study revealed a race by education interaction. The interaction indicated that the likelihood of consent was more than twice as great for Caucasian parents with more than a high school education than for Caucasian parents with a high school education or less (28). Although overall difference was small, some caution in generalizing these findings is suggested. Participation in the study required both student and parental consent. The final sample size available for this study based on complete data for the four waves was 925. University Institutional Review Board approval of the study protocol was obtained.

The cohort was formed in the 9th grade and was followed until the end of the 12th grade. Five data collection waves were completed: spring 2000, 9th grade; fall 2000, 10th grade; spring 2001, 10th grade; spring 2002, 11th grade; and spring 2003, 12th grade. Data were collected on-site, during compulsory classes (eg, health, science). The research team distributed the self-administered paper questionnaire. Completed questionnaires contained identification numbers only. Instructions were read aloud by a research team member, emphasizing confidentiality, and questions were encouraged. The questionnaires took approximately 30 minutes to complete. We assessed depressive symptoms at the four spring waves.

As indicated, the analysis involved 925 participants with complete data on the four repeated measures of depressive symptoms, and covariates. The assumption in complete case modeling (listwise deletion) is that the complete cases are a random sample of the entire sample. Essentially, data are missing completely at random (29). We tested the assumption of missing completely at random with a likelihood ratio {chi}2 test under the unrestricted latent class indicator model, a feature in Mplus 3.1. The results provided support for the missing data being completely at random ({chi}2[321, N = 1138] = 198.63; p=1.00), which indicated that the data can be considered a random sample of the entire sample. In addition, the results using complete cases did not differ qualitatively from an analysis with all available data (a form of pairwise deletion).

Instrumentation
Assessment of Depressive Symptoms
Depressive symptoms were assessed with the Center for Epidemiological Studies Depression (CES-D) inventory. The CES-D is a 20-item self-report measure designed to assess depressive symptoms in the general population (30,31). Items on the CES-D use a 4-point scale to indicate how frequently in the past week each symptom occurred (0 = none of the time, 1 = a little of the time, 2 = a moderate amount of time, 3 = most of the time). Research supports the validity and reliability of the CES-D for use with high school adolescents as a measure of depressive symptoms (18,31,32). Total CES-D scores tend to be higher for adolescents than adults; thus, the suggested clinical cutoff in adolescents is 22 for males and 24 for females, compared with 16 for adults (18). Previous research has used the CES-D to assess depressive symptoms in adolescents and their relationship with adolescent smoking behavior (7,19,33–35).

Assessment of Smoking
Baseline adolescent smoking behavior was summarized in an ordered-categorical variable with five categories representing increasing levels of smoking. The variable was generated from responses to a series of standard epidemiological questions taken from the Youth Risk Behavior Surveillance (36) regarding smoking such as, "Have you ever tried or experimented with cigarette smoking, even a few puffs?" and "Have you smoked a cigarette in the past 30 days?" The five ordered categories are as follows: 0 = never smoker; 1 = puffer (not ever having smoked a whole cigarette); 2 = experimenter (having smoked a whole cigarette but ≤100 cigarettes total in a lifetime); 3 = current smoker (smoked 1–19 days in the last 30 days and >100 cigarettes in a lifetime); and 4 = frequent smoker (≥20 days smoked in the last 30 days and >100 cigarettes in a lifetime). Adolescents who smoked more than 100 cigarettes in a lifetime but who had not smoked in the last 30 days were classified as experimenters (N = 4). This categorization is entirely consistent with the literature on adolescent smoking (37–39). Twelfth grade smoking is a proposed indicator of the latent depressive symptoms categorical variable. Twelfth grade smoking was assessed with a binary outcome (1 = current smoker, smoking 1 or more of the past 30 days, and >100 cigarettes smoked in a lifetime; 0 = else). Research supports the reliability of Youth Risk Behavior Surveillance items assessing smoking practices with {kappa} coefficients in the substantial or higher range ({kappa} ≥ 61%) (40,41). Research also supports the validity of self-report measures of smoking behavior in adolescents, particularly in nontreatment contexts in which confidentiality is emphasized (42,43).

Assessment of Covariates
Demographic variables assessed included gender (0 = male; 1 = female) and race (0 = Caucasian; 1 = non-Caucasian). Several covariates were included because research has indicated a relationship to cigarette smoking and depression in adolescents (9,18,19,33,44–47,49–50). The covariates included alcohol and marijuana use, academic performance (grades), nonsport extracurricular activity, peer smoking and household smoking, physical activity, and team sport participation. All covariates were assessed at baseline (9th grade).

Alcohol use was assessed with a single item that asked, "During your life, on how many days have you had at least one drink of alcohol?" Scores ranged from 1 = 0 days to 7 = 100 or more days. Marijuana use was assessed with a single item that asked, "During your life, how many times have you used marijuana?" Scores ranged from 1 = 0 times to 7 = 100 or more times. Grades were assessed with a single item that asked, "How do you do in school?" Scores ranged from 1 = mostly A’s to 4 = mostly D’s and F’s. This item was reverse scored so that higher scores would represent better grades. Extracurricular activity was assessed summing two items on a 5-point scale, 1 = none to 5 = four or more, representing level of activity in school, not including sport: "How many clubs are you involved in? (Example: drama, debate, foreign language, computer, band);" and "How many activities? (Example: student government, newspaper, yearbook, homecoming committee)." Scores could range from 2 to 10, representing increasing levels of activity. Peer smoking was assessed with a dichotomous item (1 = has at least one peer smoking; 0 = no peers smoking), generated from responses to items asking whether adolescents have any friends who currently smoke: "Does your best friend smoke?" "Do any of your other best male friends currently smoke?" "Do any of your other best female friends currently smoke?" Household smoking was assessed with a single dichotomous item (1 = yes; 0 = no), "Does anyone living in your household smoke?" Average physical activity was assessed summing three items that requested intensity, duration, and frequency of physical activity, and dividing by 3: scores on physical activity ranged from 0 (no physical activity) to 7 (high physical activity). These items are described thoroughly elsewhere (51). Team sport participation was assessed with a single 4-point scale item that requested the number of teams on which the individual played during the past 12 months, including those run by the "... school or community group" (1 = zero teams; 2 = one team; 3 = two teams; 4 = three or more teams).

