Addictive Behaviors 42 (2015) 154–161

Contents lists available at ScienceDirect

Addictive Behaviors

Confounders or intermediate variables? Testing mechanisms for the relationship between depression and smoking in a longitudinal cohort study Michael Chaiton a,c,⁎, Joanna E. Cohen a,c,d, Jürgen Rehm a,e,f, Mohamed Abdulle a, Jennifer O'Loughlin b a

Dalla Lana School of Public Health, University of Toronto, Canada Centre de Recherche du CHUM, University of Montreal, Montreal, Quebec, Canada c Ontario Tobacco Research Unit, University of Toronto, Canada d Johns Hopkins Bloomberg School of Public Health, United States e Centre for Addiction and Mental Health, Canada f Addiction Research Institute, Zürich, Switzerland b

H I G H L I G H T S • • • •

The relationship of depressive symptoms and smoking is a complex one. Care must be taken to differentiate between intermediate variables and confounders. Pathways between depressive symptoms and smoking varied by the direction of effect. Stress and friend smoking were important intermediate variables.

a r t i c l e

i n f o

Available online 27 November 2014 Keywords: Smoking Depression Cohort Causal pathways Adolescent

a b s t r a c t Introduction: The relationship between the onset of smoking and the onset of depression among adolescents has been well document, but the mechanisms underlying the relationship are unclear. This paper uses an empirical method to assess potential intermediate variables in the pathway between changes in depressive symptoms and cigarette smoking in a longitudinal cohort of adolescents. Methods: 837 participants from a cohort in Montreal, Canada who had not smoked and did not have elevated depressive symptoms at baseline were followed for five years from 1999 to 2003. The role of a set of 15 variables previously identified in the literature as potential confounders were systematically evaluated as predictors of exposure and outcome, for attenuation of the association by more than 10%, and for intra-individual change in the variable after onset of exposure. Results: The magnitude of the association between smoking and depressive symptoms was fully attenuated after adjustment for all variables included indiscriminately. A concept map was developed detailing the empirical associations between the variables within this data set. Stress, worry about weight, and worry about parents were identified as intermediate variables for both smoking predicting depressive symptoms and depressive symptoms predicting smoking. Cox regressions with appropriate confounders maintained statistical significance. Conclusion: Cigarette smoking is associated with higher depressive symptoms prior to and after inclusion of empirical confounders. Inclusion of intermediate variables in multivariable models can lead to the erroneous conclusion that there is no association between smoking and depression. © 2014 Elsevier Ltd. All rights reserved.

1. Introduction The association between smoking and depression or depression symptoms has been well documented over the last 30 years, but to ⁎ Corresponding author at: Dalla Lana School of Public Health, University of Toronto, T514 33 Russell St., Toronto, ON M5S 2S1, Canada. Tel.: +1 4165358501x34428. E-mail address: [email protected] (M. Chaiton).

http://dx.doi.org/10.1016/j.addbeh.2014.11.026 0306-4603/© 2014 Elsevier Ltd. All rights reserved.

date there is neither consensus on the mechanisms underlying the relationship, nor on the potential temporal order—that is, whether smoking precedes depression, or depression precedes smoking (Chaiton, Cohen, O'Loughlin, & Rehm, 2009; Paperwalla, Levin, Weiner, & Saravay, 2004). An estimated 150 million adolescents worldwide use tobacco and nearly 5 million deaths are attributed to tobacco use annually. Further, depression is the number one cause of disability worldwide.

M. Chaiton et al. / Addictive Behaviors 42 (2015) 154–161

The magnitude of the association between smoking and depression has been shown to be substantially reduced after controlling for potential confounders (Black, Zimmerman, & Coryell, 1999; Brown, Lewinsohn, Seeley, & Wagner, 1996; Dierker, Avenevoli, Stolar, & Merikangas, 2002; Federman, Costello, Angold, Farmer, & Erkanli, 1997). Chaiton et al. (2009) compiled a list of variables (Table 1) previously incorporated as confounders in longitudinal cohort analyses of the relationship between smoking and depression (Chaiton et al., 2009). These variables were included in multivariate models to, explicitly or implicitly, account for the theory that the apparent association between smoking and depression is spurious due to a common underlying variable (Breslau, Peterson, Schultz, Chilcoat, & Andreski, 1998; Park & Romer, 2007). Because these variables have been included as potential confounders in previous analyses, it is possible to suggest that, based on previous literature, all these variables should be included in future analyses. A confounder must be associated with both the exposure and the outcome; and in addition must affect the value of the estimate of the relationship between the exposure and the outcome (Boslaugh, 2008). A 10 percent difference between the unadjusted and adjusted estimate of effect has been proposed as a guideline for whether or not the variable affects the estimate of the relationship between the exposure and the outcome substantively (Rothman, Greenland, & Lash, 2008). Similar to a confounder, an intermediate variable (also termed a “mediator”) is associated with both exposure and outcome, and will also affect the estimate of the relationship of the exposure and the outcome. In contrast to a confounder, an intermediate variable is a consequence of the exposure as it lies on the causal pathway between the exposure and the outcome (Greenland, 2008). That is, the onset of the exposure should lead to changes in the mediator which occur prior to the onset of the outcome. Adding intermediate variables into the model could obscure a real association between depression and smoking by over-controlling for the causal impact (Greenland, 2008). However, Duncan and Rees argued that several purported confounders (i.e., stress, academic performance, self-esteem) might in fact be intermediate (rather than confounding) variables that lie on the causal pathway between smoking and depression (Duncan & Rees, 2005). Several theories have been proposed to explain the mechanisms underlying the association between smoking and depression (Breslau

