Article Text
Abstract
Objective: To estimate the extent to which susceptibility to smoking is associated with between-context differences (schools and classes) and to identify factors at school, class and individual levels that influence individual susceptibility to smoking among young never-smokers in South East Asia.
Methods: Cross-sectional data from the Global Youth Tobacco Survey conducted in Cambodia (2002), Laos (2003) and Vietnam (2003) are used to conduct multilevel analyses that account for the nesting of students in classes and classes in schools. The outcome variable is smoking susceptibility, defined as the absence of a firm decision not to smoke. Explanatory variables include school-level (current tobacco use prevalence in school, exposure to anti-smoking media messages and exposure to tobacco billboard advertising), class-level (classroom prevention) and individual-level influences (parents’ and friends’ smoking behaviour, knowledge of the harmful effects of and exposure to secondhand smoke at home, age, sex and pocket income).
Results: Multilevel analyses indicate that 4.5 and 4.2 of the variation in smoking susceptibility is associated with school and class differences, respectively. Students who have parents or friends who smoke, who are exposed to secondhand smoke at home and those who have access to pocket income are found to be more susceptible while greater knowledge of the harmful effects of secondhand smoke appears to diminish susceptibility to smoking. For girls only, billboard tobacco advertising increases the risk of susceptibility and classroom prevention decreases risk while for boys only, attendance at schools with higher prevalence of tobacco use increases risk of susceptibility and anti-smoking media messages decreases risk.
Conclusions: This study highlights a number of modifiable factors associated with smoking susceptibility and identifies interactions between teen sex and several factors associated with the susceptibility to smoking. This finding provides support for the call to move beyond gender-blind tobacco control policies.
Statistics from Altmetric.com
In recent decades many high-income countries have experienced steady but slow decreases in tobacco use. During the same period, however, many low-income and middle-income countries, including several in Asia, have experienced dramatic increases, especially among males.1 In many South East Asian countries smoking prevalence in men can be in excess of 50. For example, in Cambodia, Laos and Vietnam, current estimates point to a smoking prevalence in men of 54, 66 and 51, respectively.2–4 i Smoking in South East Asia is still strongly sex-linked. Smoking rates in women in Cambodia, Laos and Vietnam are orders of magnitude lower than in men (6, 16 and 3, respectively).2–4
Susceptibility to smoking—defined as the absence of a firm decision not to smoke—has been shown to be a good predictor of smoking onset. Pierce et al5 assessed the predictive validity of smoking susceptibility and found baseline susceptibility to smoking was a strong independent predictor of experimentation (the next step in the smoking onset process). Similarly, Unger et al6 found susceptible teenagers to be two to three times more likely to experiment with cigarette smoking during the subsequent two years than were non-susceptible teenagers. Choi et al,7 using data from a national and state (California) longitudinal survey of teenagers showed that susceptibility to smoking increased the risk of future established smoking at all levels of previous smoking behaviour.ii Jackson8 and Huang et al9 provide additional evidence that smoking susceptibility is a good predictor of initiation. However, Pierce et al’s measure of smoking susceptibility has never been validated in a low-income country setting. For a detailed discussion of different measures of smoking intentions, see Wakefield et al.10
Susceptibility has the advantage of being measured well before youths begin to smoke. This feature combined with the reality that smoking initiation occurs at a later age in South East Asia,2 11 12 offers important opportunities not captured by the more traditional use of smoking experimentation or participation measures.
