A systematic review of longitudinal studies on the association between depression and smoking in adolescents

Co-occurrence of tobacco use and depression has long been a major concern because of the substantially increased risk to health.[1] Most [2–12] but not all [13–15], cross-sectional studies report a strong positive association between smoking and depression. Furthermore, how these health risks inter-relate temporally remains a topic of debate. Specifically, there is no consensus on whether smoking causes depression or depression causes smoking in the natural course of tobacco use and depression.[16]

Several hypotheses have been put forward to explain the possible mechanisms underlying the association between smoking and depression. The "self-medication" hypothesis suggests that depression may lead to smoking because people smoke to relieve symptoms of depression. [17–20] In contrast, other authors have suggested that cigarette smoking causes depression as nicotine exposure may damage neurochemical pathways, such as monoamine neurotransmission. [21–23] Bi-directional mechanisms in which smoking and depression reinforce each other have also been postulated.[12, 23, 24] Finally, Kendler (1993) hypothesizes that there is no causal relationship and that the perceived association is a function of underlying shared environmental and genetic risk factors.[24]

Adolescence is regarded as the key period for the study of the onset of smoking and depression co-occurrence. The period of adolescence represents the peak of the incidence of both smoking and depression.[25] More specifically depressive symptoms tend to increase in early adolescence and then become stable or decline in later adolescence.[26] Similarly, most people who start smoking do so before the age of 18. To study the co-occurrence of smoking and depression it is necessary to investigate the association prior to onset.

While cross-sectional studies find that depression and smoking are associated among adolescents, longitudinal studies are needed to address temporality. Since Kandel and Davis [27] reported that depression in adolescents could lead to the onset of smoking, several longitudinal studies have assessed the relationship between smoking and depression in adolescents.[12, 16, 28] In previous reviews, Paperwalla et al. [12] and Schepis and Rao [28] examined the etiology of depression and smoking in general without conducting systematic search strategies or analysis. Park and Romer [16] conducted a systematic, descriptive analysis of cross-sectional and longitudinal studies published up to 2005. However, their review was limited by the lack of a quantitative component, and inclusion of several papers using data from the same cohort as independent studies.

This review focuses on longitudinal studies wherein temporality can be ascertained, and it provides a quantitative summary of estimates of the association between smoking and depression among adolescents.

English language studies published between January 1990 and December 2007 were extracted from the PubMed, OVID, and PsychInfo databases using the following abstract keywords: (a) smoking, tobacco, cigarettes; (b) depress* (c) child*, adolescen*, youth; (d) longitudinal, cohort, follow-up. Titles and abstracts were then screened to determine if publications met the following inclusion criteria: (1) the association between smoking and depression was studied longitudinally; (2) the outcome (either the onset of depression or the onset of smoking) was measured when subjects were aged 13 -19 years; (3) the study population was broadly representative of adolescents (studies of clinic populations, the unemployed, the incarcerated, and pregnant girls were excluded); and (4) the effect of smoking on the onset of depression, or the effect of depression on the onset of smoking was reported explicitly as a risk, hazard, or odds ratio. Studies that examined related but distinct mental health constructs including anxiety, stress, and negative affect were not included. Additional studies that met the inclusion criteria were identified by scanning references in the reports retained in the primary search, and through searching the Internet through Google Scholar. For cohorts with multiple relevant publications, the latest publication with the longest follow-up that reported on outcomes in adolescents was selected to avoid duplication.

The following data were abstracted from each study retained: country in which the study had been conducted; sample size; year(s) of data collection; time period over which the relationship was estimated; participants' age/grade, the definition of smoking; the definition of depression; covariates included in the multivariate modeling; the analytic method; and the parameter estimate of the unadjusted association between smoking and depression (with 95% confidence intervals if available). Finally, the estimate of effect(s) after controlling for covariates was recorded. The quality of the studies included in this review was characterized across factors that may affect generalizability and internal validity in cohort studies.[29] Several studies reported only one direction of effect (i.e., smoking initiation was the only outcome tested). Studies that reported results only by subgroup (i.e., males and females) were included as separate studies in the pooled analysis.

Sources of longitudinal studies

The database search yielded 191 publications. Eight additional reports were obtained through a search of references. Sixteen publications were excluded because they did not analyze the data longitudinally (i.e., they pooled data across data collection waves and analyzed the data as if they were cross-sectional). Screening of titles and abstracts for inclusion and exclusion criteria produced a total of 55 relevant publications representing 17 different cohorts. Four studies from separate cohorts [31–34] did not have data that could be extracted and were therefore excluded. Thirteen studies met the inclusion criteria (Figure 1).