155

et al., 1998; Johnson, Rhee, Chase, & Breslau, 2004; Kendler et al., 1993) and, depending on the theory, the same set of variables could quite legitimately be considered either as confounders or intermediate variables or both in the causal pathway. Consequently, existing theory offers little guidance as to the appropriate role of variables. This paper examines the effect of a wide range of variables previously conceptualized as confounders, on the longitudinal relationship between smoking and depressive symptoms. It quantifies the effect of the variables on the relationship between smoking and depressive symptoms, and it then uses the empirical information to develop a concept map of the factors that affect the relationship between smoking and depressive symptoms. 2. Materials and methods 2.1. Population and design The NDIT Study is a longitudinal cohort study (1999–2005) designed to investigate the natural course of early cigarette use and the development of nicotine dependence in novice smokers. The cohort included 1293 students initially aged 12–13 years recruited from all grade seven classes in a sample of ten secondary schools in Montreal, Canada. Secondary schools were purposively selected to include a mix of French and English schools, urban, suburban, and rural schools, and schools located in high and low socioeconomic neighbourhoods, all with a relatively low turnover of students. Thirteen schools were approached and agreed to participate. However, two schools were excluded due to a low return of signed parental consent forms, and one school was excluded because the school administrators could not guarantee cooperation over the entire study. All students in all Secondary I (grade 7) classes at participating schools were invited to participate. Potential participants were given an information letter to bring home to their parents. Signed informed guardian consent and student consent were obtained from all participants and from a parent or guardian. Over half of eligible students (56.2%) participated in the baseline data collection. This relatively low response was due, in part, to a labour dispute that resulted in teachers in several schools refusing to collect consent forms. Blood sampling needed for genetic analysis also likely affected the response adversely.

Table 1 Variables included in final multivariate models reported in longitudinal studies that examined the association between smoking and depression in adolescents (Eppel, O'Loughlin, Paradis, & Platt, 2006). Variable

Variable included in current study

Citation numbers of study(ies) that included variable in the final multivariate model

Age Sex Alcohol use

Age Sex Alcohol use

Parental education

Parent education

Race/ethnic group

Percent life in Canada

Smoking among peers

Friends smoking

Parental smoking Academic performance Physical activity Stress/anxiety

Parental smoking Academic Performance Physical Activity Stress/Worry about parents/Worry about weight Novelty-seeking Impulsivity Self-esteem − − −

(all) (all) Audrain-McGovern, Lerman, Wileyto, Rodriguez, & Shields (2004), Fergusson, Goodwin, & Horwood (2003), Killen et al. (1997), Patton et al. (1998), Skara, Sussman, & Dent (2001), Windle & Windle (2001), Wu & Anthony (1999) Brook, Schuster, & Zhang (2004), Brown et al. (1996), Duncan & Rees (2005), Escobedo et al. (1998), Silberg et al. (2003), Skara et al. (2001) Audrain-McGovern et al. (2004), Brown et al. (1996), Escobedo et al. (1998), Fergusson et al. (2003), Goodman & Capitman (2000), Skara et al. (2001), Wu & Anthony (1999) Audrain-McGovern et al. (2004), Fergusson et al. (2003), Killen et al. (1997), Patton et al. (1998), Silberg et al. (2003), Skara et al. (2001), Windle & Windle (2001) Escobedo et al. (1998), Patton et al. (1998), Silberg et al. (2003), Skara et al. (2001) Audrain-McGovern et al. (2004), Escobedo et al. (1998), Fergusson et al. (2003), Skara et al. (2001) Audrain-McGovern et al. (2004), Patton et al. (1998) Costello, Erkanli, Federman, & Angold (1999), Fergusson et al. (2003), O'Loughlin et al. (2002) Fergusson et al. (2003) Windle & Windle (2001) Fergusson et al. (2003) Fergusson et al. (2003) Windle & Windle (2001) Audrain-McGovern et al. (2004), Brook et al. (2004), Skara et al. (2001), Windle & Windle (2001)

− −

Fergusson et al. (2003) Fergusson et al. (2003), Silberg et al. (2003)

Novelty-seeking Temperament Self-esteem Parental attachment Social support Marijuana/other drug use Childhood adversity Conduct problems

156

M. Chaiton et al. / Addictive Behaviors 42 (2015) 154–161

Baseline data were collected in fall 1999. Participants completed a 45-minute questionnaire administered either in the classroom or in the school cafeteria to groups of classes. All participants were surveyed every 3 months on average during the 10-month school year for five years. Most data collection took place during the months of October, January, March, and May. Alternative data collection dates were scheduled for participants who were not available on the originally scheduled date. Participants who reported smoking and/or had elevated depression scores at baseline were excluded from the survival analyses. The longitudinal sample included 837 participants who completed at least 2 surveys, reported never smoking at baseline, and had depressive symptom scores lower than 3.4 at baseline. Person-years were calculated at each survey by subtracting initial age at baseline from age at each survey. 2.2. Description of study variables 2.2.1. Depressive symptoms Depressive symptoms were measured in the validated 6-item Mellinger depressive symptom scale,(Choi, Patten, Gillin, Kaplan, & Pierce, 1997; Escobedo, Reddy, & Giovino, 1998; Kandel, Huang, & Davies, 2001; Zhu & Valbo, 2002) which assessed how often in the past 3 months participants: (1) felt too tired to do things; (2) had trouble going to sleep or staying asleep; (3) felt unhappy, sad, or depressed; (4) felt hopeless about the future; (5) felt nervous or tense; (6) worried too much about things. Response choices ranged from never to often on a 4-point scale. The depressive symptom score was calculated as the mean value of the summary score based on these six items. The available items were summed and the sum was then divided by the number of valid responses to obtain a depressive symptom score between 1 and 4. Higher scores represent more depressive symptoms. Depressive symptom score cut points of 3.0 (Escobedo et al., 1998) and 3.4 (two standard deviations above the mean of 2.1 in NDIT) were used to create a dichotomous outcome to represent the onset of depression. Because the Mellinger scale may over-estimate depression compared to clinical diagnoses,(Escobedo et al., 1998) only the more conservative cut-point of 3.4 is reported, although in sensitivity analyses use of lower cut-point did not change the interpretation of results. It was assumed that an incident case developed a depressive symptom score greater than 3.4 midway between intervals of survey administration (Fischer, Najman, Williams, & Clavarino, 2012). Cronbach's alpha reliability coefficient for the measure indicates good reliability at 0.89 (Escobedo et al., 1998). At baseline, 35 participants (2.7%) had depressive symptom scores above 3.4; 218 (26%) developed an elevated depressive symptom score (above 3.4) at some point during the five-year follow-up. The rate of onset of elevated depressive symptomatology is similar to that found in other studies with close observation of adolescents over a long period, but higher than prevalence reports in single time point observations (Greenland, 2008). 2.2.2. Smoking At each survey, participants provided data on cigarette smoking for each of the three preceding months, including the number of days on which they had smoked each month and the average number of cigarettes smoked per day each month. Cigarette smoking status was also measured on a 6-point scale ranging from “never smoked a cigarette”, “took a puff on a cigarette”, “smoked a whole cigarette”, “smokes cigarettes monthly”, “smokes cigarettes weekly”, and “smoke cigarettes every day”. Smoking initiation was defined as the first report of smoking at least a puff on a cigarette. Test–retest reliability for the cigarette use frequency and intensity indicators has been evaluated as good (r = 0.78 and 0.75, respectively); reliability was lower but still acceptable for number of cigarettes smoked (ICC = 0.64) (Escobedo et al., 1998; Rodriguez, Moss, & Audrain-McGovern, 2005). Of the 1293 students in the study population, 134 participants (10%) reported daily smoking at baseline, 473 (30%) reported smoking at least a puff, and 679 participants (60%) had never smoked. During follow-up, 480 (30%) of the never smokers reported smoking at least a puff.