The Global Youth Tobacco Survey (GYTS) offers the opportunity to explore susceptibility to smoking in three countries of South East Asia: Cambodia, Laos and Vietnam. These three low-income countries, in addition to sharing long borders, generally share common social and gender norms. As indicated above smoking rates are roughly similar—very high among men and inordinately low among women in Cambodia and Vietnam and to a lesser extent in Laos. Cambodia, Laos and Vietnam have all recently ratified the Framework Convention on Tobacco Control (FCTC) demonstrating a political commitment to tobacco control. Despite this shared commitment, there is heterogeneity in the tobacco control environment of Cambodia, Laos and Vietnam. For example, mass-media advertisement of cigarettes is banned in Vietnam while a partial ban is in place in Laos. There are, however, no tobacco advertising restrictions in Cambodia.13
School-level and class-level influences
Existing studies have identified a number of school-level characteristics associated with increased susceptibility to smoking. In particular, one factor that is positively associated with susceptibility to smoking is the exposure to tobacco advertising.14–20 Evidence of the effectiveness of school-based smoking prevention programmes is mixed. It suggests that exposure to and perceived usefulness of school prevention programmes and perceived information helpfulness is associated with decreased susceptibility.21 However, it does not appear that school-based smoking prevention programmes have had much long-term success at keeping youths from starting to smoke.22 Two studies examined the association between exposure to anti-tobacco messages and smoking susceptibility and find a surprising positive association.17 20 Attendance at a school with a relatively high smoking rate among older students has been associated with increased susceptibility to smoking.23
Individual-level influences
At the student level, there is considerable evidence that youth smoking susceptibility is influenced by whether friends,15 17–20 23 24 family members,15 19 20 23 co-workers24 and/or teachers20 smoke. Additionally, the extent to which restrictions are placed on smoking at home,25 supports for smoke-free public policy26 and attitudes about the social consequences of smoking have been linked to susceptibility to smoking.27 28 The effect of sex on susceptibility to smoking is unclear. Some studies find that boys are at increased risk,16 17 20 some that it is girls who are at increased risk,18 24 28 29 while others find no significant differences.9 14 15 19
None of the above studies have been conducted in low-income countries and only two have been conducted outside of Canada or the United States. Ertas20 uses GYTS data to explore the factors associated with stages of cigarette smoking, including susceptibility, among Turkish youths. Arillo-Santillan et al30 use cross-sectional data from a school survey to examine the factors associated with smoking susceptibility in 10 Mexican cities.
Objectives
The objectives of this study are twofold. The first objective is to examine the extent to which variability in smoking susceptibility is associated with between-school versus within-school/class differences—that is, to what extent do school and classroom contexts influence susceptibility to smoking. Disaggregating these effects provides a basis for estimating the potential of contexts to influence behaviour. Given that contexts have the potential to influence susceptibility, the next objective is to identify and quantify specific influences, particularly those that might be susceptible to modification through national tobacco control policies such as strict advertising bans or school-based prevention programmes. Our second objective is thus to determine the factors that influence susceptibility to smoking among young never-smokers in South East Asia with special attention given to school-level (current tobacco use prevalence in school, exposure to anti-smoking media messages, and exposure to tobacco billboard advertising) and class-level determinants (prevention). In doing so, we allow the association between smoking susceptibility and sex to vary across schools and explore cross-level interactions to assess whether the magnitude of school and class influences on susceptibility may be modified by sex.
METHOD
Data
The World Health Organization and Centers for Disease Control and Prevention developed the GYTS to track tobacco use among young people across countries using a common methodology and core questionnaire. The GYTS is school-based and employs a two-stage sample design to collect representative data on smoking among students aged 11–17. The first stage is a probabilistic selection of schools. The second stage is a random selection of classes from participating schools.31
In this study, data are taken from four city-based surveys (Ha Noi, Hai Phong, Da Nang and Ho Chi Minh) and one province-based survey (Tuyen Quang province) conducted in Vietnam, three province-based surveys (Luang Prabang, Vientiane, and Savannakhet) and one city-based survey (Vientiane Municipality) conducted in Laos and a national survey conducted in Cambodia. The surveys were conducted in 2003 in Vietnam and Laos, and in 2002 in Cambodia. Among participating students, the number, grade levels and average age were: Cambodia: 2011 students, grades 8–10 (average age 15.2); Laos: 9720 students, grades 2–4 (average age 14.2); and Vietnam: 9507 students, grades 8–10 (average age 15.1). School response rates were 100 for all surveys with the exception of Cambodia (98) and Ho Chi Minh City (88). Student and overall response rates (that is, rates that take into account responses at both school and student levels) were high with average response rates across surveys of 91 and 90, respectively.