Statistical Analyses
Data analysis involved GGMM. Growth Mixture Modeling (GMM) is a factor mixture modeling technique that models a mixture of continuous and categorical latent variables (24). The continuous latent variables define growth within classes with factors for baseline level and trend. The latent categorical variable defines the number of unobserved developmental trajectories. GMM permits estimation of trajectory shapes (eg, linear, quadratic), trajectory classification probabilities for each participant (posterior probabilities), class-specific growth parameter variance, and regression of the latent trajectory class variable on covariates for trajectory characterization (27,52,53). Characterization allows for identification of the most likely members of a given trajectory in relation to a comparison trajectory (generally the most common trajectory, or the trajectory with mean values closest to zero, such as low depressive symptoms) with multinomial logistic regression. GGMM is a special case of GMM in which a binary variable (a distal outcome) is regressed on the latent trajectory variable and select covariates to improve model specification (27).

A critical issue in GMM is model selection. Model selection in GMM entails selecting the number of trajectories that best represents the behavior of a dependent variable (eg, depressive symptoms) in the population. Although there are no definitive criteria for selecting the optimal number of trajectories in factor mixture modeling, the Bayesian Information Criterion (BIC) is suggested as a guide (24,54,55). Low BIC values reflect model parsimony, favoring a high log likelihood estimate along with a low number of parameters.

A second useful criterion is the Lo-Mendell-Rubin Likelihood Ratio Test (LMR LRT). The LMR LRT tests significance in the –2 times Loglikelihood difference between the model with k and k – 1 (H0) classes (27,52). Use of the LMR LRT statistic for model selection with Mplus is based on the assumption of within-class conditional normality given the covariates, an assumption that is particularly useful when modeling with nonnormal covariates (eg, gender) (27). In addition, Mplus 3.1 provides an entropy summary statistic to assess classification quality, with values ranging from 0 to 1, and values closer to 1 representing good classification quality (52). The most widely accepted rule is to add trajectories as long as the BIC continues to decrease and additional trajectories are substantively meaningful (56). We used the combination of these criteria to determine the optimal number of trajectories. The analysis was conducted with an assumption of equality of growth parameter variance between classes for determination of the optimal number of trajectories. However, this assumption was relaxed to improve model specification in the final analysis by freeing growth parameter variances within classes. All analyses in this study were conducted with Mplus 3.1 software (57).


   RESULTS
TOP
 ABSTRACT
 INTRODUCTION
 METHODS
 RESULTS
DISCUSSION
 CONCLUSION
 NOTES
 REFERENCES
 
Descriptive Statistics
Means and SDs for the continuous covariates in the model appear in Table 1. The proportions for each categorical covariate in the model for the entire sample appear in Table 2. At baseline, only 4% of adolescents were smoking currently. By 12th grade, 14% of adolescents had attained this level of smoking. The change was significant ({chi}2[4,925] = 209.45; p < .0001).


View this table:
[in this window]
[in a new window]
  		TABLE 1. Bivariate Correlations for Depressive Symptoms, Grades 9 to 12, and Covariates (N = 925)

 

View this table:
[in this window]
[in a new window]
  		TABLE 2. Frequency Distributions for Categorical Covariates in the Model, for the Entire Sample and by Trajectory

 

Latent Growth Model
The Latent Growth Model (measurement growth curve model without covariates) fit the data well with a quadratic trend ({chi}2[1,971] = 0.08; p = .78; CFI = 1.00; RMSEA = 0.00 [0, 0.06], SRMR = 0.002). The linear trend was significant and positive ({eta}1 = 1.42; z = 4.21; p < .0001), and quadratic trend was significant and negative ({eta}2 = –0.47; z = –4.53; p < .0001), indicating initial growth (waves 1 and 2) followed by a period of declining symptoms (waves 3 and 4). Intercept was also significant ({eta}0 = 13.56; z = 46.06; p < .0001). Intercept and linear trend variances were significant (p < .0001), indicating significant variability in initial depressive symptoms level and linear trend. However, the quadratic trend variance was not significant (p > .05); thus, it was constrained to zero in the subsequent analyses. Table 1 presents the correlation matrix for depressive symptoms, grades 9 to 12, 12th grade smoking, and the covariates, along with their means, SDs, and Cronbach’s coefficient {alpha} reliability estimates for the repeated measures of the CES-D.

Growth Mixture Modeling
We conducted GMM without covariates to test for the existence of subpopulations of depressive symptoms in adolescents. We tested five models, beginning with a single-trajectory model, to determine the optimal number of trajectories of depressive symptoms. Model fit statistics are presented in Table 3. The BIC value decreased from BIC = 27390.974 in the single-trajectory model to BIC = 27064.055 in the four-trajectory model. However, the BIC increased with the addition of a fifth trajectory, BIC = 27077.617. The LMR LRT estimate was not significant for the five-trajectory model (p = .12), suggesting the four-trajectory model. Therefore, using a combination of model fit indices, we chose the four-trajectory model as the most parsimonious but informative description of our data, modeling without covariates at this stage.


View this table:
[in this window]
[in a new window]
  		TABLE 3. Test Statistics for Models (N = 925)

 

In the next step, we added the covariates and the binary distal outcome (12th grade smoking). To reduce the potential of an underspecified model, we freed paths from several covariates to the growth factors intercept, linear, and quadratic trends, the categorical latent variable, and the binary distal outcome. Figure 1 is a graphic depiction of the model specified for analysis. A GMM with a binary distal outcome is a GGMM. We regressed the latent class variable on the covariates to characterize class membership. Because research has found consistent gender differences in adolescent depressive symptoms (18,20), we controlled for the effects of gender by regressing the Latent Growth Model growth factors on gender, and estimated the path coefficients from gender to depressive symptoms level and trend factors for each trajectory separately. We did the same for baseline smoking to see whether there were different effects for baseline smoking on the development of depressive symptoms in each trajectory. To improve model specification, we freely estimated residual variances for the growth parameters level and linear trend for each class, but set the quadratic trend residual variance to zero (27,52). We also freely estimated the residual correlation between the level and linear trend parameters within each class.