2.2.3. Other variables In a systematic review of longitudinal studies of the relation between smoking and depression, twenty variables had been used previously in regression models as “control” or “potential confounding” variables (Chaiton et al., 2009). Fifteen of the 20 variables are available in the NDIT cohort (see Table 1) (Chaiton et al., 2009; O'Loughlin, 2002; O'Loughlin, Karp, Koulis, Paradis, & DiFranza, 2009). These included socio-demographic characteristics (age, sex, percent life spent in Canada, parental education); indicators of smoking in the social environment (parent(s) smoke (Silberg, Rutter, D'Onofrio, & Eaves, 2003), friends smoke); psychological exposures (stress symptoms (Deschesnes, 1997), worry about weight, worry about relationship with parents, impulsivity (Eysenck & Eysenck, 1977; Wills, Windle, & Cleary, 1998), novelty-seeking, self-esteem (Vallieres & Vallerand, 1990) and academic performance; lifestyle-related factors (alcohol use, level of participation in physical activity) (Sanchez et al., 2007). Detailed descriptions of each variable can be found in Appendix I. 2.3. Missing data Missing data were imputed in four steps (O'Loughlin et al., 2009). First, the same value of each variable measured only once (i.e., selfesteem) was assigned to all 20 survey cycles for each participant. Second, for variables measured 2–3 times during follow-up (parental education, academic performance), we used a “first observation carried backward” in combination with a “last observation carried forward” strategy to impute values for the other 17–18 survey cycles. Third, variables thought to be time-varying (i.e., friends smoke, stress symptoms, depressive symptoms, worry about weight, worry about relationship with parents, alcohol use, level of physical activity) were measured in all 20 survey cycles. Missing values for time-varying variables for any given participant were imputed using the “first observation carried backward” and “last observation carried forward” strategies. Fourth, a “missing” value was assigned to categorical variables for participants for whom no data were available over the entire study for that variable. 2.4. Statistical analyses 2.4.1. Crude and over-adjusted model The Cox proportional hazards model was used to obtain an estimate of the unadjusted or crude risk of smoking predicting the onset of elevated depressive symptoms. All 15 potential confounding variables were added to the model to provide an initial adjusted estimate of the effect of smoking on depressive symptoms. 2.4.2. Identification of confounding and mediating variables The role of each variable in the relationship of interest was assessed in the following sequence of analyses for each variable. This sequence first used smoking as the outcome and depression as the exposure Step 1. Test whether the variable attenuates the estimated parameter of the relationship of smoking and depression. Bivariate Cox regressions were used. Step 2. Test whether or not the variable independently predicts the onset of the outcome. Fully adjusted Cox regressions were used. Step 3. Test whether or not the variable independently predicts the onset of the exposure prior to the onset of the outcome. Fully adjusted Cox regressions were used. Step 4. Test whether or not the onset of exposure is associated with changes in the values of variables over time. Only timevarying variables were assed. A series of fixed effect regressions was used in which intra-individual change in each time-varying variable was assessed before and after exposure, prior to outcome onset. Xtlogit in Stata was used for dichotomous variables and xtreg was used for continuous variables. Step 5. Test whether change in one variable is associated with change in

M. Chaiton et al. / Addictive Behaviors 42 (2015) 154–161

other variables in the time period after exposure onset but before outcome onset. Fixed effect regressions were used. Xtlogit was used for dichotomous variables and xtreg was used for continuous variables.

157

significantly from participants who had never smoked and did not have high depressive symptom scores by age, friends smoke, parent(s) smoke, and alcohol use. 3.1. Smoking predicting depression