Concepts and measures
Dependent variable
Smoking susceptibility among South East Asian never-smokers (that is, never tried or experimented with cigarette smoking, even one or two puffs), was derived using the algorithm of Pierce et al.5 In particular, susceptibility is measured by asking students the following questions (using a four-point ordinal scale):
-
Do you think you will be smoking cigarettes five years from now?
-
If one of your best friends offered you a cigarette, would you smoke it?
-
At any time during the next 12 months do you think you will smoke a cigarette?
Students who answered “definitely not” to all three questions were considered non-susceptible; all other students were considered susceptible.
School and class-level variables
All school and class-level variables were constructed by aggregating individual responses up to the corresponding level, either classroom or school. The following school-level variables were constructed: prevalence of smoking, anti-smoking media exposure and exposure to billboard advertising. First, current tobacco use prevalence—school average—was the percentage of students in school using any form of tobacco product (for example, cigarettes, water pipe, etc) on one or more days in the 30 days preceding the survey. Second, anti-smoking media exposure was the percentage of students in school who reported having seen “a lot” of anti-smoking media messages during the past 30 days. Third, billboard tobacco advertising was the percentage of students who reported having seen “a lot” of billboard advertisements for cigarettes during the past 30 days.iii In Cambodia, Laos and Vietnam, students attending the same school, very likely reside in the same surrounding areas and therefore are exposed to similar levels of anti-smoking media and billboard advertising. Estimated at the class level, prevention was the percentage of students within each classroom who had been taught about the danger of smoking in any of their classes during the current school year.,iv The entire sample (n 21 238), including teenagers who had experimented with smoking and those who were currently smoking, was used to construct the school-level and class-level variables.
Individual-level variables
A number of individual-level variables were incorporated into the model. Parent and peer smoking behaviour were defined separately. A binary measure of mother and father smoking behaviour as reported by students is used while friends’ smoking behaviour is measured using a four-point ordinal scale (none; some; most; all). The knowledge of the harmful effects of secondhand smoke is measured using a four-point ordinal scale (from definitely not harmful to definitely harmful) while a five-point ordinal scale is used to measure exposure to secondhand smoke at home (from 0 days/week to 7 days/week). A continuous variable representing the following categories of age is included (⩽11 years; 12; 13; 14; 15; 16; ⩾17 years).v A variable for income—monthly pocket income—is also included. Because of differing response scales utilised across countries, we include a binary measure of pocket income (some versus none).
Analysis
A three-level, logistic multilevel model is employed to account for the nesting of individual students within classrooms, and classrooms within schools. Multilevel modelling explicitly accounts for between-context and between-individual heterogeneity. In addition to assessing the relative contribution of individual and contextual effects, multilevel techniques allow for an examination of cross-level effects.32
First, we specify a null or empty model to estimate the percentage of total variability in susceptibility that is the result of grouping (that is, differences between classes and differences between schools). Second, we include all explanatory variables as fixed effects. Third, we estimate a random slope model where the association between susceptibility and sex is allowed to vary across school (sex is specified as a random effect). Fourth, we determine if the strength of association between our school/class-level variables and susceptibility vary as a function of student sex (that is, test for cross-level interactions) and whether the slope of school/class-level variables differs from zero for teenage boys and girls (that is, test for simple slopes).
To facilitate the interpretation of interaction effects, the following variables are centred at their mean: income, friends’ smoking, school prevalence of any tobacco use, prevention, exposure to billboard tobacco advertising and exposure to anti-smoking media message; all school and class variables are expressed in increments of 10 (that is, the variables are coded so that a difference of one unit represents a 10 change in magnitude). All models are estimated using Stata/SE 10.0 for Macintosh and re-estimated using MlwiN 2.02 (results not reported) to prevent against programming coding errors.
RESULTS
Table 1 presents sample characteristics of young South East Asian never-smokers. About 13.2 of students are susceptible to smoking, from 14.1 in Cambodia to 12.4 in Laos. There are relatively few students less than 14 years old. About 50 of students have a father who smokes but less than 5 have a mother who smokes. Most students (67.6) have non-smoking friends. About half of the students (48.3) report being exposed to secondhand smoke at home, including 16.1 who report being exposed every day of the week. Most students (90.0) believe that secondhand smoke is harmful.