View larger version (28K):
[in this window]
[in a new window]
  		Figure 1. GGMM for depressive symptoms and smoking.

 

Statistics for the four-trajectory model with covariates and distal outcome, controlling for gender and baseline smoking within each trajectory, were BIC = 26479.043 and entropy = 0.58. The LRM LRT was no longer significant (p = .62), suggesting the three-trajectory over the four-trajectory solution, also supporting the importance of proper model specification, before selecting the optimal number of trajectories (27). We fit a three-trajectory model to the data. The BIC value for the three-trajectory model was lower, supporting the three-trajectory model, BIC = 26389.492. The entropy value was higher, entropy = 0.61, and the average latent class probabilities for the most likely class membership were respectable at 0.82, 0.89, and 0.81 for trajectories 1 to 3, respectively. Moreover, the LMR LRT was now significant (p = .03), favoring the three-trajectory over the two-trajectory model. A test of the within-trajectory conditional normality assumption revealed acceptable skewness and kurtosis estimates for all depressive symptoms variables, with the exception of 10th grade depressive symptoms in trajectories 2 and 3. This suggests that the within-trajectory conditional normality assumption is tenable, supporting the use of the LMR LRT for model selection, but some caution is still advisable when relying solely on the statistical criteria for final model selection (27).

Before selecting the three-trajectory model, we wanted to ensure the three-trajectory solution was not a local solution, based on a small set of random start values. Thus, we repeated the analysis with 100 random start values, and 10 optimizations for each of the 100 sets of start values, instead of the Mplus 3.1 default of 10 sets of random start values and 1 optimization for each of the 10 sets (58). One hundred random sets of start values is suggested for a more accurate search of the multivariate sample space, along with 10 optimizations (58). The more comprehensive analysis found the identical solution as found with the 10 sets of random start values, supporting the three-trajectory solution.

The final step in model selection was substantive. We thought the three-trajectory solution was feasible for several reasons. First, 23% of our sample was in the trajectory with the highest average CES-D scores, which is similar to the 20% lifetime rate of clinical cases found in the Oregon Adolescent Depression Project (18). In addition, there appears to be a group of adolescents (33%) who do not experience elevated depressive symptoms. With two extreme trajectories, it is reasonable to conclude a moderate trajectory. Indeed, one study proposed the existence of subthreshold levels of psychiatric conditions in adolescents, including depression (59). These adolescents may have subsyndromal levels of depressive symptoms but are at risk for future elevations (18). Second, the three trajectories differ from each other in key ways, with respect to baseline smoking and the remaining covariates, as will be seen shortly, whereas the four-trajectory model resulted in two very similar low-symptoms trajectories that did not differ on the covariates. Third, the addition of a fourth trajectory resulted in a less parsimonious solution. Thus, we believe the three-trajectory solution is the best solution for our data. Figure 2 is a graphic representation of the final three-trajectory model, controlling for all covariates.



View larger version (15K):
[in this window]
[in a new window]
  		Figure 2. Trajectories of depressive symptoms based on the GGMM results.

 

Characteristics of the Three Trajectories
Trajectories of Depressive Symptoms
Inspection of the estimated and observed means for each trajectory resulted in the following classification: class 1, high symptoms (N = 175; 23%); class 2, moderate symptoms (N = 410; 44%); class 3, low symptoms (N = 340; 33%). The sample sizes for each trajectory reflect the participants’ most likely latent class membership based on the estimated posterior probabilities. Of the 175 adolescents with overall high symptoms, 69% had baseline CES-D scores above the clinical cutoff for MDD (≥22 for males; ≥24 for females). In the 10th and 11th grades, this proportion fell to 47% and 38%, respectively, reflecting the downward portion of the inverse quadratic trend. However, the proportion climbed again to 46% by 12th grade. The proportions for the adolescents with moderate symptoms were 4%, 18%, 25%, and 11%, respectively, reflecting the inverted U shape. Finally, the proportion of adolescents with low symptoms above the clinical cutoff was zero for all waves, except for a 3% spike in 10th grade.

Effects of Gender on Depressive Symptoms
Gender had a significant effect on the linear ({gamma}1 = 4.60; z = 2.61; p < .01) and quadratic ({gamma}2 = –1.44; z = –2.68; p < .01) trends for adolescents with moderate depressive symptoms. This suggests that for adolescents with moderate depressive symptoms, being female is associated with higher symptom acceleration in waves 1 and 2 (linear trend), followed by a slowing of deceleration (quadratic trend) in waves 3 and 4.1 There were no other significant effects for gender on the growth parameters in the remaining two trajectories (p > .05).

Effects of Baseline Smoking on Depressive Symptoms
Baseline smoking had a significant negative effect on the linear ({gamma}1 = –2.70; z = –2.46; p < .05) and a significant positive effect on the quadratic ({gamma}2 = 0.63; z = 2.10; p < .05) trends for adolescents with high depressive symptoms. This suggests that baseline smoking is associated with decelerating growth (linear trend), followed by an acceleration of decline (quadratic trend) in depressive symptoms. Conversely, for adolescents with moderate depressive symptoms, baseline smoking had a positive effect on the linear trend ({gamma}1 = 3.11; z = 2.85; p < .005) and a negative effect on the quadratic trend ({gamma}2 = –0.85; z = –2.42; p < .05). This suggests that for adolescents with moderate depressive symptoms, baseline smoking was associated with acceleration in growth (linear trend) followed by a deceleration of decline (quadratic trend) in depressive symptoms. However, baseline smoking did not have a significant effect on baseline depressive symptoms for adolescents in any trajectory, nor did baseline smoking have a significant effect on the linear or quadratic trends for adolescents with low depressive symptoms.