All analyses used an alpha of b0.05 to determine significance, except for Step 1 which used a N 10% change in the estimate as the criterion for designating a substantive effect. Interaction terms for each of the selected variables were tested in both directional models (step 1) and were not significant for either the depressive symptoms or smoking outcomes (data not shown). The analysis was repeated in male and female subgroups. However, the same set of variables had a substantial effect (step 1) among males and females. Sample sizes were too small to conduct the fixed effect analyses by sex, so analyses were conducted in males and females combined. All analyses used Taylor series linearization to account for clustering within schools (Armitage & Colton, 2005). The survival analyses were stratified by the number of survey cycles to control for time in the study (Boslaugh, 2008). 2.4.3. Development of concept map A concept map (Greenland, Pearl, & Robins, 1999) was drawn according to the following sequence: (i) only variables that affect the estimate by more than 10% (in step 1) were included; (ii) arrows were drawn directly to the outcome for variables that were statistically significant predictors of the outcome in step 2; (iii) arrows were drawn to the exposure from variables that predicted the onset of exposure (in step 3) prior to the onset of the outcome; (iv) arrows were drawn from the exposure to a variable if the onset of exposure predicted change in the variable (step 4); (v) arrows were drawn between variables if one variable predicted changes in the second variable (step 5). 3. Results Four hundred and twenty one participants (33% of 1293 respondents at baseline) reported smoking prior to baseline (Table 2) and 34 (3%) had depressive symptom scores N 3.4. A total of 3163 person-years were contributed by 837 participants in the longitudinal subsample of participants who did not report smoking and did not have high depressive symptom scores at baseline. Fifty-two percent of the 837 contributed at least 19 survey cycles; 19% provided data in all 20 survey cycles. A total of 616 respondents of the longitudinal sample (74%) provided data in the last survey cycle. Half of respondents contributed at least 3.9 of a possible 5 person-years to the study. There were no significant differences in the population demographics between the initial analytic sample and those of the sample that provided data in the last cycle. In the longitudinal sample, 397 never smokers started smoking (incidence rate = 18 (95% Confidence Interval (CI): 16, 19) per 100 person-years). There were 199 participants whose depressive symptom score increased to 3.4 or higher (incidence rate = 7 (95% CI: 6, 8)) per 100 person years. Participants who at baseline had smoked at least a puff or who had elevated depressive symptom scores differed Table 2 Descriptive statistics for all participants at baseline. Nicotine Dependence in Teens Study, Montreal, Canada. n = 837. n Age (mean, (sd)) Sex (%) Male Female Parental education (%) High school or less More than high school No data Percent lifetime spent in Canada (%) 100 b100

837

12.7 (0.1)

421 416

50 50

52 605 180

6 72 22

743 94

89 11

The unadjusted relative hazard for smoked at least a puff predicting depression was 2.1 (95% CI: 1.6, 2.7). After controlling for all potential confounders, the estimate was attenuated to 1.1 (0.8, 1.5). Sex, alcohol use, stress, worry about weight, worry about parents, impulsivity and friends smoke each attenuated the estimate by more than 10% (Table 3). Smoking initiation was associated with increases in the friends smoke indicator, stress, and worry about weight (Table 4). Sex, alcohol use and impulsivity were classified as confounders, while friends smoke, stress, worry about weight and worry about parents were classified as intermediate variables (Table 5). Fig. 1 illustrates the empirical relationships between these variables, smoking and depression. After including only variables identified as confounders, the hazard ratio was 1.7 (95% CI: 1.2, 2.3) for smoking predicting depressive symptoms. 3.2. Depression predicting smoking The unadjusted relative hazard for depressive symptom scores N 3.4 predicting smoking initiation was 2.1 (95% CI: 1.4, 3.1). The fully adjusted relative hazard was 0.9 (95% CI: 0.6, 1.4). Alcohol use, stress, worry about weight, worry about parents, impulsivity and friends smoke attenuated the estimate by more than 10% (Table 3, Fig. 2). Friends smoke and impulsivity were categorized as confounders, while alcohol use, stress, worry about weight and worry about parents were intermediate variables. After including only variables identified as confounders in the Cox regression, the hazard ratio was 1.8 (95% CI: 1.2, 2.8) for depressive symptoms predicting smoking. 4. Discussion While depressive symptoms predicted smoking initiation, and smoking predicted the onset of depressive symptoms prior to adjustment, including all variables identified as confounders in previous reports as potential confounders in this current analysis, attenuated the magnitude of effects to the null. Interpretation of the association between smoking and depressive symptoms therefore depends critically on decisions made about which variables to include as confounders in the modelling process (MacKinnon, 2008). This article used a repeated measures design to ensure that the temporal relationship between smoking, depression, and possible mediators was clearly defined. In a review of longitudinal studies on smoking and depression in adolescents, Chaiton et al. (2009) reported that the pooled estimate for smoking predicting depression was 1.7 (95% CI: 1.3, 2.4), while the pooled estimate for depression predicting smoking was 1.4 (95% CI: 1.2, 1.6). These estimates are consistent with the adjusted estimates found in this current analysis (i.e., 1.8 (95% CI: 1.2, 2.8) for depressive symptoms predicting smoking, and 1.7 (95% CI: 1.2, 2.3) for smoking predicting depressive symptoms) (Chaiton et al., 2009). Our results suggest that the pathways from depressive symptoms to smoking differ from the pathways that relate smoking to depressive symptom onset. It is possible that the relationship between smoking and depression varies between subgroups not assessed in this analysis (Fischer et al., 2012). In some subgroups, smoking may lead to depression while in others, depression leads to smoking (Hooshmand, Willoughby, & Good, 2012; Killen et al., 1997; Rodriguez et al., 2005). Previous studies have shown that in adolescent girls, an increase in smoking uptake predicted an increase in depressive symptoms (Beal, Negriff, Dorn, Pabst, & Schulenberg, 2013; Boden, Fergusson, & Horwood, 2013). Other potential subgroups include those with genetic polymorphisms associated with different susceptibilities to depression and to smoking (Wills, Sandy, & Yaeger, 2002). Furthermore, different levels of social

158

M. Chaiton et al. / Addictive Behaviors 42 (2015) 154–161

Table 3 Results of assessment of variables on the relationship between smoking and depression in the Nicotine Dependence in Teens study, Montreal, Canada. (n = 837) (steps 1−3).

