Background
Tobacco smoking disproportionately affects groups of low socioeconomic position (SEP) [
1]. In the past few decades, there has been an overall downward trend in smoking prevalence across most demographic groups in countries with advanced tobacco control programs [
2,
3]. However, the declines are generally greater in less disadvantaged groups, contributing to a widening disparity in smoking by SEP [
2,
4]. For example, in Australia, the gap in smoking prevalence between low and high SEP groups among people aged 14 years and over widened from 8.6% in 1998 [
5] to 13.2% in 2013 [
6].
Of importance to the role of SEP in smoking behaviour is the reasons for the association. Children from socioeconomically disadvantaged families are more often exposed to parental smoking [
7], have more favourable attitudes toward smoking [
8], a greater intention to smoke [
9] and early smoking experimentation [
10] than those from less disadvantaged backgrounds. In turn, these factors are associated with increased risk of future smoking [
7,
11‐
16]. Few longitudinal studies have explored the extent to which these factors might account for the SEP differences in later smoking [
7] and none have taken into account the impact of adult SEP, which is closely related to childhood SEP and adulthood smoking [
1].
Understanding whether SEP at different life stages differentially impacts later smoking and the underlying mechanisms may help to inform policies to reduce the high prevalence in low SEP groups. Various models have been proposed to describe how exposures such as SEP may operate over the life course [
17]. The critical period model assumes the effect of SEP is important at a limited time window and that there is no influence outside this time period. The sensitive period model assumes the effect of SEP is stronger at one time period than at other times. The accumulation of risk model assumes SEP affects the outcome cumulatively and equally over the life course. Social mobility models vary across different definitions. The intra-generational (adult) mobility model assumes that any downwards change in SEP in adulthood would be harmful to the outcome and any upwards mobility in adulthood would be beneficial, independent of childhood social background. Any mobility model hypothesises that all downward changes in the life course are equally harmful to the outcome and all upward shifts are equally beneficial. There is evidence that these models are useful for understanding the development of health across the life course. Using data on adult body mass index (BMI) and SEP measured once in childhood and twice in adulthood from the Medical Research Council National Survey of Health and Development study, Mishra et al. [
17] concluded that only considering one life course model may produce misleading results and recommended considering all possible models in such analyses. A systematic review of models of life course socioeconomic factors also recommended to test multiple life course models and use multiple SEP measures in one sample in future analyses [
18].
Utilising longitudinal data at different life stages, several studies have tried to understand the relationship of SEP across the life course and smoking status in later life. These were limited in that they included only one potential life course model such as the critical period model (only childhood reflected by parental SEP) [
19,
20], the sensitive period model [
21‐
23], the social mobility model [
24], and the accumulation model [
25,
26]. However, no study has investigated how all of the possible life course models might describe the association between SEP and smoking in later life in one sample. The aim of this study was to examine the importance of timing and duration of exposure to low SEP, and mobility in SEP, for mid-adulthood smoking in an Australian national cohort. We also investigated whether smoking-related variables in childhood mediated the relationship.
Discussion
This is the first study to examine the relationship between SEP trajectories over the life course and smoking later life using a structured regression framework to examine a series of theoretical life course models. For individual-level SEP, the sensitive period model and the accumulation model best fit the data. The risk of being a current smoker in mid-adulthood was higher in those exposed to low SEP in childhood and mid-adulthood and for those exposed with greater cumulative exposure. For area-level SEP, the model that best described the data was the sensitive period model in which the smoking risk was highest in those exposed to low SEP in early and mid-adulthood. This association was moderately explained by exposure to parental smoking and intention to smoke in childhood.
The sensitive period model was supported by our individual- and area-level SEP data. Being exposed to low individual-level SEP in the “sensitive periods” of childhood and mid-adulthood and to low area-level SEP in young- and mid-adulthood increased the risk of being a current smoker in mid-adulthood when SEP at all three life stages were mutually adjusted. There is considerable evidence that smoking in adulthood is influenced by childhood and adulthood socioeconomic disadvantage [
7,
18,
45,
46]. For example, according to Kestila et al. [
45], young adults whose parents had the lowest educational attainment were about five times more likely to be a daily smoker than those with parents in the highest education category.
The results for individual-level SEP shows strong support for the accumulation model for adult smoking. This finding is consistent with the study by Smith et al. [
25] which assessed the influence of SEP over three life stages on risk factors of cardiovascular disease including smoking among 5766 men. They revealed a positive graded association between the number of time periods belonging to manual occupation social class and the risk of being current smokers. There is also evidence for a similar association in women, where belonging to a manual occupation social class in both childhood and adulthood increased the odds of being current smokers by 75% compared with staying in non-manual social class at both time points [
47]. Nevertheless, it should be noted that direct comparison with other studies is limited because their analyses were not framed in terms of life course models.
We found inconsistent results when using socioeconomic indicators at the individual and area levels: both the sensitive period model and the accumulation model were found to best fit the data when using individual-level SEP where only the sensitive period model fit when using area-level SEP. Previous evidence of the validity of using area-based SEP measures as proxies of individual-level indicators is conflicting [
48,
49]. One of the possible explanations is the different constructs of area and individual-level socioeconomic measures [
48]. Using data from three large population-based epidemiologic studies, Diez and colleagues reported that area and individual-level indicators were somewhat correlated but actually provided complementary information on living circumstances [
48]. Presence of contextual area effects may help explain discrepancies between area- and individual-based estimates of socioeconomic differences in smoking [
48]. This involves mechanisms through which contextual effects of area on smoking could be mediated, including greater risk of being exposed to smoking and greater availability of places that sell cigarettes in low SEP areas [
1]. Another possible reason is that the SEP of area as a whole will not always represent the SEP of individuals (the “ecological fallacy”). Some individuals with a higher individual-level SEP may reside in a relatively disadvantaged area and in contrast, some individuals with a lower individual-level SEP may live in an area which relatively lacks disadvantage [
50].
As expected, the observed higher risk of smoking in mid-adulthood among people exposed to low SEP in childhood and mid-adulthood and for a greater number of periods was partially explained via exposure to parental smoking during childhood. This finding is in line with past studies by Paul et al. [
12] and Fergusson et al. [
7], which concluded that smoking in adulthood was predicted by exposure to parental smoking that could account for over 25% of the relationship between childhood social background and later smoking [
7]. In the current study intention to smoke, self-rated importance to be a non-smoker and smoking experimentation in childhood were explored as potential explanations the SEP gradient in smoking for the first time. Our results suggest childhood socioeconomic disadvantage influenced smoking in mid-adulthood partially through intention to smoke in childhood.
Our findings reiterate the important roles of exposure to parental smoking and intention to smoke in childhood in the relationship of SEP across the early life span and smoking in mid-adulthood. The increased risk of smoking in offspring of people that smoke along with the well-established health problems and illnesses of second-hand smoke may be used to encourage parents and those who will become parents to quit smoking. This approach is likely to have a large impact as parents are strongly motivated to adopt healthy behaviours for the sake of their children [
51]. For adult smokers in low SEP, increasing tobacco taxes are believed to have the greatest potential to achieve reduction in smoking [
52,
53].
Some limitations should be considered. First, self-report could result in the misclassification of smoking status; however, it is most likely that people have under-reported smoking which would likely mean we have underestimated the effect of SEP on current smoking [
54]. Second, dichotomising SEP is very simplified but necessary for the modelling framework. We could not explore whether there was a gradient of effects across socioeconomic levels. Third, we did not know the full duration of exposure so the accumulation model in this study does not refer to the exact length of exposure to low or high SEP. Fourth, we were somewhat limited in the approaches we could use in mediation analyses because our potential explanatory variables were measured at a single point in time that was concurrent with one of our exposures – childhood SEP. This precludes using a method such as path analysis with structural equation modelling where it is recommended to use temporally separate exposures, mediators and outcomes. Fifth, our mediation results should be interpreted with caution since the traditional approach to mediation analyses we have used relies on fairly strong assumptions including control for mediator-outcome, exposure-mediator, and exposure-induced mediator-outcome confounding to be interpreted causally [
55]. Failure to control for these assumptions may produce flawed results [
38,
55]. In our study, genetics and particular personality traits (i.e. extraversion, neuroticism and conscientiousness) that affect both early smoking experimentation and established smoking patterns in adulthood, are potential mediator-outcome confounders [
56,
57]. Unfortunately, genetic data were not available and adjusting for personality traits measured by the NEO-Five Factor Inventory in CDAH-1 showed no evidence of mediator-outcome confounding. No exposure-mediator interaction was observed for exposure to parental smoking and a moderate interaction was present between childhood SEP and children’s intention to smoke. Therefore, the natural direct effect which incorporated the interaction effects was estimated using Richiardi and colleagues’ approach [
38]. Due to the low prevalence of intention to smoke (2.3%) in our cohort, the estimated natural direct effect is almost identical to the direct effect shown in Table
4, indicating our conclusions are largely unchanged by this interaction. The smoking behaviour of peers has been identified as a strong predictor of both intention to smoke and smoking uptake among adolescents [
57,
58] but is unlikely to act as a substantial exposure-induced mediator-outcome confounder. This is because of its weak association with childhood SEP [
58]. The direct effect estimated in Table
4 (Model 3 with adjustment for intention to smoke) might be overestimated, depending on the strength of the independent relationship between peers smoking and mid-adulthood smoking.
The strengths of this study include its large national sample, the 25 years follow-up period, the use of a novel methodology – a structured regression framework to modelling the effects of binary exposure variables over the life course and the efforts to explore the underlying mechanisms. Although several studies have examined the association of SEP and smoking status using a life course approach, none of them has tested multiple life course models in the same sample. As concluded by Pollitt and colleagues in a systematic review [
18], analyses using data followed from childhood to adulthood, multiple SEP measures and multiple life course designs within the same sample offer the best approach to test which theories best describe the association between life course SEP and the outcome. The structured regression framework we used to compare a set of nested models to an all-inclusive (fully saturated) model is an improvement over traditional regression models in which results are interpreted from a single pre-specified hypothesis without considering the merits of alternative life course hypotheses. For example, if we only considered a single model, such as the accumulation model, we might conclude that there is evidence for its fit. However, the sensitive period model would not be identified even though it fits the data just as well as the fully saturated model with both individual- and area-level SEP.
To conclude, childhood, young- and mid-adulthood are all important, but SEP in childhood and mid-adulthood may be of more importance in determining smoking status in mid-adulthood. Exposure to parental smoking and intention to smoke in childhood seems to moderately mediate the associations.