Residual Variances and Correlations
Residual variance for baseline depressive symptoms and linear trend was significant for adolescents with high and moderate depressive symptoms (p < .0001). Residual variance was not significant, however, for baseline or linear trend among adolescents with low depressive symptoms (p > .05). The residual correlation between baseline level and linear trend was significant and negative in the high and moderate symptoms trajectories (p < .0001).

Characteristics of Trajectory Membership
We next characterized trajectory membership by estimating the likelihood of adolescents with different levels of covariates belonging to a trajectory in relation to a comparison trajectory. The proportions for each categorical covariate in the model, for the entire sample and by trajectory, appear in Table 2. Trajectory specific means and SDs for the continuous covariates appear in Table 4. There were no significant differences on the covariates when comparing moderate and low depressive symptoms (p > .05). However, there were significant differences in the remaining trajectory comparisons.


View this table:
[in this window]
[in a new window]
  		TABLE 4. Means and SDs for Continuous Covariates by Trajectory

 

In comparing high with low depressive symptoms, adolescents with higher grades at baseline (odds ratio [OR] = 0.67; confidence interval [CI] = 0.46–0.97) were less likely to have high than low depressive symptoms. Adolescents higher in nonsport extracurricular activity at baseline (OR = 1.24; CI = 1.04–1.48) were more likely to have high than low symptoms. Finally, adolescents higher in physical activity at baseline were marginally less likely (OR = 0.84; CI = 0.71–0.996) to have high than low depressive symptoms. There were no other significant differences on the remaining covariates in comparing adolescents with high and low depressive symptoms (p > .05).

In comparing adolescents with high and moderate depressive symptoms, adolescents with higher levels of lifetime alcohol use at baseline (OR = 1.36; CI = 1.08–1.71) were more likely to have high than moderate symptoms. Adolescents with greater physical activity at baseline (OR = 0.80; CI = 0.67–0.95) were less likely to have high than moderate depressive symptoms. Finally, adolescents higher in nonsport extracurricular activity involvement at baseline (OR = 1.35; CI = 1.07–1.70) were more likely to have high than moderate symptoms. There were no other significant differences on the remaining covariates (p > .05).

Our final analysis assessed the effects of the latent trajectory classification on 12th grade smoking. The probability of being a current smoker in the 12th grade for adolescents with high, moderate, and low depressive symptoms was 0.20, 0.15, and 0.09, respectively. The difference was significant ({chi}2[2,925] = 11.35; p = .003), suggesting that adolescents with higher depressive symptoms smoked more at follow-up. To ensure this finding represents a meaningful effect of trajectory membership on 12th grade smoking, we also controlled for the effects of baseline smoking (Figure 1). Baseline smoking had a significant effect on 12th grade smoking (ß = 1.19; z = 9.84; p < .0001). We tested whether this effect differed by trajectory with a likelihood ratio test, comparing the model with the paths constrained equal across the three trajectories, to the model with the path from baseline to 12th grade smoking freely estimated within each trajectory. The difference was not significant. This result suggests that although there was a significant difference in the proportion of current smokers in 12th grade based on trajectory membership, we cannot conclude that adolescents with higher levels of depressive symptoms smoked more because of their depressive symptoms. In other words, we cannot use the finding of a significant difference in 12th grade smoking to inform research about the direction of the relationship betweens smoking and depressive symptoms in adolescents, because baseline smoking had a positive effect on 12th grade smoking that did not differ by trajectory.


   DISCUSSION
TOP
 ABSTRACT
 INTRODUCTION
 METHODS
 RESULTS
 DISCUSSION
CONCLUSION
 NOTES
 REFERENCES
 
This study assessed trajectories of adolescent depressive symptoms, and characterized these trajectories on covariates associated with depressive symptoms and smoking. We also assessed the effects of trajectory on 12th grade smoking, controlling for baseline smoking. The findings revealed three latent trajectories of depressive symptoms (ie, high, moderate, and low) that differed with respect to baseline and 12th grade smoking, and several covariates related to smoking and depressive symptoms. Collectively, these findings suggest that the mixed results regarding directionality of the smoking-depression relationship in the literature may be attributable, in part, to the presence of latent subpopulations of adolescents with different developmental trajectories of depressive symptoms. As such, identifying subpopulations at risk for depressive symptoms and smoking with properly specified growth models can provide valuable insight into the relationship between smoking and depressive symptoms in adolescents and may facilitate smoking prevention and cessation efforts by targeting adolescents at special risk for smoking.

Consistent with our first hypothesis, we found multiple trajectories of depressive symptoms, which included high, moderate, and low symptoms. This finding suggests the existence of at least three subpopulations of adolescents who follow different developmental trajectories with respect to depressive symptoms. This developmental heterogeneity indicates that treating constructs like depressive symptoms from a single population assumption may mask potentially critical differences by averaging over contrasting developmental trends. For instance, there was a single quadratic trajectory for depressive symptoms when treating adolescents as a single population, with an initial increase (waves 1 and 2), followed by a period of declining symptoms (waves 3 and 4). However, GGMM identified trajectories of adolescents with high (N = 175) and moderate (N = 410) symptoms, and a third trajectory of adolescents with low depressive symptoms (N = 340). This suggests that the inconsistent findings regarding the relationship between smoking and depression in previous research may have resulted, in part, from an assumption of a single population of adolescents equally influenced by depression and smoking. Our results showed distinct differences in the effects of baseline smoking on the rate of change in depressive symptoms thereafter among the three trajectories. Thus, issues such as proper model specification and selection are critical to understanding complicated behavioral relationships like smoking and depressive symptoms in adolescents and should receive greater attention (27).