Sex Age in years Percent life in Canada Parent education Parent smokes Friends smoke Stress symptoms Self-esteem Impulsivity Novelty score Worry about weight Worry about parents Academic performance Alcohol use Physical activity

Predicts depression adjusteda

Predict % change in smoking Smoking initiation Predicts Predicts % change in Onset of depression smoking prior hazard predicting associated with smoking depression depression hazard associated with change to depressionb depressionc change in variabled adjusteda prior to smokingb predicting in variabled smokingc

Y N N Y N N N N N N Y Y Y N N

N Y N N Y Y N Y Y N N N N Y N

22 2 0 7 1 26 49 3 13 5 35 33 5 13 5

Y Y

Y N Y

N N N N N Y N N N N Y N N Y Y

Y Y Y Y N N Y N N N N N N N Y

8 −5 −2 9 3 42 87 8 13 −1 42 30 1 32 8

Y Y

Y Y N

y = yes, meets criteria; n = no, does not meet criteria. a Cox regression, fully adjusted, in which variables predict outcome onset; y = alpha b 0.05. b Cox regression, fully adjusted, in which variables predict exposure onset, pre outcome onset; y = alpha b 0.05. c Bivariate Cox regressions difference between estimates of main effect. d Fixed effect regressions before and after exposure, prior to outcome onset; y = alpha b 0.05.

acceptability of smoking could affect how the relationship between smoking and depression was expressed (Kassel, Stroud, & Paronis, 2003). Results of the concept mapping process are comparable to previous research that examined the influence of third variables on the depression and smoking relationship. For example, stress, worry about parents, and worry about weight increased after smoking initiation. These variables therefore met the criterion for being intermediate variables of the relationship between smoking and depression. While Wills et al. (2002) found that smoking increased with stress but not the reverse (Kassel et al., 2003; Wills et al., 2002), Parrott (1999) claimed that smoking lead to increases in stress (Patton et al., 1998). Our model can be used as a theoretically and empirically basis to test the hypothesized relationships in future longitudinal studies. There are also techniques that would allow an investigator to include or condition on intermediate variables appropriately in a model without exclusion of that variable. For example, structural equation modelling or a marginal structural model could be used to test the model developed. An

Table 4 Results of fixed effect regression examining changes in variables associated with smoking and depressive symptoms. Nicotine Dependence in Teens study, Montreal, Canada. (n = 837). Outcomes are across the top.

intermediate variable, particularly in a longitudinal cohort, may also be a confounder if it is a time dependent risk factor that influences the subsequent exposure, and is itself influenced by the exposure. Nevertheless, it is important to note that in each case, a decision about the interpretation and the modelling of included variables rests on either theoretical assumptions of the causal role of the variable or empirical testing such as that in this paper. Our results were consistent with a study by Boden et al. that examined a number of potential mechanisms underpinning the association between depression and smoking in a birth cohort. These investigators found that increasing levels of nicotine-dependence symptoms were significantly and directly associated with the rise in rate of depressive symptoms and that there were also alternative pathways that suggested a common cause mechanism (Boden et al., 2013). The results were broadly similar between the Canadian and New Zealand cohort, but this paper suggests in addition an active pathway, albeit primarily social, between depression and smoking as well as an increased role of stress in the smoking to depression pathway. Differences may be due to the shorter follow up periods and younger ages in this current study. Friends smoking has been shown to be a mediator in the relationship of smoking to depression (Audrain-McGovern, Rodriguez, & Kassel, 2009). Increased depressive symptoms may lead to increases in either the perception of, or actual number of friends who smoke, which in turn may lead to increased smoking (Patton et al., 1998). Further, the

After smoking, before depression Alcohol use

Alcohol Use

Stress symptoms

Worry about weight

Worry about relationship with parents

+

+

+

ns

+

+

ns

+

ns

Stress symptoms

+

Worry about weight

+

+

Worry about relationship with parents Friends smoke

ns

+

+

ns

ns

+

Friends smoke

ns ns

After depression, before smoking Alcohol use

Alcohol use

Stress symptoms

Worry about weight

Worry about relationship with parents

Friends smoke

ns

+

ns

ns

+

+

ns

+

ns

Stress symptoms

ns

Worry about weight

ns

+

Worry about relationship with parents Friends smoke

+

+

+

ns

ns

ns

ns

ns

+ indicates significant intra-individual increases in outcome with increases in variable.

Table 5 Classification of variables in the relationship between smoking and depressive symptoms in the Nicotine Dependence in Teens study, Montreal, Canada. No association Smoking to depression Percent life in Canada Novelty-seeking score Physical activity

Depression to smoking Novelty-seeking score Parent smokes Academic performance

Predictor of outcome

Confounder

Intermediate variable

Self esteem Sex Age in years Impulsivity Parent education Alcohol use Parent smokes Academic performance

Friends smoke Stress symptoms Worry about weight Worry about parents

Self esteem Impulsivity Age in years Parent education Physical activity Sex Percent life in Canada

Alcohol use Stress symptoms Worry about weight Worry about parents Friends smoke

M. Chaiton et al. / Addictive Behaviors 42 (2015) 154–161

159

Fig. 1. Empirically driven concept map of the role of selected variables that affect the relationship between smoking predicting depressive symptoms. Nicotine Dependence in Teens study, Montreal. Canada. (n = 837).

onset of smoking led to increases in (perceptions of) the number of friends who smoked, which were associated with the onset of depressive symptoms. Friends smoking could increase exposure to SHS which may be a biologic underpinning. Limitations include that the study is limited to depression developing during adolescence, and the results may not be generalizable to adult onset depression. The sample was restricted to individuals who had not started smoking by study entry, who may therefore be viewed as having “later emerging smoking” (i.e., after age 12–13 years). A delayed age of onset may be protective of smoking emerging subsequent to depression. There may be heterogeneity in depression such that the early onset phenotype has different predictors than the later onset phenotype (Black et al., 1999). On the other hand, smoking and depression that occurred prior to baseline may have been subject to recall bias and not reported. Particularly, this is an issue with the reporting of periods of elevated depressive symptoms as the study was designed to prompt for smoking history. Research supports the validity of self-report measures of smoking behaviour in adolescents, especially in contexts in which confidentiality is emphasized (Armitage & Colton, 2005). Nevertheless, if potential confounders are measured with error, those confounders may not have been identified nor would their effect have been fully taken into account (Ezzati, 2004). Generally, we were limited to variables assessed in this cohort and other variable known and unknown could also play an important role in the relationship of smoking and depression. Some of the included variables were measured only 1–3 times over the course of the study; however, these variables may have reflected important changes had they been measured at more frequent intervals. Multiple imputation was not used, as the same variables have been used as dependent and independent variables, and missing value imputation could be problematic in this situation. However, this meant that the