The present review suggests that there is evidence in longitudinal studies of adolescents that depression predicts smoking, and that smoking predicts depression. The pooled estimates for depression predicting smoking (1.41 (95% CI: 1.21, 1.63)) and smoking predicting depression (1.73 (95% CI: 1.32, 2.40)) were both statistically significant despite the relatively small number of studies and wide variability in study designs and measures.

This analysis revealed moderately strong evidence supporting a relationship whereby smoking predicts depression among adolescents. Only one study [11] found that smoking did not significantly predict depression, but this study had the smallest sample size of the studies included, and the magnitude of the effect was large (1.74). The magnitude of the pooled effect for smoking predicting depression was also larger than that of depression predicting smoking. There is evidence that smoking leads to dysregulation of the hypothalamic-pituitary-adrenal system and hypersecretion of cortisol [49], which can regulate biological and psychological reactions to sources of stress. [50–52] Consequently, smoking might interfere with natural adaptive coping mechanisms and induce depression.[53]

Conversely, the results also may be interpreted to suggest that depression leading to smoking is the more likely pathway. Smokers may use nicotine as a means of 'self-medicating' symptoms such as depressed mood and fatigue that are related to depression; nicotine has been shown to produce an elevation in mood.[24] Nicotine may also have anti-depressant properties, acting in a similar manner to common antidepressants through the serotonin neurotransmitters. [54–56] In the few studies that tested competing models directly, depression predicting smoking was the stronger effect. When more specific measures of depression are used, depression is better able to predict smoking, suggesting that random misclassification of depression may bias the effect of depression predicting smoking towards the null in some studies. Furthermore, smoking may be more easily quantified and recalled by adolescents--more studies measure previous smoking history (i.e., identify never smokers) and control for it than measure previous depression or control for previous depressive episodes. If depression at a young age went undiagnosed at the time, it is possible and perhaps likely that previous depression would have been forgotten or mischaracterized. While it is possible to forget smoking experiences, smoking is a tangible event compared to a depressive episode. Controlling for previous smoking without controlling for previous depression will create bias by design.

The current analysis cannot refute the hypothesis that the relationship between smoking and depression is spurious. It is possible that if the two conditions are causally related, the relationship may reflect several potential causal mechanisms in which: (a) depression precedes smoking, (b) smoking precedes depression, (c) there is a bidirectional relationship between smoking and depression, (d) the relationship is a function of unknown risk factors, or (e) there are subgroups in which the relationship between smoking and depression differs. To address these issues, the nature of the relationship between depression and smoking should be investigated further. While temporality is an important condition for causality, there are other issues in the current literature of prospective studies that limit our ability to draw robust conclusions about the nature, causal or otherwise, of the relationship between smoking and depression. These issues should inform future research and are discussed below.

This paper used strict inclusion criteria to include as homogenous a population of studies as possible in order to perform the pooled analysis. As such, studies that used techniques such as growth curve modeling [33, 69, 70], cross-lagged analysis [71], or correlations[34, 72] were not included. While broader would have included more studies and different perspectives, the advantage of a focused research question allows the important information to be highlighted and quantified in a way that a broader survey could not.

The pooled estimates should be interpreted with caution because they are based on studies with heterogeneous measures of exposure and outcomes, and differing study designs. Estimates adjusted for covariates were selected from each study, but the variables included to control for confounding varied from study to study. There may therefore be residual confounding of the pooled estimate, which could inflate the estimate away from the 'true' value.[73]

Several studies were excluded because the outcome was measured in young adulthood rather than adolescence.[47, 74] It could be argued that the young adult measure is more acceptable as it reduces the number of censored observations. However, one major purpose of studying adolescents is to capture the initiation of the relationship between smoking and depression. The strength of the association between smoking and depression is well established before age 20 and consequently the initiation of the relationship must be established by this point. Dramatic developmental and neurobiological changes in the adolescent brain may influence drug use vulnerability and future susceptibility to depression and, consequently, the likelihood that an association exists may differ across ages.[75]

While the association between smoking and depression is not in doubt, the reasons underlying the association matter both for those affected by both depression and smoking, but also for the development and implementation of interventions. Understanding the causal mechanisms underlying the relationship could allow for the development of interventions for adolescents with these conditions, and possibly prevent the implementation of interventions based on an erroneous causal mechanism that may have negligible or even iatrogenic effects. If, for example, some individuals smoke in order to control symptoms of depression, interventions that focus solely on smoking could be ineffective, or indeed, potentially dangerous. However, if smoking exacerbates depression symptoms, excluding those who are depressed from treatment could lead to further harm. Further research into identifying the nature of the association should be initiated.