Consistent with our second hypothesis, we found characteristics that distinguish subpopulations of adolescents, providing tentative profiles of adolescents at particular risk for depressive symptoms and smoking. Past research has identified gender as one key source of developmental heterogeneity in trajectories of adolescent depressive symptoms (20–22). One study found that girls and boys had similar levels of depressive symptoms before 8th grade, but afterward, depressive symptoms differed for girls and boys (20). Consistent with these findings, we also found that being female was a risk factor for higher depressive symptoms, but only for adolescents with moderate symptoms, indicating that whereas females were at risk for an acceleration in depressive symptoms and deceleration in decline, they were not necessarily at greater risk than males for clinically high levels of depressive symptoms.

Smoking at baseline had an effect on adolescent depressive symptoms, but the relationship was opposite for adolescents with high and moderate symptoms. Although there was variability in the initial status and linear trend of depressive symptoms for both trajectories, it appears that for adolescents with high symptoms, smoking was associated with an overall deceleration in the rate of increase in depressive symptoms. For adolescents with moderate symptoms, however, smoking was associated with an overall acceleration in the rate of increase in depressive symptoms. This suggests that for adolescents with higher symptoms, smoking may be a form of self-medication.

Theory and research in adults are consistent with the possible antidepressant effects of nicotine. One hypothesis proposes that nicotine, particularly nicotine delivered through cigarette smoking, elevates mood by increasing arousal and decreasing negative affect for individuals with chronic or acute depressive symptoms (61). In support of this hypothesis, one study found that current female smokers scored higher on three facets of depression (ie, anhedonia, depressed affect, and somatic features) than exsmokers or nonsmokers (62). Another study found that the effect of depression on nicotine dependence was mediated by negative affect smoking (ie, smoking to lift mood) and smoking to increase arousal (63). An alternative hypothesis has linked the antidepressant effects of nicotine to its possible stimulating effects on the serotonergic system (64,65). Thus, it is possible that adolescents with high depressive symptoms benefited from the antidepressant effects of nicotine, although the research design precludes assignment of causality. Adolescents in the high depressive symptoms group had the highest rates of current smoking in 9th and 12th grades, but we do not know whether smoking or higher depressive symptoms came first, because we do not have data on either before 9th grade.

For adolescents with moderate symptoms, baseline smoking was associated with an overall increase in depressive symptoms. This relationship is less telling, because average depressive symptoms in the moderate trajectory were lower than the suggested CES-D cutoff for clinically elevated depressive symptoms in adolescents, 22 for males and 24 for females (18), and few were smoking at baseline. Nevertheless, the finding that the moderate and high depressive symptoms trajectories differed on several covariates, including baseline smoking, is intriguing. For instance, we found that adolescents higher in physical activity at baseline were more likely to have moderate than high depressive symptoms. This suggests that physical activity may be one factor that reduced these adolescents’ risk of high depressive symptoms and possibly smoking progression (51,66–69). However, the opposite is also possible, because previous research has found that 9th grade depressive symptoms were negatively associated with physical activity trend (51), indicating that higher levels of depressive symptoms were related to greater declines in physical activity over time.

Another intriguing yet perplexing finding is the greater likelihood of adolescents with higher levels of nonsport extracurricular activity to have high than low or moderate depressive symptoms. Increasing involvement in pleasurable activities is a fundamental component in behaviorally oriented prescriptions for depression (18). In addition, factors like low popularity, inferior peer relationships, and not having many friends are positively related to adolescent depression (33,70). Extracurricular activity involvement seems to run counter to these factors (71). However, greater extracurricular activity may not necessarily equate with the amount of reward derived from these activities, such as greater interpersonal success. For instance, one study found that depressed adolescents behave in ways that evoke peer rejection (72). Unfortunately, we did not assess interpersonal competence beliefs or the rewarding nature of these activities.

We also found that adolescents with high depressive symptoms were more likely to have lower grades at baseline than were adolescents with low depressive symptoms. Further, we found that alcohol use was associated with a greater likelihood of high than moderate depressive symptoms. These findings are consistent with past research findings that adolescents with elevated depressive symptoms are not likely to do as well as their nondepressed counterparts in academics (33,34,73). Moreover, they are more likely to use substances like alcohol (18).

Limitations
The present study has several limitations. First, because there is no standard approach for determining the optimal number of trajectories in GMM, it is possible that more trajectories of depressive symptoms were appropriate than the three selected. However, the nonsignificant LMR LRT did not support more trajectories. Further, we used substantive criteria along with statistical criteria to inform our final model choice. Other researchers may have selected a model with a different number of trajectories depending on different substantive criteria. Second, although we included an extensive number of covariates in this analysis, other covariates that may influence depressive symptoms that were not available for this analysis could have resulted in a different set of findings. For instance, research has shown that factors like socioeconomic status, delinquent behavior, and weight management are related to adolescent smoking (19,35,49), and factors like anxiety, stress, attributional style, and pubertal timing are related to depression in adolescents (18,20–22). Future studies should assess the effects of these and other covariates on depression and smoking progression. However, we believe that our choice of covariates in the present study reflects important cofactors of smoking and depressive symptoms in adolescents. Third, we were not able to make statements about specific racial groups. Although 36% of our participants were non-Caucasian, breaking down the non-Caucasian racial group would have resulted in a dramatic decrease in the number of participants selected to each subcategory of this covariate and could negatively influence model convergence (74). Fourth, the CES-D is a self-report measure of depressive symptoms, which were not confirmed with a clinical diagnostic interview. Thus, the results of this analysis may not inform clinical decisions.

Finally, although 72% of the parents of the eligible participants returned a response regarding study participation, we are unable to make comparisons with the 28% who did not provide a response. Seventy-five percent of those who responded did provide consent, and the differences between those who provided consent and those who declined were relatively small and few (28). However, some caution is warranted in generalizing the results of this study. It is important to note that the percentage of adolescents in our sample with low depressive symptoms scores and scores at or above the clinical cutoff is similar to the prevalence reported in epidemiological studies of adolescent depression (eg, 18).