study may have erred towards maintaining inclusion of variables in the model. This paper uses set criteria for statistical testing at p b 0.05 and a 10% change in the parameter estimate. While the p b 0.05 significance testing approach is limited for selecting the best set of potential confounders (Greenland, 2008; Rothman et al., 2008), this paper used significance testing appropriately to investigate if the variable of interest was associated with the outcome. Nevertheless, type I and type II errors where variables are included or excluded erroneously are possible. 5. Conclusion The results suggest that there are significant covariates of the relationship of depressive symptoms and smoking but that care must be taken to differentiate between intermediate variables and confounders. Furthermore, the paper generated several hypotheses about the role of covariates in the pathway between smoking and depression including the impact of stress and friend smoking as intermediate variables. Role of funding sources This work is supported by Canadian Cancer Society grant #702160. The funders had no role in the study design, collection, analysis or interpretation of the data, writing the manuscript, or the decision to submit the paper for publication. Contributors MC drafted the manuscript and conducted the analysis. JC, JR, JOL, MA provided substantial intellectual contributions to the writing, interpretation of results. JOL contributed to the design of the underlying cohort study. All authors were involved in the design and conceptualization of the study and provided final approval. Conflict of interest All authors declare that they have no conflicts of interest.

Worry Weight

Worry Parents

Alcohol Use

Stress Elevated Depression Symptoms

Impulsivity

Smoking

Friend Smoking

Fig. 2. Empirically driven concept map of the role of selected variables in the relationship between depressive symptoms predicting smoking. Nicotine Dependence in Teens study, Montreal, Canada. (n = 837).

160

M. Chaiton et al. / Addictive Behaviors 42 (2015) 154–161

Appendix I. Description of predictor variables

Predictor variable

Data collected in survey cycle(s)

Item(s)

Response choices (score assigned)

Sex Age in years Percent life in Canada Parent education

1–20 1–20 1–20

Are you a boy or a girl? Date of birth, Date of survey Time spent in Canada

Male, female

13, 17

How much education has your father had? How much education has your mother?

Did not finish high school, high school graduate, vocational, technical school, CEGEP, university, don't know, other (for each parent)

Parent smokes

1–20

No, yes (for each parent)

Friends smoke

1–20

Stress symptoms

1–20

Self esteem

0:00

Impulsivity

14,18

Novelty score

14,19

Worry about weight

1–20

Does your father currently smoke cigarettes? Does your mother currently smoke cigarettes? Now think about your friends. How many of the people whom you usually hang out with smoke his cigarettes? During the past 3 months, have you been worried or stressed by any of the following: (i) your parents separating or divorcing (ii) loneliness (iii) breaking up with your boyfriend or girlfriend (iv) your relationship with your brother(s)/sister(s) (v) your relationship with your friends (vi) a health problem (such as acne or asthma) (vii) sex (viii) your new family (parents remarried) (ix) financial problems in your family (x) school work (xi) other For each of the following statements, indicate the response which best describes your situation. (i) I think I am someone who has something valuable to offer, at least as much as other people do (ii) I think I have a certain number of good qualities (iii) Everything considered, I tend to think I'm a failure (iv) I think I am capable of doing things as well as other people my age (v) There's little reason to be proud of myself; (vi) I have a positive attitude towards myself; (vii) difficult to accept myself as I am; (viii) Sometimes I think I'm really useless; (ix) I've thought of myself as a good for-nothing on occasion How true are each of the following statements for you: (i) I often do things without stopping to think (ii) I am an impulsive person (iii) I often talk quickly, before thinking things out (iv) I often get involved in things I later wish I could get out of (v) I need to use a lot of self-control to keep out of trouble (vi) I often get into trouble because I do things without thinking (vii) I get carried away be new and exciting ideas, but I don't think of the possible problems Here are some things people may say about themselves. How true are each of the following statements for you: (i) I often try new things just for fun or thrills, even if most people think it is a waste of time (ii) When nothing new is happening, I usually start looking for something that is exciting (iii) I can usually get people to believe me, even when what I'm saying isn't quite true (iv) I often do things based on how I feel at the moment (v) I sometimes get so excited that I lose control of myself (vi) I like it when people can do whatever they want, without strict rules and regulations (vii) I often follow my instincts, without thinking through all the details (viii) I can do a good job of “stretching the truth” when I'm talking to people (ix) I change my interests a lot, because my attention often shifts During the past 3 months, have you been worried or stressed about your weight?

Worry about parents

1–20

Academic performance Alcohol use Physical activity

16,20 1–20 1–20

Re-coded

100% or less than 100%

None, a few, about half, more than half, most or all

No, yes (mother and/or father completed university) No, yes (1–2 parents smoke) None, a few or more

Not at all (Chaiton et al., 2009), a little bit (Paperwalla et al., 2004), quite a bit (Black et al., 1999), a whole lot (Brown et al., 1996)

I find it not at all true (Chaiton et al., 2009), a little true (Paperwalla et al., 2004), very true (Black et al., 1999)

Not at all true (Chaiton et al., 2009), a little true (Paperwalla et al., 2004), somewhat true (Black et al., 1999), pretty true (Brown et al., 1996), very true (Dierker et al., 2002)

Not at all true (Chaiton et al., 2009), a little true (Paperwalla et al., 2004), somewhat true (Black et al., 1999), pretty true (Brown et al., 1996), very true (Dierker et al., 2002)

Not at all (Chaiton et al., 2009), a little bit (Paperwalla et al., 2004), quite a bit (Black et al., 1999), a whole lot (Brown et al., 1996) Not at all (Chaiton et al., 2009), a little bit (Paperwalla During the past 3 months, have you been worried or et al., 2004), quite a bit (Black et al., 1999), a whole lot stressed about your relationship with your mother? (Brown et al., 1996) ….your father? How true is the following for you: I'm doing well at school Very true, a bit true, not at all true this year Do you drink alcohol? Never, Ever, weekly? Never, Ever, weekly No, yes (for each physical activity, Monday to Sunday) Now, think about the physical activities that you did last week from Monday to Sunday outside your regular school gym class. For each activity that you did for 5 min or more at one time, mark an “X” to show the day(s) on which you did that activity (list of 29 different activities).

No, yes (a little bit or more) No, yes (a little bit or more) Not at all true, a bit/very true Total no. of activities checked “yes”

M. Chaiton et al. / Addictive Behaviors 42 (2015) 154–161

References Armitage, P., & Colton, T. (2005). Encyclopedia of biostatistics (2nd ed.). Chichester, Hoboken, NJ: John Wiley. Audrain-McGovern, J., Lerman, C., Wileyto, E. P., Rodriguez, D., & Shields, P. G. (2004, Jul). Interacting effects of genetic predisposition and depression on adolescent smoking progression. The American Journal of Psychiatry, 161(7), 1224–1230. Audrain-McGovern, J., Rodriguez, D., & Kassel, J. D. (2009, Jun). Adolescent smoking and depression: Evidence for self-medication and peer smoking mediation. Addiction, 22. Beal, S. J., Negriff, S., Dorn, L. D., Pabst, S., & Schulenberg, J. (2013). Longitudinal associations between smoking and depressive symptoms among adolescent girls. Prevention Science, 1–10. Black, D. W., Zimmerman, M., & Coryell, W. H. (1999, Sep). Cigarette smoking and psychiatric disorder in a community sample. Annals of Clinical Psychiatry, 11(3), 129–136. Boden, J. M., Fergusson, D. M., & Horwood, L. J. (2013). Cigarette smoking and depression: Tests of causal linkages using a longitudinal birth cohort. The British Journal of Psychiatry, 196(6), 440–446. Boslaugh, S. (2008). Encyclopedia of epidemiology. Los Angeles: Sage Publications. Breslau, N., Peterson, E. L., Schultz, L. R., Chilcoat, H. D., & Andreski, P. (1998, Feb). Major depression and stages of smoking. A longitudinal investigation. Archives of General Psychiatry, 55(2), 161–166. Brook, J. S., Schuster, E., & Zhang, C. (2004, Aug). Cigarette smoking and depressive symptoms: A longitudinal study of adolescents and young adults. Psychological Reports, 95(1), 159–166. Brown, R. A., Lewinsohn, P. M., Seeley, J. R., & Wagner, E. F. (1996, Dec). Cigarette smoking, major depression, and other psychiatric disorders among adolescents. Journal of the American Academy of Child and Adolescent Psychiatry, 35(12), 1602–1610. Chaiton, M. O., Cohen, J. E., O'Loughlin, J., & Rehm, J. (2009). A systematic review of longitudinal studies on the association between depression and smoking in adolescents. BMC Public Health, 9(1), 356. Choi, W. S., Patten, C. A., Gillin, J. C., Kaplan, R. M., & Pierce, J. P. (1997, Winter). Cigarette smoking predicts development of depressive symptoms among U.S. adolescents. Annals of Behavioral Medicine, 19(1), 42–50. Costello, E. J., Erkanli, A., Federman, E., & Angold, A. (1999, Sep). Development of psychiatric comorbidity with substance abuse in adolescents: effects of timing and sex. Journal of Clinical Child Psychology, 28(3), 298–311. Deschesnes, M. (1997). Tome 1 (Secteur general). [Chapter 1]. Style de vie des jeunes du secondaire en Outaouais [Life style of youth in secondary schools in Outaouais]. Hull, Quebec, Canada: Direction de la sante publique. Regie regionale de la sante et des services sociaux (pp. 150)Public Health Directorate, Regional Directorate of Health and Social Services (pp. 150). Dierker, L. C., Avenevoli, S., Stolar, M., & Merikangas, K. R. (2002, Jun). Smoking and depression: An examination of mechanisms of comorbidity. The American Journal of Psychiatry, 159(6), 947–953. Duncan, B., & Rees, D. I. (2005, Sep 1). Effect of smoking on depressive symptomatology: A reexamination of data from the National Longitudinal Study of Adolescent Health. American Journal of Epidemiology, 162(5), 461–470. Eppel, A., O'Loughlin, J., Paradis, G., & Platt, R. (2006, Sep). Reliability of self-reports of cigarette use in novice smokers. Addictive Behaviors, 31(9), 1700–1704. Escobedo, L. G., Reddy, M., & Giovino, G. A. (1998, Mar). The relationship between depressive symptoms and cigarette smoking in US adolescents. Addiction, 93(3), 433–440. Eysenck, S. B., & Eysenck, H. J. (1977, Feb). The place of impulsiveness in a dimensional system of personality description. The British Journal of Social and Clinical Psychology, 16(1), 57–68. Ezzati, M. (2004). World Health Organization. Comparative quantification of health risks: Global and regional burden of disease attributable to selected major risk factors. Geneva: World Health Organization. Federman, E. B., Costello, E. J., Angold, A., Farmer, E. M., & Erkanli, A. (1997, Mar 14). Development of substance use and psychiatric comorbidity in an epidemiologic study of white and American Indian young adolescents the Great Smoky Mountains Study. Drug and Alcohol Dependence, 44(2–3), 69–78. Fergusson, D. M., Goodwin, R. D., & Horwood, L. J. (2003, Nov). Major depression and cigarette smoking: results of a 21-year longitudinal study. Psychological Medicine, 33(8), 1357–1367. Fischer, J. A., Najman, J. M., Williams, G. M., & Clavarino, A. M. (2012). Childhood and adolescent psychopathology and subsequent tobacco smoking in young adults: findings from an Australian birth cohort. Addiction, 107(9), 1669–1676. Goodman, E., & Capitman, J. (2000, Oct). Depressive symptoms and cigarette smoking among teens. Pediatrics, 106(4), 748–755. Greenland, S. (2008, Mar 1). Invited commentary: Variable selection versus shrinkage in the control of multiple confounders. American Journal of Epidemiology, 167(5), 523–529 (discussion 30–1).

161

Greenland, S., Pearl, J., & Robins, J. M. (1999, Jan). Causal diagrams for epidemiologic research. Epidemiology, 10(1), 37–48. Hooshmand, S., Willoughby, T., & Good, M. (2012). Does the direction of effects in the association between depressive symptoms and health-risk behaviors differ by behavior? A longitudinal study across the high school years. Journal of Adolescent Health, 50(2), 140–147. Johnson, E. O., Rhee, S. H., Chase, G. A., & Breslau, N. (2004). Comorbidity of depression with levels of smoking: An exploration of the shared familial risk hypothesis. Nicotine & Tobacco Research, 6(6), 1029–1038. Kandel, D. B., Huang, F. Y., & Davies, M. (2001, Oct 1). Comorbidity between patterns of substance use dependence and psychiatric syndromes. Drug and Alcohol Dependence, 64(2), 233–241. Kassel, J. D., Stroud, L. R., & Paronis, C. A. (2003, Mar). Smoking, stress, and negative affect: Correlation, causation, and context across stages of smoking. Psychological Bulletin, 129(2), 270–304. Kendler, K. S., Neale, M. C., MacLean, C. J., Heath, A. C., Eaves, L. J., & Kessler, R. C. (1993, Jan). Smoking and major depression. A causal analysis. Archives of General Psychiatry, 50(1), 36–43. Killen, J. D., Robinson, T. N., Haydel, K. F., Hayward, C., Wilson, D. M., Hammer, L. D., et al. (1997, Dec). Prospective study of risk factors for the initiation of cigarette smoking. Journal of Consulting and Clinical Psychology, 65(6), 1011–1016. MacKinnon, D. P. (2008). Introduction to statistical mediation analysis. New York, NY: Lawrence Erlbaum Associates. O'Loughlin, J., DiFranza, J., Tarasuk, J., Meshefedjian, G., McMillan-Davey, E., Paradis, G., et al. (2002, Dec). Assessment of nicotine dependence symptoms in adolescents: A comparison of five indicators. Tobacco Control, 11(4), 354–360. O'Loughlin, J., Karp, I., Koulis, T., Paradis, G., & DiFranza, J. (2009). Determinants of first puff and daily cigarette smoking in adolescents. American Journal of Epidemiology, 170(5), 585–597. Paperwalla, K. N., Levin, T. T., Weiner, J., & Saravay, S. M. (2004, Nov). Smoking and depression. The Medical Clinics of North America, 88(6), 1483–1494 (x-xi). Park, S., & Romer, D. (2007, Mar). Associations between smoking and depression in adolescence: An integrative review. Taehan Kanho Hakhoe Chi, 37(2), 227–241. Parrott, A. C. (1999, Oct). Does cigarette smoking cause stress? The American Psychologist, 54(10), 817–820. Patton, G. C., Carlin, J. B., Coffey, C., Wolfe, R., Hibbert, M., & Bowes, G. (1998, Oct). Depression, anxiety, and smoking initiation: A prospective study over 3 years. American Journal of Public Health, 88(10), 1518–1522. Rodriguez, D., Moss, H. B., & Audrain-McGovern, J. (2005, Mar–Apr). Developmental heterogeneity in adolescent depressive symptoms: Associations with smoking behavior. Psychosomatic Medicine, 67(2), 200–210. Rothman, K. J., Greenland, S., & Lash, T. L. (2008). Modern epidemiology (3rd ed.). Philadelphia: Wolters Kluwer/Lippincott Williams & Wilkins. Sanchez, A., Norman, G. J., Sallis, J. F., Calfas, K. J., Cella, J., & Patrick, K. (2007, Feb). Patterns and correlates of physical activity and nutrition behaviors in adolescents. American Journal of Preventive Medicine, 32(2), 124–130. Silberg, J., Rutter, M., D'Onofrio, B., & Eaves, L. (2003, Jul). Genetic and environmental risk factors in adolescent substance use. The Journal of Child Psychology and Psychiatry, 44(5), 664–676. Skara, S., Sussman, S., & Dent, C. W. (2001, Mar− Apr). Predicting regular cigarette use among continuation high school students. American Journal of Health Behavior, 25(2), 147–156. Vallieres, E., & Vallerand, R. (1990). Traduction et validation Canadienne-Francaise de l'echelle de l'estime de soi de Rosenberg [Translation to French (Canadian) and validation of Rosenberg's self-esteem scale]. International Journal of Psychology, 25, 305–316. Wills, T. A., Sandy, J. M., & Yaeger, A. M. (2002, Mar). Stress and smoking in adolescence: A test of directional hypotheses. Health Psychology, 21(2), 122–130. Wills, T. A., Windle, M., & Cleary, S. D. (1998, Feb). Temperament and novelty seeking in adolescent substance use: convergence of dimensions of temperament with constructs from Cloninger's theory. Journal of Personality and Social Psychology, 74(2), 387–406. Windle, M., & Windle, R. C. (2001, Apr). Depressive symptoms and cigarette smoking among middle adolescents: prospective associations and intrapersonal and interpersonal influences. Journal of Consulting and Clinical Psychology, 69(2), 215–226. Wu, L. T., & Anthony, J. C. (1999, Dec). Tobacco smoking and depressed mood in late childhood and early adolescence. American Journal of Public Health, 89(12), 1837–1840. Zhu, S. H., & Valbo, A. (2002, Jul–Aug). Depression and smoking during pregnancy. Addictive Behaviors, 27(4), 649–658.

Confounders or intermediate variables? Testing mechanisms for the relationship between depression and smoking in a longitudinal cohort study.

The relationship between the onset of smoking and the onset of depression among adolescents has been well document, but the mechanisms underlying the ...
533KB Sizes 0 Downloads 4 Views