Table 2 presents class and school characteristics. The sample is taken from students nested in 516 classes nested in 260 schools (49 in Cambodia, 100 in Laos and 111 in Vietnam). On average, about 82 students are sampled from each school. School average prevalence of any tobacco use (for example, cigarette, and tobacco smoking such as water pipe) is 8.6: 25.4 of students report having seen “a lot” of billboard tobacco advertising and 62.7 report having seen “a lot” of anti-smoking media messages. Finally, 63.9 of students report having been taught in classes about the dangers of smoking.
Tables 3 and 4 present the results of three-level multilevel logistic regressions (table 3 presents the multilevel logistic regressions (coefficients, standard errors and random effects variances at levels 3 and 2) while table 4 converts the fixed effects parameters in table 3, models 2–4, to odds ratios and 95 confidence intervals). In model 1, the school and class intercept variances, at 0.161 and 0.152, are both statistically significant. Expressed as intra-class correlation coefficients, these variances indicate that 4.5 and 4.2 of the variability in susceptibility to smoking is associated with between-school (level 3) and between-class (level 2) differences.vi As is usually the case, between-individual differences are far more important.3334
Model 2 incorporates all explanatory variables into the null model. Smoking prevalence among students of the same school has a positive and significant impact on students’ susceptibility to smoking. A 10 increase in prevalence increases the odds of being susceptible by about 16 (OR 1.16, 95 CI 1.04 to 1.30). Exposure to anti-smoking messages has a protective effect, as it significantly reduces the odds of being susceptible (OR 0.93, 95 CI 0.87 to 0.99). The associations between smoking susceptibility, exposure to billboard tobacco advertising and classroom prevention (that is, having been taught in classes about the dangers of smoking) are not significant. Parental smoking increases the likelihood of smoking susceptibility by nearly 20 if the father smokes (OR 1.20, 95 CI 1.08 to 1.34) and by more than 100 if the mother smokes (OR 2.12, 95 CI 1.68 to 2.66). Boys are found to be almost 250 more likely to be susceptible to smoking than girls (OR 3.48, 95 CI 3.12 to 3.87) while age is not associated with smoking susceptibility (OR 1.02, 95 CI 0.98 to 1.06). Exposure to secondhand smoke at home increases the odds of being susceptible (OR 1.08, 95 CI 1.04 to 1.12), while knowledge about the harmful effects of secondhand smoke has a protective effect (OR 0.74, 95 CI 0.69 to 0.78). Students who have pocket income are more than 60 more likely to be susceptible than students who have none (OR 1.64, 95 CI 1.47 to 1.84). Lastly, levels of susceptibility are not associated with country.
In model 3, the regression of susceptibility on student sex is allowed to vary across schools. The slope variance at 0.395 is significant, indicating that the relation between sex and susceptibility varies from school to school. In about 68 of schools the regression of susceptibility on student sex will go from 0.64 to 1.90 ((SD 1.269 (√0.395)). While the effect of being a boy is positive in all schools, the magnitude of that risk is much stronger in some school than others. The high effects are more than three times larger than low effects.vii
Model 4 builds on model 3 and incorporates cross-level interactions. The asymptotic variance/covariance matrix of fixed effects necessary to probe the significance of interactions is available upon request.35 36 For girls only, billboard tobacco advertising increases the risk for susceptibility and classroom prevention decreases risk. For boys only, school prevalence increases the risk for susceptibility and anti-smoking media decreases risk. The influences of billboard tobacco advertising and classroom prevention are not significantly different from zero for boys while the influences of school prevalence and anti-smoking media are not significantly different from zero for girls.
DISCUSSION
Principal findings
We find the association between sex and susceptibility varies significantly across schools. For girls only, the association between advertising and susceptibility is positive and significant while the association between classroom prevention and susceptibility is negative and significant. For boys only, the association between school prevalence and susceptibility is positive and significant while the association between anti-smoking media messages and susceptibility is negative and significant.