   CONCLUSION
TOP
 ABSTRACT
 INTRODUCTION
 METHODS
 RESULTS
 DISCUSSION
 CONCLUSION
NOTES
 REFERENCES
 
The findings of the present study revealed the existence of distinct subpopulations of adolescent depressive symptoms with respect to smoking, and covariates related to both. We found differences in the effects of 9th grade smoking on how depressive symptoms change with time and characteristics that increase the likelihood of different trajectories of depressive symptoms in adolescents. Such characterization may be useful in smoking prevention and cessation efforts by targeting adolescents at particular risk for depression and smoking. Further, although there were differences in 12th grade smoking by trajectory, these differences were not attributable to trajectory membership separate from the effects of baseline smoking. This finding accentuates the importance of model specification (eg, controlling for baseline smoking in the analysis of the effects of latent class on 12th grade smoking), because underspecified models can result in spurious findings. Collectively, the findings of this study suggest that the question of directionality in the smoking-depression relationship may be answered better by examining developmental heterogeneity in these processes.


   NOTES
TOP
 ABSTRACT
 INTRODUCTION
 METHODS
 RESULTS
 DISCUSSION
 CONCLUSION
 NOTES
REFERENCES
 
1Interpretation of effects of covariates on the linear and quadratic trends is somewhat complicated. The linear trend applies primarily to the first two waves, because the weights (factor loadings) are identical (0 and 1) for both trends, whereas the quadratic trend applies primarily to the last two waves. Values for the final two waves differ for the linear (2 and 3) and quadratic (4 and 9) trends, because the latter is the square of the former (linear = x; quadratic = x2). Back

This study was supported by National Cancer Institute and National Institute on Drug Abuse grant P50 CA/DA 84718

Received for publication March 29, 2004; revision received October 1, 2004.

DOI:10.1097/01.psy.0000156929.83810.01


   REFERENCES
TOP
 ABSTRACT
 INTRODUCTION
 METHODS
 RESULTS
 DISCUSSION
 CONCLUSION
 NOTES
 REFERENCES
 

  1. [CDC] CoDCaP: Tobacco information and prevention source (TIPS): OSH summary for 2002, n.d.
  2. Brooks TL, Harris SK, Thrall JS, Woods ER. Association of adolescent risk behaviors with mental health symptoms in high school students. J Adolesc Health 2002;31:240–6.[Medline]
  3. Coogan PF, Adams M, Geller AC, Brooks D, Miller DR, Lew RA, Koh HK. Factors associated with smoking among children and adolescents in Connecticut. Am J Prev Med 1998;15:17–24.
  4. Covey LS, Tam D. Depressive mood, the single-parent home, and adolescent cigarette smoking. Am J Public Health 1990;80:1330–3.[Abstract/Free Full Text]
  5. Fergusson DM, Lynskey MT, Horwood LJ. Comorbidity between depressive disorders and nicotine dependence in a cohort of 16-year-olds. Arch Gen Psychiatry 1996:1043–7.
  6. Glied S, Pine DS. Consequences and correlates of adolescent depression. Arch Pediatr Adolesc Med 2002;156:1009–14.[Abstract/Free Full Text]
  7. Whalen CK, Jamner LD, Henker B, Delfino RJ. Smoking and moods in adolescents with depressive and aggressive dispositions: evidence from surveys and electronic diaries. Health Psychol 2001;20:99–111.[CrossRef][Medline]
  8. Kelder SH, Murray NG, Orpinas P, Prokhorov A, McReynolds L, Zhang Q, Roberts R. Depression and substance use in minority middle-school students. Am J Public Health 2001;91:761–6.[Abstract]
  9. Patton GC, Carlin JB, Coffey C, Wolfe R, Hibbert M, Bowes G. Depression, anxiety, and smoking initiation: a prospective study over 3 years. Am J Public Health 1998;88:1518–22.[Abstract/Free Full Text]
  10. Killen JD, Robinson TN, Haydel KF, Hayward C, Wilson DM, Hammer LD, Litt IF, Taylor CB. Prospective study of risk factors for the initiation of cigarette smoking. J Consult Clin Psychol 1997;65:1011–6.[CrossRef][Medline]
  11. Albers AB, Biener L. The role of smoking and rebelliousness in the development of depressive symptoms among a cohort of Massachusetts adolescents. Prev Med 2002;34:625–31.[CrossRef][Medline]
  12. Choi WS, Patten CA, Gillin JC, Kaplan RM, Pierce JP. Cigarette smoking predicts development of depressive symptoms among U.S. adolescents. Ann Behav Med 1997;19:42–50.[Medline]
  13. Goodman E, Capitman J. Depressive symptoms and cigarette smoking among teens. Pediatrics 2000;106:748–55.[Abstract/Free Full Text]
  14. Wu LT, Anthony JC. Tobacco smoking and depressed mood in late childhood and early adolescence. Am J Public Health 1999;89:1837–40.[Abstract/Free Full Text]
  15. Breslau N, Kilbey MM, Andreski P. Nicotine dependence and major depression: new evidence from a prospective investigation. Arch Gen Psychiatry 1993;50:31–5.[Abstract/Free Full Text]
  16. Breslau N, Peterson EL, Schultz LR, Chilcoat HD, Andreski P. Major depression and stages of smoking. Arch Gen Psychiatry 1998;55:161–6.[Abstract/Free Full Text]
  17. Brown RA, Lewinsohn PM, Seeley JR, Wagner EF. Cigarette smoking, major depression, and other psychiatric disorders among adolescents. J Am Acad Child Adolesc Psychiatry 1996;35:1602–10.[CrossRef][Medline]
  18. Lewinsohn PM, Rohde P, Seeley JR. Major depressive disorder in older adolescents: prevalence, risk factors, and clinical implications. Clin Psychol Rev 1998;18:765–94.[CrossRef][Medline]
  19. Windle M, Windle RC. Depressive symptoms and cigarette smoking among middle adolescents: prospective associations and intrapersonal and interpersonal influences. J Consult Clin Psychol 2001;69:215–26.[CrossRef][Medline]
  20. Ge X, Conger RD, Elder GH. Pubertal transition, stressful life events, and the emergence of gender differences in adolescent depressive symptoms. Dev Psychol 2001;37:404–17.[CrossRef][Medline]
  21. Ge X, Lorenz FO, Conger RD, Elder GH, Simons RL. Trajectories of stressful life events and depressive symptoms during adolescence. Dev Psychol 1994;30:467–83.[CrossRef]
  22. Garber J, Keiley MK, Martin C. Developmental trajectories of adolescents’ depressive symptoms: predictors of change. J Consult Clin Psychol 2002;70:79–95.[CrossRef][Medline]
  23. Goodman E, Huang B, Wade TJ, Kahn RS. A multilevel analysis of the relation of socioeconomic status to adolescent depressive symptoms: does school context matter? J Pediatr 2003;143:451–6.[CrossRef][Medline]
  24. Muthén BO. Second-generation structural equation modeling with a combination of categorical and continuous latent variables: new opportunities for latent class/latent growth modeling. In: Collins L, Sayer A, editors. New methods for the analysis of change. Washington, DC: APA; 2001:291–322.
  25. Muthén BO, Muthén LK. Integrating person-centered and variable-centered analyses: growth mixture modeling with latent trajectory classes. Alcoholism: clinical and experimental research 2000;24:882–91.[CrossRef][Medline]
  26. Lubke G, Muthen B. Investigating population heterogeneity with factor mixture models, 2003.
  27. Muthén BO. Latent variable analysis: growth mixture modeling and related techniques for longitudinal data. In: Kaplan D, editor. Handbook of quantitative methodology for the social sciences. Nebury Park, CA: Sage Publications; 2004:345–68.
  28. Audrain J, Tercyak KP, Goldman P, Bush A. Recruiting adolescents into genetic studies of smoking behavior. Cancer Epidemiol Biomarkers Prev 2002;11:249–52.[Abstract/Free Full Text]
  29. Gold MS, Bentler PM, Kim KH. A comparison of maximum-likelihood and asymptotically distribution-free methods of treating incomplete nonnormal data. Structural Equation Modeling 2003;10:47–79.[CrossRef]
  30. Radloff LS. The CES-D scale: a self-report depression scale for research in the general population. Appl Psychol Meas 1977;1:385–401.[CrossRef]
  31. Radloff LS. The use of the center for epidemiologic studies depression scale in adolescents and young adults. J Youth Adolesc 1991;20:149–66.[CrossRef]
  32. Roberts RE, Andrews JA, Lewinsohn PM, Hops H. Assessment of depression in adolescents using the center for epidemiologic studies depression scale. Psychol Assess 1990;2:122–8.[CrossRef]
  33. Field T, Diego M, Sanders C. Adolescent depression and risk factors. Adolescence 2001;36:491–8.[Medline]
  34. Garrison CZ, Schluchter MD, Schoenbach VJ, Kaplan BK. Epidemiology of depressive symptoms in young adolescents. J Am Acad Child Adolesc Psychiatry 1989;28:343–51.[Medline]
  35. Goodman E, Huang B. Socioeconomic status, depressive symptoms, and adolescent substance use. Arch Pediatr Adolesc Med 2002;156:448–53.[Abstract/Free Full Text]
  36. Kann L, Kinchen SA, Williams BI, Ross JG, Lowry R, Hill CV, Grunbaum JA, Blumson PS, Collins JL, Kolbe LJ. Youth risk behavior surveillance—United States, 1997. State and Local YRBSS Coordinators. J Sch Health 1998;68:355–69.[Medline]
  37. Grunbaum JA, Kann L, Kinchen S, Ross J, Hawkins J, Lowry R, Harris WA, McManus T, Chyen D, Collins J. Youth risk behavior surveillance—United States, 2003. MMWR Surveill Summ 2004;53:1–96.[Medline]