We provide additional evidence that teenagers who attend schools with higher prevalence of tobacco use are more likely to be susceptible to smoking.23 Contrary to existing findings17 20 we find exposure to anti-smoking media messages is negatively associated with smoking susceptibility.viii Consistent with the findings of other studies, we find teenagers who have parents or friends who smoke are more susceptible to smoking.15–2023 24 Additionally we find teenagers who are exposed to secondhand smoke at home and those who have access to pocket income to be more susceptible, while better knowledge of the harmful effects of secondhand smoke appears to diminish susceptibility to smoking. We do not find age is significantly associated with the odds of being susceptible to smoking. This result may be explained by the fairly homogeneous sample of students used in this analysis (that is, most are between the age of 14 and 16). Individual-level influences are robust across specifications.
Strengths and weaknesses of the study
The study presented is sound and innovative in several respects. First, to our knowledge, it is the first that examines the factors associated with the susceptibility to smoking in low-income countries. In this context, it takes advantage of a relatively untapped and rich dataset of more than 20 000 youth in three South East Asian countries. Second, the methodological approach is sound in that its utilisation of multilevel modelling permits the examination of contextual and individual-level determinants and variation between classes and schools.
There are, however, some weaknesses that merit discussion. First, the measure of susceptibility developed by Pierce et al5 has not been validated in low-income countries. As discussed earlier, there is considerable evidence that Pierce et al’s measure of susceptibility is a good predictor of experimentation. Additionally, Pierce et al examined two-way interactions between all their predictor variables (age, sex, race, adult education, family income, perceived school performance, exposure to other smokers) and found that none of the two-way interactions between susceptibility to smoking and the other predictor variables were significant. These results provide support, albeit limited, that South East Asian youths may not differ markedly from North American youths in their progression through the smoking uptake continuum despite differing social and gender norms regarding smoking.
Second, anti-smoking media messages and to a lesser extent, exposure to billboard tobacco advertising may be endogenous (that is, not independent of teenagers’ susceptibility). For example, the differences in reported exposure to anti-smoking media messages by students may be explained by differences in susceptibility (less susceptible teenagers being more aware of anti-smoking media messages). That said, our measures of anti-smoking media messages and exposure to billboard tobacco advertising represent school averages and thus, may alleviate the problem of endogeneity because differences in exposure may reflect regional differences (especially across regions from different countries). An additional concern with respect to our measure of tobacco advertising is that nearly 20 of Vietnamese respondents report having seen “a lot” of billboard advertisements for cigarettes during the past 30 days when, in theory, such advertisements are banned in Vietnam. It may be the case that point-of-purchase promotion, which is often substantial in Vietnam, was perceived as being equivalent to billboard advertisements.
Third, GYTS is a school-based survey and cannot capture teenagers no longer in school. Secondary school enrolment is high in Vietnam (72 in girls and 75 in boys) but relatively low in Cambodia (24 in girls and 35 in boys) and Laos (39 in girls and 52 in boys).37 Additionally, the surveys conducted in Laos and Vietnam are not nationally representative. Four out of five surveys conducted in Vietnam took place mostly in urban areas, whereas the urban population as a percentage of total population is only about 25 in Vietnam.37 This characteristic of the GYTS limits the generalisability of our findings. Fourth, the data utilised are not longitudinal and, as such, causation cannot be inferred. Additionally, cross-sectional data do not allow for control of residual heterogeneity (that is, the possibility of substantial variation between similar individuals owing to unmeasured variables).38
Meaning of the study: implications for policymakers
Cambodia, Laos and Vietnam have all recently ratified the Framework Convention on Tobacco Control which demonstrates a commitment from policymakers to curb the tobacco epidemic. Findings from this study provide insight to policymakers in addressing tobacco use initiation among young boys and girls.
First, this study’s most important contribution is the finding that sex can modify factors associated with the susceptibility to smoking. This finding provides support to the call to move beyond gender-blind tobacco control policies.39–41 Second, our results indicate the potential benefits of banning billboard tobacco advertising in regions where it is not already banned and further existing bans to include point-of-purchase promotion. Tobacco industry documents have revealed tobacco companies have used such promotion in response to advertising restrictions.42 Third, the positive association between exposure to secondhand smoke at home and the negative association between knowledge of the harmful effects of secondhand smoke with susceptibility to smoking suggest potential benefits to be gained from the provision of information about the harmful effects of secondhand smoke. In particular, it may improve teenager knowledge and reduce their exposure to secondhand smoke at home. Fourth, our results underscore the important influence that parents and friends have on the susceptibility to smoking among youths. As such, interventions proved to be effective at reducing tobacco use overall, such as price increases, can be expected to have an additional benefit in reducing teenage susceptibility to smoking.
Unanswered questions and future research
The results of this study illustrate the advantages of using multilevel modelling with GYTS data from multiple countries. In addition to studying smoking susceptibility, GYTS data can be used to examine factors associated with smoking initiation—a particular void in the current literature—and to explore exposure to tobacco advertising among South East Asian teenagers, especially in the presence of comprehensive advertising bans. Using multilevel approaches will shed additional light on the benefits of the GYTS data for understanding the determinants of tobacco initiation and use among adolescents in surveyed countries. Given the lack of research on the predictive validity of Pierce et al’s5 measure of susceptibility outside of the United States, research exploring smoking susceptibility’s predictive validity in low-income and middle-income countries is warranted.
What is already known
-
Susceptibility to smoking has been shown to be a good predictor of smoking onset.
-
School-level influences including exposure to tobacco advertising and attendance at a school with a relatively high smoking rate among older students have been shown to be associated with increased susceptibility to smoking while the evidence with respect to school prevention programmes is mixed.
What this study adds
-
This is the first study to examine factors associated with susceptibility to smoking in low-income countries.
-
It highlights the importance of contextual interactions as sex is found to modify the association between susceptibility to smoking and school-level and class-level influences.
Acknowledgments
We thank the CDC and GYTS country investigators for making these data available. In particular we thank Yel Daravuth, Phan Thi Hai, Nathan Jones, Lam Nguyen Tuan, Sin Sovann and Wick C Warren. We thank Paul Contoyannis and Lam Nguyen Tuan for their comments and we acknowledge support from colleagues in Advanced Analysis of Survey Data.
REFERENCES
Footnotes
-
Funding: GEG is supported by a Social Sciences and Humanities Research Council of Canada (SSHRC) Canada Graduate Scholarship (CGS). During this study, GEG was supported by an Ashley Studentships for Research in Tobacco Control, Ontario Tobacco Research Unit and KG was supported by a SSHRC post-doctoral fellowship award. MHB is supported by a Canada Research Chair in the Social Determinants of Child Health.
-
Competing interests: GEG is a former employee of the World Health Organization—Tobacco Free Initiative whose objective is to reduce the global burden of disease and death caused by tobacco.
-
↵i Note that the survey conducted in Cambodia uses a different methodology (for example, definition of smokers) from that of Laos and Vietnam. Such differences undermine the comparability of the studies.
-
↵ii The smoking continuum was defined as: never smoker; puffer; non-recent experimenter; recent experimenter; former established; current established.
-
↵iii The question reads: During the past 30 days (one month), how many advertisements for cigarettes have you seen on billboards? [A lot; A few; None.]
-
↵iv The question reads: During this school year, were you taught in any of your classes about the dangers of smoking? [Yes; No; Not sure.]
-
↵v Note that the category 11 years includes people under the age of 11 and the category 17 years includes people older than 17 years. Given that so few observations fall into these categories, we are confident the outcome will not be affected in a meaningful way.
-
↵vi and where ϕ02 school-level intercept variance and τ02 class-level intercept variance.
-
↵vii “High” and “low” refer to values occurring in schools with, respectively, the top 15, and the bottom 15, of the school-dependent effect.
-
↵viii This finding is only true when the relation between susceptibility and anti-smoking messages is not allowed to vary across sex (that is, no contextual interaction between sex and anti-smoking messages).