  38. Grunbaum JA, Kann L, Kinchen SA, Williams B, Ross JG, Lowry R, Kolbe L. Youth risk behavior surveillance—United States, 2001. J Sch Health 2002;72:313–28.[Medline]
  39. Choi WS, Pierce JP, Gilpin EA, Farkas AJ, Berry CC. Which adolescent experimenters progress to established smoking in the United States. Am J Prev Med 1997;13:385–91.[Medline]
  40. Brener ND, Collins JL, Kann L, Warren CW, Williams BI. Reliability of the Youth Risk Behavior Survey questionnaire. Am J Epidemiol 1995;141:575–80.[Abstract/Free Full Text]
  41. Brener ND, Kann L, McManus T, Kinchen SA, Sundberg EC, Ross JG. Reliability of the 1999 Youth Risk Behavior Survey questionnaire. J Adolesc Health 2002;31:336–42.[CrossRef][Medline]
  42. Patrick DL, Cheadle A, Thompson DC, Diehr P, Koepsell T, Kinne S. The validity of self-reported smoking: a review and meta-analysis. Am J Public Health 1994;84:1086–93.[Abstract/Free Full Text]
  43. Botvin GJ, Botvin EM. Adolescent tobacco, alcohol, and drug abuse: prevention strategies, empirical findings, and assessment issues. J Dev Behav Pediatr 1992;13:290–301.[Medline]
  44. Rodriguez D, Audrain-McGovern J. Team sport participation and smoking. J Pediatr Psychol 2004;29:299–308.[Abstract/Free Full Text]
  45. Abrams K, Skolnik N, Diamond JJ. Patterns and correlates of tobacco use among suburban Philadelphia 6th - through 12th grade students. Fam Med 1999;31:128–32.[Medline]
  46. Audrain-McGovern J, Rodriguez D, Tercyak KP, Epstein LH, Goldman P, Wilyeto EP. Applying a behavioral economic framework to understanding adolescent smoking. Psychol Addict Behav 2004;18.
  47. Duncan TE, Duncan SC. A multivariate latent growth curve analysis of adolescent substance use. Structural Equation Modeling 1996;3:323–47.
  48. Deleted in proof.
  49. Smimantov E, Schoen C, Klein JD. Health-comprising behaviors: why do adolescents smoke or drink?: identifying underlying risk and protective factors. Arch Pediatr Adolesc Med 2000;154:1025–33.[Abstract/Free Full Text]
  50. Lloyd-Richardson EE, Papandonatos G, Kazura A, Stanton C, Niaura R. Differentiating stages of smoking intensity among adolescents: stage-specific psychological and social influences. J Consult Clin Psychol 2002;70:998–1009.[CrossRef][Medline]
  51. Audrain-McGovern J, Rodriguez D, Moss HB. Smoking progression and physical activity. Cancer Epidemiol Biomark Prev 2003;12:1121–9.[Abstract/Free Full Text]
  52. Muthén BO, Muthén LK: Mplus short course: traditional latent variable modeling using Mplus, 2002.
  53. Muthén BO, Shedden K. Finite mixture modeling with mixture outcomes using the EM algorithm. Biometrics 1999;55:463–9.[CrossRef][Medline]
  54. Raskin White H, Pandina RJ, Chen P-H. Developmental trajectories of cigarette use in adolescence into young adulthood. Drug Alcohol Depend 2002;65:167–78.[CrossRef][Medline]
  55. Nagin D, Tremblay RE. Trajectories of boys’ physical aggression, opposition, and hyperactivity on the path to physically violent and nonviolent juvenile delinquency. Child Dev 1999;70:1181–96.[CrossRef][Medline]
  56. Nagin D, Tremblay RE. Analyzing developmental trajectories of distinct but related behaviors: a group-based method. Psychol Methods 2001;6:18–34.[CrossRef][Medline]
  57. Muthén BO, Muthén LK. Mplus user’s guide. Los Angeles, CA: Muthen & Muthen; 2001.
  58. Muthén LK, Muthén BO. Mplus user’s guide. Los Angeles, CA: Muthén & Muthén; 1998–2004.
  59. Lewinsohn PM, Shankman SA, Gau JM, Klein DN. The prevalence and co-morbidity of subthreshold psychiatric conditions. Psychol Med 2004;34:613–22.[CrossRef][Medline]
  60. Deleted in proof.
  61. Pomerleau CS. Co-factors for smoking and evolutionary psychobiology. Addiction 1997;92:397–408.[CrossRef][Medline]
  62. Pomerleau CS, Zucker AN, Stewart AJ. Patterns of depressive symptomatology in women smokers, ex-smokers, and never-smokers. Addict Behav 2003;28:575–82.[CrossRef][Medline]
  63. Lerman C, Audrain J, Orleans CT, Boyd R, Gold K, Main D, Caporaso N. Investigation of mechanisms linking depressed mood to nicotine dependence. Addict Behav 1996;21:9–19.[CrossRef][Medline]
  64. Salin-Pascual RJ, Drucker-Colin R. A novel effect of nicotine on mood and sleep in major depression. Neuroreport 1998;9:57–60.[Medline]
  65. Vazquez J, Guzman-Marin R, Salin-Pascual RJ, Drucker-Colin R. Transdermal nicotine on sleep and PGO spikes. Brain Res 1996;737:317–20.[CrossRef][Medline]
  66. Motl RW, Birnbaum AS, Kubik MY, Dishman RK. Naturally occurring changes in physical activity are inversely related to depressive symptoms during early adolescence. Psychosom Med 2004;66:336–42.[Abstract/Free Full Text]
  67. Fulkerson JA, Sherwood NE, Perry CL, Neumark-Sztainer D, Story M. Depressive symptoms and adolescent eating and health behaviors: a multifaceted view in a population-based sample. Prev Med 2004;38:865–75.[CrossRef][Medline]
  68. Goodwin RD. Association between physical activity and mental disorders among adults in the United States. Prev Med 2003;36:698–703.[CrossRef][Medline]
  69. Vickers KS, Patten CA, Lane K, Clark MM, Croghan IT, Schroeder DR, Hurt RD. Depressed versus nondepressed young adult tobacco users: differences in coping style, weight concerns and exercise level. Health Psychol 2003;22:498–503.[CrossRef][Medline]
  70. Petersen AC, Compas BE, Brooks-Gunn J, Stemmler M, Ey S, Grant KE. Depression in adolescence. Am Psychol 1993;48:155–68.[CrossRef][Medline]
  71. Mahoney JL, Cairns BD, Farmer TW. Promoting interpersonal competence and educational success through extracurricular activity participation. J Educ Psychol 2003;95:409–18.[CrossRef]
  72. Baker M, Milich R, Manolis MB. Peer interactions of dysphoric adolescents. J Abnorm Child Psychol 1996;24:241–55.[CrossRef][Medline]
  73. Windle M, Windle RC. Coping strategies, drinking motives, and stressful life events among middle adolescents: associations with emotional and behavioral problems and with academic functioning. J Abnorm Psychol 1996;105:551–60.[CrossRef][Medline]
  74. Hosmer DW, Lemeshow S. Applied logistic regression. New York: John Wiley & Sons, Inc; 2000.



This article has been cited by other articles:


Home page
 PediatricsHome page
J. Audrain-McGovern, D. Rodriguez, V. Patel, M. S. Faith, K. Rodgers, and J. Cuevas
How Do Psychological Factors Influence Adolescent Smoking Progression? The Evidence for Indirect Effects Through Tobacco Advertising Receptivity
Pediatrics, April 1, 2006; 117(4): 1216 - 1225.
[Abstract] [Full Text] [PDF]

This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrowRequest Permissions
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Rodriguez, D.
Right arrow Articles by Audrain-McGovern, J.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Rodriguez, D.
Right arrow Articles by Audrain-McGovern, J.
Related Collections
Right arrow Depression

HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS